Merging r286998:

------------------------------------------------------------------------ r286998 | chfast | 2016-11-15 19:29:24 +0100 (wto, 15 lis 2016) | 12 lines Integer legalization: fix MUL expansion Summary: This fixes the runtime results produces by the fallback multiplication expansion introduced in r270720. For tests I created a fuzz tester that compares the results with Boost.Multiprecision. Reviewers: hfinkel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D26628 llvm-svn: 288086
author: Pawel Bylica <chfast@gmail.com> 2016-11-28 22:45:01 +0000
committer: Pawel Bylica <chfast@gmail.com> 2016-11-28 22:45:01 +0000
commit: cf874a215545dd6b8685b5a28ec0ff3e04d87b3b (patch)
tree: 57abba27e9710adf4328f4f2e8bf959819cce4e1
parent: 97214e4fb8f12738c3b1cc0d4af6d7f579986068 (diff)
download: llvm-cf874a215545dd6b8685b5a28ec0ff3e04d87b3b.tar.gz
4 files changed, 4363 insertions, 3841 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index e7dd913da2ea..9a18943291c8 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -2207,7 +2207,7 @@ void DAGTypeLegalizer::ExpandIntRes_MUL(SDNode *N,
     SDValue RLH = DAG.getNode(ISD::SRL, dl, NVT, RL, Shift);
 
     SDValue U = DAG.getNode(ISD::ADD, dl, NVT,
-                            DAG.getNode(ISD::MUL, dl, NVT, LLH, RLL), TL);
+                            DAG.getNode(ISD::MUL, dl, NVT, LLH, RLL), TH);
     SDValue UL = DAG.getNode(ISD::AND, dl, NVT, U, Mask);
     SDValue UH = DAG.getNode(ISD::SRL, dl, NVT, U, Shift);
 
@@ -2216,14 +2216,14 @@ void DAGTypeLegalizer::ExpandIntRes_MUL(SDNode *N,
     SDValue VH = DAG.getNode(ISD::SRL, dl, NVT, V, Shift);
 
     SDValue W = DAG.getNode(ISD::ADD, dl, NVT,
-                            DAG.getNode(ISD::MUL, dl, NVT, LL, RL),
+                            DAG.getNode(ISD::MUL, dl, NVT, LLH, RLH),
                             DAG.getNode(ISD::ADD, dl, NVT, UH, VH));
-    Lo = DAG.getNode(ISD::ADD, dl, NVT, TH,
+    Lo = DAG.getNode(ISD::ADD, dl, NVT, TL,
                      DAG.getNode(ISD::SHL, dl, NVT, V, Shift));
 
     Hi = DAG.getNode(ISD::ADD, dl, NVT, W,
                      DAG.getNode(ISD::ADD, dl, NVT,
-                                 DAG.getNode(ISD::MUL, dl, NVT, RH, LL), 
+                                 DAG.getNode(ISD::MUL, dl, NVT, RH, LL),
                                  DAG.getNode(ISD::MUL, dl, NVT, RL, LH)));
     return;
   }
diff --git a/llvm/test/CodeGen/X86/mul-i1024.ll b/llvm/test/CodeGen/X86/mul-i1024.ll
index 768afc98f4f1..60933b1e1fa1 100644
--- a/llvm/test/CodeGen/X86/mul-i1024.ll
+++ b/llvm/test/CodeGen/X86/mul-i1024.ll
@@ -11,49 +11,171 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    andl $-8, %esp
-; X32-NEXT:    subl $2496, %esp # imm = 0x9C0
-; X32-NEXT:    movl 12(%ebp), %edx
-; X32-NEXT:    movl 8(%ebp), %ecx
-; X32-NEXT:    movl 40(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 44(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl (%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 4(%ecx), %eax
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    movl 32(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 36(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 8(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 12(%ecx), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 8(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 12(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 64(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 68(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 72(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 76(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl (%edx), %ecx
-; X32-NEXT:    movl 4(%edx), %edx
+; X32-NEXT:    subl $2640, %esp # imm = 0xA50
+; X32-NEXT:    movl 8(%ebp), %eax
+; X32-NEXT:    movl 64(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 68(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 72(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 76(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 80(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 84(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 88(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 92(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 96(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 100(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 104(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 108(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 112(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 116(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 120(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 124(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl (%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 4(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 8(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 12(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 16(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 20(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 24(%eax), %ecx
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl 28(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 32(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 36(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 40(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 44(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 48(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 52(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 56(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 60(%eax), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 12(%ebp), %eax
+; X32-NEXT:    movl 48(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 52(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 56(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 60(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 32(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 36(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 40(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 44(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 16(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 20(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 24(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 28(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 8(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 12(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 112(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 116(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 120(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 124(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 96(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 100(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 104(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 108(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 80(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 84(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 88(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 92(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 64(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 68(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 72(%eax), %edi
+; X32-NEXT:    movl 76(%eax), %esi
+; X32-NEXT:    movl (%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 4(%eax), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    pushl %ecx
-; X32-NEXT:    movl %ecx, %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edx
+; X32-NEXT:    pushl %ecx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl %esi, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -61,27 +183,24 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -89,11 +208,13 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -101,9 +222,8 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -112,146 +232,76 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
-; X32-NEXT:    seto %al
-; X32-NEXT:    lahf
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    popl %eax
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    seto %al
-; X32-NEXT:    lahf
-; X32-NEXT:    movl %eax, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    seto %al
-; X32-NEXT:    lahf
-; X32-NEXT:    movl %eax, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addb $127, %al
-; X32-NEXT:    sahf
-; X32-NEXT:    popl %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 8(%ebp), %edx
-; X32-NEXT:    movl 68(%edx), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    addb $127, %al
-; X32-NEXT:    sahf
-; X32-NEXT:    popl %eax
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 64(%edx), %ecx
-; X32-NEXT:    movl 12(%ebp), %edx
-; X32-NEXT:    movl 44(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 40(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 36(%edx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 32(%edx), %ebx
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %ecx
-; X32-NEXT:    movl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -262,15 +312,13 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -278,74 +326,36 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    seto %al
-; X32-NEXT:    lahf
-; X32-NEXT:    movl %eax, %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    seto %al
-; X32-NEXT:    lahf
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    popl %eax
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %eax, %eax
-; X32-NEXT:    addb $127, %al
-; X32-NEXT:    sahf
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 8(%ebp), %eax
-; X32-NEXT:    movl 76(%eax), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 72(%eax), %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edx
-; X32-NEXT:    pushl %ecx
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -353,33 +363,13 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    addb $127, %al
-; X32-NEXT:    sahf
-; X32-NEXT:    popl %eax
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -387,148 +377,42 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %ecx
-; X32-NEXT:    movl 92(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 88(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 84(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 80(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 108(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 104(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 100(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 96(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 124(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 120(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 116(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 112(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 28(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 24(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 20(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 16(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 60(%ecx), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 56(%ecx), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 52(%ecx), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 48(%ecx), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 12(%ebp), %ecx
-; X32-NEXT:    movl 60(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 56(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 52(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 48(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 28(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 24(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 20(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 16(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 124(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 120(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 116(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 112(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 108(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 104(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 100(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 96(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 92(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 88(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 84(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 80(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addb $127, %al
-; X32-NEXT:    sahf
-; X32-NEXT:    popl %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %edx
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -536,104 +420,119 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl %ebx, %esi
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -641,14 +540,12 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -657,11 +554,11 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -680,29 +577,53 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -710,79 +631,91 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    pushl %ecx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl %ebx, %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -791,13 +724,12 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -806,11 +738,12 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -818,7 +751,8 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, %esi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -827,43 +761,42 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -871,154 +804,144 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl %edi, %ebx
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    calll __multi3
-; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -1026,26 +949,24 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
@@ -1056,7 +977,8 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
@@ -1072,16 +994,18 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -1092,11 +1016,10 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
@@ -1105,16 +1028,15 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1125,8 +1047,9 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1139,8 +1062,8 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1151,35 +1074,36 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl %ebx, %esi
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1188,12 +1112,12 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1201,146 +1125,132 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    calll __multi3
-; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1348,78 +1258,78 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %ebx
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -1427,20 +1337,37 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl %edi, %ebx
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -1453,10 +1380,11 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1477,11 +1405,10 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1492,62 +1419,76 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl %esi, %edi
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1558,7 +1499,7 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
@@ -1567,10 +1508,9 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -1581,11 +1521,26 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -1593,29 +1548,27 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1623,11 +1576,11 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1641,7 +1594,7 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1649,65 +1602,115 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %ebx
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
@@ -1716,13 +1719,25 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1735,7 +1750,7 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -1747,22 +1762,22 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -1774,22 +1789,22 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -1797,66 +1812,93 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    pushl %ecx
-; X32-NEXT:    movl %ecx, %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1867,8 +1909,10 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1877,13 +1921,10 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1895,7 +1936,7 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -1906,141 +1947,189 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, %edi
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movl %edx, %eax
+; X32-NEXT:    adcl %esi, %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl (%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    sbbl %eax, %eax
-; X32-NEXT:    andl $1, %eax
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    movl $0, %ebx
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %eax, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %ecx, %edx
+; X32-NEXT:    addl %eax, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %ecx, %esi
 ; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %eax, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %ecx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl %ecx, %edi
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    adcl %ebx, %ecx
 ; X32-NEXT:    movl $0, %edx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    sbbl %edx, %edx
-; X32-NEXT:    andl $1, %edx
+; X32-NEXT:    sbbl %eax, %eax
+; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edi, %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %edx, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl %edi, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %ecx, %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl %ecx, %ebx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -2053,23 +2142,22 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
@@ -2078,135 +2166,150 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %edx, %edx
-; X32-NEXT:    andl $1, %edx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl $0, %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    sbbl %esi, %esi
+; X32-NEXT:    andl $1, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %ebx
-; X32-NEXT:    movl (%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %eax, %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %ebx
+; X32-NEXT:    movl %ebx, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %edx, %esi
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    sbbl %edx, %edx
 ; X32-NEXT:    andl $1, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl $0, %eax
@@ -2221,165 +2324,232 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    sbbl %eax, %eax
 ; X32-NEXT:    andl $1, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    sbbl %ecx, %ecx
 ; X32-NEXT:    andl $1, %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, %edx
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    movl %edi, %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edi, %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl $0, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl %edx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl %ebx, %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl $0, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    sbbl %eax, %eax
+; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    movl %edi, %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edx, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edi, %edx
+; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ebx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl $0, %edi
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    sbbl %eax, %eax
+; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    sbbl %ecx, %ecx
@@ -2390,109 +2560,204 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    addl %edx, %eax
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl $0, %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    sbbl %esi, %esi
-; X32-NEXT:    andl $1, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edx, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    sbbl %eax, %eax
+; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %ebx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
@@ -2502,138 +2767,89 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
 ; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl %eax, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    movl $0, %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    sbbl %eax, %eax
-; X32-NEXT:    andl $1, %eax
-; X32-NEXT:    movl %eax, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %eax, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    addl %eax, %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl (%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %ecx, %edi
+; X32-NEXT:    adcl %eax, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %esi, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %edx, %ecx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl %esi, %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl %ebx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
@@ -2654,112 +2870,127 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    movl $0, %edx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    sbbl %edx, %edx
-; X32-NEXT:    andl $1, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    sbbl %esi, %esi
+; X32-NEXT:    andl $1, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %ebx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    sbbl %esi, %esi
-; X32-NEXT:    andl $1, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    movl %edi, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -2770,20 +3001,19 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edx, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -2796,91 +3026,93 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    sbbl %eax, %eax
 ; X32-NEXT:    andl $1, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl $0, %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    sbbl %edi, %edi
-; X32-NEXT:    andl $1, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edi, %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %ebx, %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
@@ -2889,36 +3121,38 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %esi, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl %ebx, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -2946,52 +3180,69 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl $0, %eax
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl $0, %eax
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl $0, %eax
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    sbbl %eax, %eax
-; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    movl $0, %ebx
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    sbbl %eax, %eax
+; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl (%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ebx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
@@ -3013,30 +3264,42 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl %eax, %edx
 ; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movl $0, %eax
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl $0, %edi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    sbbl %ebx, %ebx
 ; X32-NEXT:    andl $1, %ebx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl %ebx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %ebx, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -3056,228 +3319,355 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl %eax, %edx
 ; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movl $0, %ebx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    sbbl %eax, %eax
-; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    movl %edi, %ecx
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %edx, %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %edx
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edx, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %esi, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    sbbl %eax, %eax
+; X32-NEXT:    andl $1, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ebx, %ebx
+; X32-NEXT:    andl $1, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl %ebx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %eax, %edx
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edi
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl %eax, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %edi, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edx, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    sbbl %esi, %esi
-; X32-NEXT:    andl $1, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl %eax, %edx
 ; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ebx, %ebx
+; X32-NEXT:    andl $1, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    movl %esi, %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edi, %esi
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -3290,53 +3680,68 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    sbbl %eax, %eax
 ; X32-NEXT:    andl $1, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl $0, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ebx, %ebx
+; X32-NEXT:    andl $1, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %edx, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %ebx, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %esi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
@@ -3472,51 +3877,35 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    addl %edx, %eax
 ; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %ebx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    sbbl %edx, %edx
-; X32-NEXT:    andl $1, %edx
+; X32-NEXT:    movl $0, %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    sbbl %esi, %esi
+; X32-NEXT:    andl $1, %esi
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl %edi, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    adcl %ebx, %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %eax, %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
@@ -3564,25 +3953,79 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %esi, %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ebx, %ebx
+; X32-NEXT:    andl $1, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -3604,53 +4047,66 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    sbbl %eax, %eax
 ; X32-NEXT:    andl $1, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl $0, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl $0, %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    sbbl %esi, %esi
+; X32-NEXT:    andl $1, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %eax, %ebx
+; X32-NEXT:    adcl %ecx, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
@@ -3699,199 +4155,153 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl %edx, %eax
+; X32-NEXT:    movl $0, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    sbbl %edi, %edi
+; X32-NEXT:    andl $1, %edi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl %eax, %ebx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %edi, %esi
-; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
@@ -3899,82 +4309,70 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl 16(%ebp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 32(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 36(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 40(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, (%eax)
-; X32-NEXT:    movl %esi, 4(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 44(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 8(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 48(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 52(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 12(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 56(%eax)
-; X32-NEXT:    movl %ebx, 16(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 60(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, 20(%eax)
-; X32-NEXT:    movl %edi, 24(%eax)
-; X32-NEXT:    movl %ecx, 28(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 64(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 68(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 72(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 76(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 80(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 84(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 88(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 92(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 96(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 100(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 104(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 108(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 112(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 116(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 120(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 124(%eax)
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 16(%ebp), %ebx
+; X32-NEXT:    movl %ecx, 4(%ebx)
+; X32-NEXT:    movl 16(%ebp), %ecx
+; X32-NEXT:    movl %eax, (%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 8(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 12(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 16(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 20(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 24(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 28(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 32(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 36(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 40(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 44(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 48(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 52(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 56(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 60(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 64(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 68(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 72(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 76(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 80(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 84(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 88(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 92(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 96(%ecx)
+; X32-NEXT:    movl %edx, 100(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 104(%ecx)
+; X32-NEXT:    movl %esi, 108(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 112(%ecx)
+; X32-NEXT:    movl %edi, 116(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 120(%ecx)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 124(%ecx)
 ; X32-NEXT:    leal -12(%ebp), %esp
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    popl %edi
@@ -3990,691 +4388,733 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X64-NEXT:    pushq %r13
 ; X64-NEXT:    pushq %r12
 ; X64-NEXT:    pushq %rbx
-; X64-NEXT:    subq $496, %rsp # imm = 0x1F0
+; X64-NEXT:    subq $360, %rsp # imm = 0x168
 ; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 56(%rdi), %rbp
-; X64-NEXT:    movq 48(%rdi), %r11
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq (%rsi), %r14
-; X64-NEXT:    movq 8(%rsi), %rcx
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %r8
-; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rbx, %rbp
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %r10
-; X64-NEXT:    movq %r10, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    addq %rbp, %r12
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    addq %rsi, %rcx
-; X64-NEXT:    sbbq %rsi, %rsi
-; X64-NEXT:    andl $1, %esi
-; X64-NEXT:    movq %r11, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    imulq %r10, %rax
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    leaq (%r9,%rax), %rax
-; X64-NEXT:    movq %r9, %r10
-; X64-NEXT:    imulq %r14, %r8
-; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r8, %rax
+; X64-NEXT:    movq 48(%rdi), %r9
+; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq 40(%rdi), %rcx
+; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq 32(%rdi), %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rbx, %rcx
-; X64-NEXT:    adcq %rax, %rsi
+; X64-NEXT:    movq %rdi, %r10
 ; X64-NEXT:    xorl %r8d, %r8d
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %rax
 ; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    movq %r15, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rbx, %r15
-; X64-NEXT:    adcq %r11, %r9
-; X64-NEXT:    addq %rcx, %r15
-; X64-NEXT:    adcq %rsi, %r9
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 32(%rdi), %r13
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    movq %r14, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 40(%rdi), %r11
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rbp, %rsi
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq -{{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rbx, %r10
-; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rax, %r12
-; X64-NEXT:    movq %r12, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    adcq $0, %r9
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    movq 16(%r10), %rdi
-; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %r12, %rcx
-; X64-NEXT:    movq %r12, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq 24(%r10), %rsi
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rcx, %r14
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    addq %rdi, %rsi
-; X64-NEXT:    sbbq %rbp, %rbp
-; X64-NEXT:    andl $1, %ebp
-; X64-NEXT:    movq %r13, %rcx
-; X64-NEXT:    imulq %rbx, %rcx
-; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %r11
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    leaq (%r10,%rcx), %rcx
-; X64-NEXT:    addq %r11, %rcx
-; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r12, %rsi
-; X64-NEXT:    adcq %rcx, %rbp
-; X64-NEXT:    xorl %edi, %edi
-; X64-NEXT:    mulq %rdi
+; X64-NEXT:    addq %rdi, %rcx
 ; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    addq %r11, %rcx
 ; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdi, %rbp
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %rbx, %rbp
+; X64-NEXT:    sbbq %rbx, %rbx
+; X64-NEXT:    andl $1, %ebx
+; X64-NEXT:    addq %rax, %rbp
+; X64-NEXT:    adcq %rdx, %rbx
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %rcx, %rdi
-; X64-NEXT:    adcq %rbx, %r8
-; X64-NEXT:    addq %rsi, %rdi
-; X64-NEXT:    adcq %rbp, %r8
-; X64-NEXT:    addq %r10, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    movq %r14, (%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    adcq $0, %r8
-; X64-NEXT:    addq %r15, %rdi
-; X64-NEXT:    adcq %r9, %r8
-; X64-NEXT:    movl $0, %eax
-; X64-NEXT:    adcq $0, %rax
+; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r11, %r13
+; X64-NEXT:    addq %rax, %r13
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rdi, %r15
+; X64-NEXT:    movq %r15, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %rdx, %rax
+; X64-NEXT:    addq %rbp, %r13
+; X64-NEXT:    movq %r13, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %rbx, %rax
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    sbbq %r15, %r15
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq {{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    movq %r13, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rsi, %r8
+; X64-NEXT:    movq (%r8), %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    xorl %ebp, %ebp
+; X64-NEXT:    mulq %rbp
 ; X64-NEXT:    movq %rax, %r14
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %r11, %rcx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    andl $1, %r15d
-; X64-NEXT:    addq %rbp, %r14
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    movq 8(%r8), %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    xorl %r9d, %r9d
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rcx, %r12
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %r14, %r12
+; X64-NEXT:    movq %rcx, %rbx
+; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    addq %rbp, %rbx
 ; X64-NEXT:    sbbq %rbp, %rbp
 ; X64-NEXT:    andl $1, %ebp
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %r13
-; X64-NEXT:    movq %r13, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
-; X64-NEXT:    movq %r9, %rsi
-; X64-NEXT:    imulq %r10, %rsi
-; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    leaq (%r10,%r13), %rdx
-; X64-NEXT:    addq %rsi, %rdx
-; X64-NEXT:    movq %rbx, %r12
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
-; X64-NEXT:    movq %r12, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r12, %r14
-; X64-NEXT:    adcq %rdx, %rcx
-; X64-NEXT:    addq %r10, %rdi
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r8 # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    adcq %r15, %rcx
-; X64-NEXT:    xorl %r10d, %r10d
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    movq %r13, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    adcq %r15, %rdx
-; X64-NEXT:    movq %rsi, %rbx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    movq %r15, %rsi
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    adcq %rbp, %rsi
 ; X64-NEXT:    addq %rax, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rdx, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    sbbq %r9, %r9
-; X64-NEXT:    andl $1, %r9d
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    adcq %r11, %r9
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdx # 8-byte Reload
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    adcq %r11, %rbp
-; X64-NEXT:    addq %rax, %rdx
-; X64-NEXT:    adcq %r9, %rbp
+; X64-NEXT:    adcq %rdx, %rbp
+; X64-NEXT:    movq 16(%r8), %rax
+; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r14, %rdi
+; X64-NEXT:    movq %rdi, %r9
+; X64-NEXT:    addq %rax, %r9
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    adcq %rdx, %rax
+; X64-NEXT:    addq %rbx, %r9
+; X64-NEXT:    adcq %rbp, %rax
+; X64-NEXT:    movq %rax, %rbp
 ; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    addq %r15, %rax
-; X64-NEXT:    adcq %r13, %r10
-; X64-NEXT:    adcq %rbx, %rdx
-; X64-NEXT:    adcq %rsi, %rbp
+; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    adcq %rcx, %r15
+; X64-NEXT:    movq %r15, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq (%r10), %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r8, %r10
-; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r14, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rcx, %rbp
-; X64-NEXT:    movq %rbp, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq 16(%rcx), %rax
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq 24(%rcx), %rax
-; X64-NEXT:    movq %rcx, %r11
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    movq %r13, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %r12
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r8, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    mulq %rcx
+; X64-NEXT:    xorl %r15d, %r15d
+; X64-NEXT:    mulq %r15
 ; X64-NEXT:    movq %rdx, %rsi
 ; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rbx, %r14
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    addq %rbp, %rsi
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    adcq %rcx, %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    imulq %rcx, %rdi
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rcx, %r9
-; X64-NEXT:    movq %r13, %rcx
-; X64-NEXT:    imulq %r12, %rcx
-; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r10, %rbx
-; X64-NEXT:    leaq (%rbx,%rdi), %rbp
-; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    addq %r8, %rsi
-; X64-NEXT:    adcq %rdx, %rbp
+; X64-NEXT:    movq 32(%r8), %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    xorl %r8d, %r8d
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r14, %rcx
+; X64-NEXT:    addq %rax, %rcx
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    adcq %rdx, %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    addq %rdi, %r11
+; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    adcq %r12, %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r9, %r13
+; X64-NEXT:    movq %r13, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r9, %rdi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    adcq %rbp, %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rbp, %r9
+; X64-NEXT:    movq 8(%r10), %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    xorl %ecx, %ecx
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    addq %rsi, %r15
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    addq %r14, %r15
+; X64-NEXT:    movq %rsi, %rbp
+; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %rbx, %rbp
+; X64-NEXT:    sbbq %r8, %r8
+; X64-NEXT:    andl $1, %r8d
+; X64-NEXT:    addq %rax, %rbp
+; X64-NEXT:    adcq %rdx, %r8
+; X64-NEXT:    movq 16(%r10), %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r14, %rbx
+; X64-NEXT:    addq %rax, %rbx
+; X64-NEXT:    movq %rsi, %r10
+; X64-NEXT:    adcq %rdx, %r10
+; X64-NEXT:    addq %rbp, %rbx
+; X64-NEXT:    adcq %r8, %r10
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    movq %r14, (%rsp) # 8-byte Spill
+; X64-NEXT:    addq %r11, %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r15, %rcx
+; X64-NEXT:    adcq %rcx, %r12
+; X64-NEXT:    movq %r12, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %rbx, %rdi
+; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rbx, %r8
+; X64-NEXT:    adcq %r10, %r9
+; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r13 # 8-byte Reload
+; X64-NEXT:    movq 40(%r13), %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    adcq %r15, %rcx
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    adcq %rbp, %rcx
-; X64-NEXT:    movq %rcx, %rbp
-; X64-NEXT:    movq %r11, %rcx
-; X64-NEXT:    movq (%rcx), %r8
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    movq %r12, %rsi
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 8(%rcx), %r11
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %rsi
+; X64-NEXT:    xorl %edx, %edx
+; X64-NEXT:    mulq %rdx
+; X64-NEXT:    xorl %r11d, %r11d
 ; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r9 # 8-byte Reload
+; X64-NEXT:    addq %r9, %rsi
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
 ; X64-NEXT:    addq %rdi, %rsi
-; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    movq %r9, %rbp
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %rbx, %rbp
+; X64-NEXT:    sbbq %rbx, %rbx
+; X64-NEXT:    andl $1, %ebx
+; X64-NEXT:    addq %rax, %rbp
+; X64-NEXT:    adcq %rdx, %rbx
+; X64-NEXT:    movq 48(%r13), %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %r12
-; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    movq %rdi, %r15
+; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    addq %rax, %r15
+; X64-NEXT:    movq %r9, %rdi
+; X64-NEXT:    adcq %rdx, %rdi
+; X64-NEXT:    addq %rbp, %r15
+; X64-NEXT:    adcq %rbx, %rdi
+; X64-NEXT:    addq %r11, %r14
+; X64-NEXT:    movq %r14, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %rsi, %rcx
+; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r15, %r8
+; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %rdi, %r10
+; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    addq %r11, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    adcq %r9, %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r12
-; X64-NEXT:    addq %r10, %rbx
-; X64-NEXT:    movq %rbx, %r10
-; X64-NEXT:    adcq %rax, %r14
-; X64-NEXT:    adcq $0, %r13
-; X64-NEXT:    movq %r13, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r13
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    addq %r11, %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
+; X64-NEXT:    movq %r15, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
+; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %r15
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    movq 56(%rax), %rax
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r10
 ; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %r9, %rbp
+; X64-NEXT:    addq %rbx, %rbp
 ; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    mulq %rcx
 ; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    addq %rbp, %r8
 ; X64-NEXT:    adcq $0, %rbx
 ; X64-NEXT:    addq %rdi, %rbx
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    movq %r8, %rdi
-; X64-NEXT:    imulq %rcx, %rdi
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r11, %rcx
-; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    imulq %r15, %rcx
-; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    leaq (%r13,%rdi), %rsi
-; X64-NEXT:    addq %rcx, %rsi
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rcx, %r11
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    adcq %rdi, %rdx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r12 # 8-byte Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %r15
+; X64-NEXT:    adcq %rdx, %r12
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %r10, %rsi
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %r10, %rbx
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rdi, %r10
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    addq %rbp, %rdi
+; X64-NEXT:    sbbq %rbp, %rbp
+; X64-NEXT:    andl $1, %ebp
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rcx, %rsi
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %r13
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rdi, %rbx
+; X64-NEXT:    adcq %rbp, %r13
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
 ; X64-NEXT:    addq %r9, %rbx
+; X64-NEXT:    adcq %r8, %r13
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    movq %r10, %rbp
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rsi, %r8
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rcx, %rdi
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    movq 24(%rax), %r14
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r11 # 8-byte Reload
+; X64-NEXT:    addq %r11, %rbp
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rbp
 ; X64-NEXT:    adcq %rdx, %rsi
+; X64-NEXT:    addq %rbx, %r9
+; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r13, %rdi
+; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    addq %r15, %rbp
+; X64-NEXT:    adcq %r12, %rsi
+; X64-NEXT:    movl $0, %r10d
+; X64-NEXT:    adcq $0, %r10
+; X64-NEXT:    sbbq %r15, %r15
+; X64-NEXT:    andl $1, %r15d
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
 ; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    movq %r8, %r15
-; X64-NEXT:    movq %r15, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    mulq %rcx
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rbx
 ; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
+; X64-NEXT:    addq %r12, %rdi
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r14
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq %rbx, %rdi
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    addq %r13, %r10
-; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r14, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %rdi, %r9
 ; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    movq %r15, %rcx
-; X64-NEXT:    imulq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    imulq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    addq %rcx, %r11
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
-; X64-NEXT:    movq %r12, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rax, %r11
-; X64-NEXT:    movq %r11, %rdi
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r9, %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    adcq %r11, %rdx
-; X64-NEXT:    addq %r12, %rcx
-; X64-NEXT:    movq %rcx, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    addq %rbx, %rcx
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %r14, %r12
+; X64-NEXT:    addq %rcx, %rax
 ; X64-NEXT:    adcq %rdi, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    addq %r11, %rcx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
 ; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %r11, %rax
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    adcq %rdx, %rdi
+; X64-NEXT:    addq %rbp, %r13
+; X64-NEXT:    adcq %rsi, %r9
+; X64-NEXT:    adcq %r10, %rcx
+; X64-NEXT:    adcq %r15, %rdi
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
+; X64-NEXT:    movq %r13, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
+; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
 ; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rdx, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r10, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq %rbp, (%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq $0, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq $0, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq $0, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq $0, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r13 # 8-byte Reload
-; X64-NEXT:    movq 32(%r13), %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
+; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    movq 24(%rax), %rax
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r13
+; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %r12
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    mulq %rbp
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rbx, %rsi
-; X64-NEXT:    movq %rbx, %r14
-; X64-NEXT:    movq %r14, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    addq %rbx, %r15
 ; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    movq 40(%r13), %rbx
+; X64-NEXT:    addq %rdi, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rbp, %r14
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r8 # 8-byte Folded Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # 8-byte Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %r8
+; X64-NEXT:    adcq %rdx, %r10
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
 ; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %rbx
+; X64-NEXT:    mulq %r13
 ; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    addq %rsi, %r10
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
 ; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    addq %rcx, %rdi
-; X64-NEXT:    sbbq %rcx, %rcx
-; X64-NEXT:    andl $1, %ecx
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    imulq %rbx, %rax
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rsi, %r9
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    addq %rbx, %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    imulq %r12, %r11
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rsi, %rbx
+; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    addq %r11, %rbx
+; X64-NEXT:    adcq %r15, %rcx
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    adcq $0, %r10
+; X64-NEXT:    movq %r9, %rsi
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rbp, %r14
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r13
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %r9, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %r12
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %r12
+; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    adcq %rdi, %rdx
+; X64-NEXT:    movq (%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    addq %r14, %rsi
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rsi
+; X64-NEXT:    adcq %rdx, %rbp
+; X64-NEXT:    addq %rbx, %r11
 ; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rbp
-; X64-NEXT:    leaq (%rbp,%rax), %rax
-; X64-NEXT:    addq %r11, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r14, %rdi
-; X64-NEXT:    adcq %rax, %rcx
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    mulq %rdx
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
-; X64-NEXT:    addq %rax, %r9
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
-; X64-NEXT:    adcq %rdx, %r15
-; X64-NEXT:    addq %rdi, %r9
 ; X64-NEXT:    adcq %rcx, %r15
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
+; X64-NEXT:    movq %r15, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %r8, %rsi
+; X64-NEXT:    adcq %r10, %rbp
+; X64-NEXT:    movl $0, %r10d
+; X64-NEXT:    adcq $0, %r10
+; X64-NEXT:    sbbq %r15, %r15
+; X64-NEXT:    andl $1, %r15d
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
 ; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %r9, %rbx
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    movq %r12, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq %r14, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %rsi, %rcx
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %rbx, %r9
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rdi, %rcx
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
 ; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %rbx
+; X64-NEXT:    mulq %r12
 ; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r14, %rbp
-; X64-NEXT:    adcq %rax, %r10
-; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r9
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    movq 48(%r13), %rsi
-; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    movq %r11, %r14
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %r12, %rdi
-; X64-NEXT:    movq %r12, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq 56(%r13), %rcx
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %r13
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    adcq $0, %r13
-; X64-NEXT:    addq %rsi, %r13
-; X64-NEXT:    sbbq %rsi, %rsi
-; X64-NEXT:    andl $1, %esi
-; X64-NEXT:    imulq %rcx, %r8
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %r14
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    leaq (%rbx,%r8), %rcx
+; X64-NEXT:    adcq %rdi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    addq %r14, %rcx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r12 # 8-byte Reload
+; X64-NEXT:    movq %r12, %r13
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rcx
+; X64-NEXT:    adcq %rdx, %r13
+; X64-NEXT:    addq %rsi, %r11
+; X64-NEXT:    adcq %rbp, %r9
+; X64-NEXT:    adcq %r10, %rcx
+; X64-NEXT:    adcq %r15, %r13
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
+; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
 ; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r12, %r13
-; X64-NEXT:    adcq %rcx, %rsi
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    mulq %rdx
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
+; X64-NEXT:    adcq $0, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq $0, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq $0, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq $0, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    addq %rax, %r10
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    adcq %rdx, %r11
-; X64-NEXT:    addq %r13, %r10
-; X64-NEXT:    adcq %rsi, %r11
-; X64-NEXT:    addq %rbx, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    adcq $0, %r11
-; X64-NEXT:    addq %r9, %r10
-; X64-NEXT:    adcq %r15, %r11
-; X64-NEXT:    movl $0, %ebp
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    sbbq %r15, %r15
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rcx, %r11
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rdi, %rbp
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rsi, %rax
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq {{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rbp, %r10
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rbx, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r14
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r9 # 8-byte Reload
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
+; X64-NEXT:    movq %r12, %r15
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %r9
+; X64-NEXT:    adcq %rdx, %r15
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %r11, %rdi
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %r8, %rbp
+; X64-NEXT:    adcq $0, %rbx
 ; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    movq %rcx, %r8
+; X64-NEXT:    mulq %r14
 ; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rbx, %rcx
+; X64-NEXT:    sbbq %rbp, %rbp
+; X64-NEXT:    andl $1, %ebp
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rsi, %rdi
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq %rbp, %rsi
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq %r10, %rsi
+; X64-NEXT:    adcq $0, %r9
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    movq %r8, %rbp
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %rcx
 ; X64-NEXT:    addq %r14, %rcx
-; X64-NEXT:    addq %r8, %rcx
-; X64-NEXT:    andl $1, %r15d
-; X64-NEXT:    addq %r9, %r13
-; X64-NEXT:    adcq %rbx, %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdx # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    sbbq %r8, %r8
-; X64-NEXT:    andl $1, %r8d
+; X64-NEXT:    adcq $0, %r12
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r12 # 8-byte Reload
-; X64-NEXT:    movq %r12, %rbx
-; X64-NEXT:    imulq %rdi, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
-; X64-NEXT:    leaq (%r9,%rsi), %rsi
-; X64-NEXT:    addq %rbx, %rsi
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r8, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rdx, %r13
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    addq %r9, %r10
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq %rbp, %r13
-; X64-NEXT:    adcq %r15, %rcx
-; X64-NEXT:    xorl %r8d, %r8d
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    addq %rdi, %r9
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rdi, %r9
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    adcq %rbx, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r14 # 8-byte Reload
-; X64-NEXT:    addq %rdi, %r14
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    adcq %rbx, %rdi
-; X64-NEXT:    movq %rbx, %r15
-; X64-NEXT:    addq %rax, %r14
+; X64-NEXT:    movq 56(%rax), %rdi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %r12, %rbp
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    movq (%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
+; X64-NEXT:    addq %r11, %rcx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r12 # 8-byte Reload
+; X64-NEXT:    adcq %r12, %rdi
+; X64-NEXT:    addq %rax, %rcx
 ; X64-NEXT:    adcq %rdx, %rdi
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    movq %r12, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    adcq %r8, %rdx
-; X64-NEXT:    movq %rbx, %rsi
-; X64-NEXT:    movq %rbx, %rbp
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq %r8, %rbx
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rdx, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    addq %r15, %rax
-; X64-NEXT:    adcq %r12, %r9
-; X64-NEXT:    adcq %rsi, %r14
-; X64-NEXT:    adcq %rbx, %rdi
-; X64-NEXT:    addq %r10, %rax
-; X64-NEXT:    movq %rax, %rdx
-; X64-NEXT:    adcq %r11, %r9
-; X64-NEXT:    adcq %r13, %r14
-; X64-NEXT:    adcq %rcx, %rdi
+; X64-NEXT:    addq %rbx, %r8
+; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %rsi, %r10
+; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    addq %r9, %rcx
+; X64-NEXT:    adcq %r15, %rdi
+; X64-NEXT:    movl $0, %r8d
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    sbbq %r9, %r9
+; X64-NEXT:    andl $1, %r9d
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %r10, %rbp
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    addq %rbp, %r15
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    addq %rbx, %rsi
+; X64-NEXT:    sbbq %rbp, %rbp
+; X64-NEXT:    andl $1, %ebp
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    adcq %rbp, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
+; X64-NEXT:    addq %r11, %r10
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    adcq %r12, %r14
+; X64-NEXT:    addq %rax, %r10
+; X64-NEXT:    adcq %rdx, %r14
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    adcq %r8, %r15
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rdi, %r15
+; X64-NEXT:    adcq %r8, %r10
+; X64-NEXT:    adcq %r9, %r14
 ; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
-; X64-NEXT:    movq %r15, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rax, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    movq (%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rax, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    adcq $0, %r9
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    addq %rcx, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq %rcx, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq %rcx, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq %r13, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq $0, %rax
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    adcq $0, %r10
 ; X64-NEXT:    adcq $0, %r14
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    movq %r14, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movl $0, %eax
 ; X64-NEXT:    adcq $0, %rax
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
@@ -4683,764 +5123,806 @@ define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movl $0, %eax
 ; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    movq %rax, (%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    sbbq %rax, %rax
 ; X64-NEXT:    andl $1, %eax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r13 # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movl $0, %edi
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    sbbq %rbp, %rbp
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r9 # 8-byte Reload
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq {{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    andl $1, %ebp
-; X64-NEXT:    addq %r8, %rax
-; X64-NEXT:    adcq %rsi, %rdx
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
-; X64-NEXT:    movq %r13, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rdi, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
-; X64-NEXT:    movq %r9, %rsi
-; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, %rax
 ; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rcx, %r9
 ; X64-NEXT:    movq %rdx, %rbx
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %r8, %rcx
+; X64-NEXT:    addq %rsi, %rcx
 ; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq {{[0-9]+}}(%rsp) # 8-byte Folded Reload
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    addq %rcx, %r15
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rcx, %r13
 ; X64-NEXT:    adcq $0, %rsi
 ; X64-NEXT:    addq %rbx, %rsi
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r12 # 8-byte Reload
-; X64-NEXT:    addq %r14, %r12
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq %r8, %rsi
-; X64-NEXT:    adcq %rax, %r12
-; X64-NEXT:    addq %rdi, %rsi
-; X64-NEXT:    adcq %rbp, %r12
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r12 # 8-byte Reload
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
 ; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r8 # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %r10, %rbp
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    addq %rax, %r12
+; X64-NEXT:    adcq %rdx, %r8
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rdi, %r9
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %rbp, %rbx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %rcx, %rsi
+; X64-NEXT:    adcq %rdi, %r11
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq %r13, %r11
+; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %r9, %rdi
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rbx, %r9
+; X64-NEXT:    mulq %rcx
 ; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r10, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq -{{[0-9]+}}(%rsp) # 8-byte Folded Reload
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %rbx, %rcx
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
-; X64-NEXT:    addq %r11, %r9
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq %r10, %rdx
-; X64-NEXT:    adcq %rax, %r9
-; X64-NEXT:    addq %rdi, %rdx
-; X64-NEXT:    adcq %r8, %r9
-; X64-NEXT:    addq %r14, %r11
-; X64-NEXT:    adcq %r15, %rcx
-; X64-NEXT:    adcq %rsi, %rdx
-; X64-NEXT:    adcq %r12, %r9
-; X64-NEXT:    addq %r13, %r11
+; X64-NEXT:    addq %r8, %rcx
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    addq %rbp, %rbx
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rdi
+; X64-NEXT:    adcq %rdx, %rcx
+; X64-NEXT:    addq %rsi, %r13
+; X64-NEXT:    adcq %r11, %r8
+; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %r12, %rdi
+; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movl $0, %r12d
+; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    sbbq %r9, %r9
+; X64-NEXT:    andl $1, %r9d
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rbp, %rbx
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rsi
+; X64-NEXT:    adcq %rdx, %rcx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq %r12, %rsi
+; X64-NEXT:    adcq %r9, %rcx
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
 ; X64-NEXT:    addq %rax, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rax, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rax, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rax, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq %r15, {{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq %r10, %r13
+; X64-NEXT:    movq %r13, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r14, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    movq %rbx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
 ; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq (%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    movq 64(%rsi), %rcx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq 64(%rsi), %r14
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq 72(%rsi), %rcx
+; X64-NEXT:    movq %rsi, %r13
+; X64-NEXT:    movq %rbp, %rax
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
-; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %rcx, %rsi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rbx, %r10
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rdi, %rcx
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rsi, %r8
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq %rdi, %rsi
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    xorl %ecx, %ecx
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r14
 ; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r8, %rbx
-; X64-NEXT:    adcq $0, %r14
-; X64-NEXT:    movq 72(%rsi), %rbp
-; X64-NEXT:    movq %rsi, %r10
+; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r15 # 8-byte Reload
+; X64-NEXT:    addq %rbx, %r15
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r12 # 8-byte Reload
+; X64-NEXT:    adcq %r11, %r12
+; X64-NEXT:    addq %rbp, %r15
+; X64-NEXT:    adcq %rsi, %r12
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %r14, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    mulq %r14
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    addq %r14, %rsi
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    imulq %rbp, %rdi
-; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %r15
-; X64-NEXT:    movq %r15, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    imulq %rcx, %r9
-; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    leaq (%r11,%rdi), %rbp
+; X64-NEXT:    movq %rax, %rbp
 ; X64-NEXT:    addq %r9, %rbp
-; X64-NEXT:    addq %r8, %rsi
-; X64-NEXT:    adcq %rax, %rbp
+; X64-NEXT:    adcq $0, %rsi
 ; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    xorl %r14d, %r14d
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rcx, %r14
+; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbp, %rax
 ; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
-; X64-NEXT:    addq %rax, %r8
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rdx, %rax
-; X64-NEXT:    addq %rsi, %r8
-; X64-NEXT:    adcq %rbp, %rax
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq %rbx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r13 # 8-byte Reload
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rbx, %rsi
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
 ; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, %rbx
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r9, %r11
-; X64-NEXT:    movq %r11, (%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rax, %r12
-; X64-NEXT:    adcq $0, %r8
-; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 80(%r10), %rdi
-; X64-NEXT:    movq %rbx, %rsi
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    movq %r13, %r8
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %r11, %rbp
-; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq 88(%r10), %r15
+; X64-NEXT:    movq %rdi, %r9
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rbx
+; X64-NEXT:    adcq %rdx, %r11
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq %r10, %r11
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    movq 80(%r13), %rbp
+; X64-NEXT:    movq %r14, %rsi
 ; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r15
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rbp
 ; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %r8, %rcx
 ; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    addq %rbx, %r10
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    movq %rsi, %r13
-; X64-NEXT:    imulq %r15, %r13
-; X64-NEXT:    imulq %rdi, %r8
-; X64-NEXT:    leaq (%r9,%r13), %rbx
-; X64-NEXT:    addq %r8, %rbx
-; X64-NEXT:    addq %r11, %r10
-; X64-NEXT:    adcq %rax, %rbx
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    addq %r10, %r14
-; X64-NEXT:    adcq %rbx, %rdx
-; X64-NEXT:    addq %r9, (%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq %r12, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r14
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
+; X64-NEXT:    movq 88(%r13), %r13
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    addq %rcx, %r8
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    addq %r10, %rdi
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rdi, %rcx
+; X64-NEXT:    adcq %rsi, %r10
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    xorl %edx, %edx
+; X64-NEXT:    mulq %rdx
 ; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movl $0, %ebp
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    sbbq %r12, %r12
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq -{{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    adcq %rdx, %rax
+; X64-NEXT:    addq %rcx, %rsi
+; X64-NEXT:    adcq %r10, %rax
+; X64-NEXT:    addq %rbx, %r14
+; X64-NEXT:    movq %r14, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r11, %r8
+; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    adcq $0, %rax
+; X64-NEXT:    addq %r15, %rsi
+; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r12, %rax
 ; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    addq %r8, %r9
-; X64-NEXT:    addq %r13, %r9
+; X64-NEXT:    movl $0, %r15d
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    sbbq %r12, %r12
 ; X64-NEXT:    andl $1, %r12d
-; X64-NEXT:    addq %r11, %r10
-; X64-NEXT:    adcq %rbx, %r9
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    sbbq %r11, %r11
-; X64-NEXT:    movq %rsi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %r8, %rbx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    imulq %rsi, %rdx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
-; X64-NEXT:    addq %r8, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    imulq %rcx, %rbx
-; X64-NEXT:    addq %rdx, %rbx
-; X64-NEXT:    andl $1, %r11d
-; X64-NEXT:    movq %rax, %rdx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r11, %rbx
-; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rdx, %r10
-; X64-NEXT:    adcq %rbx, %r9
-; X64-NEXT:    addq %r8, %r14
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %r10
-; X64-NEXT:    movq %r10, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r12, %r9
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rsi
+; X64-NEXT:    adcq %rdx, %rcx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
+; X64-NEXT:    movq %r14, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r10, %rbx
+; X64-NEXT:    movq %rbx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r15, %rsi
+; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r12, %rcx
+; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %r15
+; X64-NEXT:    imulq %rax, %r13
 ; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r15, %rdx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r13 # 8-byte Reload
-; X64-NEXT:    imulq %r13, %rdi
-; X64-NEXT:    addq %rdx, %rdi
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %r13, %rdx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    imulq %rdi, %rbp
+; X64-NEXT:    addq %rdx, %rbp
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    imulq %rsi, %rbp
-; X64-NEXT:    movq %rsi, %r10
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    imulq %rbx, %rsi
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    imulq %rcx, %r11
-; X64-NEXT:    addq %rdx, %r11
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    adcq %rdi, %r11
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rsi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    imulq %rcx, %rax
+; X64-NEXT:    addq %rdx, %rax
+; X64-NEXT:    addq %r9, %r10
+; X64-NEXT:    adcq %rbp, %rax
+; X64-NEXT:    movq %rax, %r9
 ; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rcx, %rbp
 ; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %r15, %rbp
-; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r13
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    addq %rbx, %rdi
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    imulq %r13, %rcx
-; X64-NEXT:    imulq %r8, %r10
-; X64-NEXT:    addq %rsi, %rcx
-; X64-NEXT:    addq %rcx, %r10
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq %r15, %rdi
-; X64-NEXT:    adcq %rax, %r10
-; X64-NEXT:    addq %r12, %rdi
-; X64-NEXT:    adcq %r11, %r10
-; X64-NEXT:    movq %r10, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rbx, %r12
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %rbx, %r14
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %rsi, %rbp
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %r12, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    addq %rbp, %r8
+; X64-NEXT:    adcq %rcx, %r12
+; X64-NEXT:    addq %r10, %r8
+; X64-NEXT:    adcq %r9, %r12
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdx # 8-byte Reload
 ; X64-NEXT:    movq 120(%rdx), %rcx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %rcx
-; X64-NEXT:    movq 112(%rdx), %rbx
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
+; X64-NEXT:    imulq %r9, %rcx
+; X64-NEXT:    movq 112(%rdx), %rsi
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rax, %r15
 ; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    imulq %rsi, %rbx
-; X64-NEXT:    addq %rdx, %rbx
-; X64-NEXT:    movq 96(%r10), %rbp
-; X64-NEXT:    movq 104(%r10), %r8
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # 8-byte Reload
+; X64-NEXT:    imulq %r10, %rsi
+; X64-NEXT:    addq %rdx, %rsi
+; X64-NEXT:    movq 96(%rdi), %rbp
+; X64-NEXT:    movq 104(%rdi), %rbx
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    imulq %r8, %rcx
+; X64-NEXT:    imulq %rbx, %rcx
 ; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    movq %rax, %r13
 ; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rbp, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    addq %r13, %r15
-; X64-NEXT:    adcq %rbx, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    imulq %rbp, %r11
+; X64-NEXT:    addq %rdx, %r11
+; X64-NEXT:    addq %r15, %r13
+; X64-NEXT:    adcq %rsi, %r11
+; X64-NEXT:    movq %r11, %r15
 ; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %r13
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    addq %rbx, %r10
-; X64-NEXT:    adcq $0, %r13
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    adcq $0, %rdi
 ; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %r10, %rsi
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %r13, %rdx
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    imulq %rcx, %rbp
-; X64-NEXT:    imulq %r12, %r8
-; X64-NEXT:    addq %r11, %rbp
-; X64-NEXT:    addq %rbp, %r8
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq %rbx, %rdx
-; X64-NEXT:    adcq %rax, %r8
-; X64-NEXT:    addq %r15, %rdx
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r8 # 8-byte Folded Reload
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    addq %r13, %rax
+; X64-NEXT:    adcq %r15, %rdx
 ; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    adcq %rdi, %rdx
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r8 # 8-byte Folded Reload
-; X64-NEXT:    addq %r14, %r11
-; X64-NEXT:    movq %r11, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r14, %rbp
+; X64-NEXT:    adcq %r8, %rax
+; X64-NEXT:    adcq %r12, %rdx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
+; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r9, %r8
-; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
 ; X64-NEXT:    movq 80(%rsi), %rax
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
 ; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq %rdx, %r15
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq 88(%rsi), %rax
-; X64-NEXT:    movq %rsi, %r9
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rsi, %r11
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdi, %r14
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r8, %rbx
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rdi, %rbx
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rcx, %rdi
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    addq %rdi, %rsi
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, %r9
+; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %rdi, %r14
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    addq %rsi, %rbp
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rbp, %rcx
+; X64-NEXT:    adcq %rdi, %rsi
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    xorl %r13d, %r13d
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    imulq %rbp, %rdi
-; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %r10
-; X64-NEXT:    imulq %r14, %rcx
-; X64-NEXT:    movq %rcx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    leaq (%r15,%rdi), %rbp
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    movq (%rsp), %r10 # 8-byte Reload
+; X64-NEXT:    addq %r10, %r12
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r9 # 8-byte Reload
+; X64-NEXT:    adcq %r9, %r8
+; X64-NEXT:    addq %rcx, %r12
+; X64-NEXT:    adcq %rsi, %r8
+; X64-NEXT:    movq %r11, %rsi
+; X64-NEXT:    movq 64(%rsi), %rax
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rbx, %rdi
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq 72(%rsi), %rbx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
 ; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    addq %r8, %rsi
-; X64-NEXT:    adcq %rdx, %rbp
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    mulq %rcx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    mulq %r15
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq %rsi, %rdi
-; X64-NEXT:    adcq %rbp, %rcx
-; X64-NEXT:    movq %r9, %rsi
-; X64-NEXT:    movq 64(%rsi), %rax
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq %rdi, %rsi
+; X64-NEXT:    movq %r11, %rdi
+; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %r11
 ; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    movq %r14, %rbp
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq %rbx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    addq %r13, %r10
+; X64-NEXT:    adcq %r11, %r9
+; X64-NEXT:    addq %rbp, %r10
+; X64-NEXT:    adcq %rsi, %r9
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
+; X64-NEXT:    movq %r10, (%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r14, %r9
 ; X64-NEXT:    movq %r9, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 72(%rsi), %r8
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    movq %rbx, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %rbp
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rbx, %rsi
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r9, %r15
-; X64-NEXT:    movq %r15, %rbp
-; X64-NEXT:    adcq %rax, %r11
-; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    movq %r13, %rax
+; X64-NEXT:    addq %r9, %rsi
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r10 # 8-byte Reload
 ; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rsi, %rdi
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rbp, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %r12, %rsi
-; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rsi, %r14
-; X64-NEXT:    adcq $0, %r9
-; X64-NEXT:    addq %rbx, %r9
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    imulq %r15, %r13
-; X64-NEXT:    movq %r8, %r15
-; X64-NEXT:    imulq %r10, %r15
-; X64-NEXT:    leaq (%r11,%r13), %rsi
-; X64-NEXT:    addq %r15, %rsi
-; X64-NEXT:    addq %r12, %r9
-; X64-NEXT:    adcq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    xorl %ebx, %ebx
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
-; X64-NEXT:    addq %rax, %r8
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    adcq %rdx, %rax
-; X64-NEXT:    addq %r9, %r8
-; X64-NEXT:    adcq %rsi, %rax
-; X64-NEXT:    addq %r11, %rbp
-; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    movq %r14, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    adcq %r14, %r11
+; X64-NEXT:    addq %rax, %r13
+; X64-NEXT:    adcq %rdx, %r11
+; X64-NEXT:    addq (%rsp), %r15 # 8-byte Folded Reload
+; X64-NEXT:    movq %r15, (%rsp) # 8-byte Spill
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
+; X64-NEXT:    movq %rdi, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %r13
+; X64-NEXT:    adcq $0, %r11
+; X64-NEXT:    addq %r12, %r13
+; X64-NEXT:    adcq %r8, %r11
+; X64-NEXT:    movl $0, %r8d
 ; X64-NEXT:    adcq $0, %r8
-; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    addq %rdi, %r8
-; X64-NEXT:    adcq %rcx, %rax
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    adcq $0, %rbx
 ; X64-NEXT:    sbbq %r9, %r9
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq -{{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    addq %r15, %r14
-; X64-NEXT:    addq %r13, %r14
 ; X64-NEXT:    andl $1, %r9d
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    adcq %rdi, %r14
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    sbbq %rdx, %rdx
-; X64-NEXT:    movq %r12, %rcx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r11 # 8-byte Reload
-; X64-NEXT:    imulq %r11, %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
-; X64-NEXT:    addq %r15, %rcx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r13 # 8-byte Reload
-; X64-NEXT:    imulq %rbp, %r13
-; X64-NEXT:    addq %rcx, %r13
-; X64-NEXT:    andl $1, %edx
-; X64-NEXT:    movq %rdi, %rcx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rdx, %r13
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    adcq %r13, %r14
-; X64-NEXT:    addq %r15, %r8
-; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rbx, %rax
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r9, %r14
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    movq 96(%rsi), %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    imulq %rcx, %rdi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r15 # 8-byte Reload
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %r12, %rbp
+; X64-NEXT:    adcq $0, %rsi
 ; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rdi, %rdx
-; X64-NEXT:    movq 104(%rsi), %r15
-; X64-NEXT:    imulq %r15, %r10
-; X64-NEXT:    addq %rdx, %r10
-; X64-NEXT:    movq 112(%rsi), %rax
-; X64-NEXT:    movq %rsi, %rbx
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbp, %rax
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    movq %r11, %rdi
-; X64-NEXT:    imulq %rdi, %rbp
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    movq 120(%rbx), %rbx
-; X64-NEXT:    imulq %rsi, %rbx
-; X64-NEXT:    addq %rdx, %rbx
-; X64-NEXT:    addq %r8, %r11
-; X64-NEXT:    adcq %r10, %rbx
-; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq %r14, %rcx
+; X64-NEXT:    addq %rax, %rsi
+; X64-NEXT:    adcq %rdx, %rcx
+; X64-NEXT:    addq %r13, %rdi
+; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r11, %rbp
+; X64-NEXT:    movq %rbp, {{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r8, %rsi
+; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r9, %rcx
+; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq 96(%rbp), %rcx
+; X64-NEXT:    imulq %rcx, %r10
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %r10, %rdx
+; X64-NEXT:    movq 104(%rbp), %r8
+; X64-NEXT:    imulq %r8, %r15
+; X64-NEXT:    addq %rdx, %r15
+; X64-NEXT:    movq 112(%rbp), %rax
+; X64-NEXT:    movq %rbp, %rdi
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    imulq %rbx, %rsi
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rsi, %rdx
+; X64-NEXT:    movq 120(%rdi), %rdi
+; X64-NEXT:    imulq %rbp, %rdi
+; X64-NEXT:    addq %rdx, %rdi
+; X64-NEXT:    addq %r9, %r13
+; X64-NEXT:    adcq %r15, %rdi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, %r9
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %r8, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %rbp, %rsi
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %r10, %r9
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rsi, %r12
 ; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    addq %r9, %rax
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    addq %rbp, %rdi
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    movq %rsi, %rbp
-; X64-NEXT:    imulq %r15, %rbp
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdx # 8-byte Reload
-; X64-NEXT:    imulq %rcx, %rdx
-; X64-NEXT:    addq %r8, %rbp
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq %r10, %rdi
-; X64-NEXT:    adcq %rax, %rdx
-; X64-NEXT:    addq %r11, %rdi
-; X64-NEXT:    adcq %rbx, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %rbp, %r9
+; X64-NEXT:    adcq %rcx, %r8
+; X64-NEXT:    addq %r13, %r9
+; X64-NEXT:    adcq %rdi, %r8
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    imulq %rax, %rbx
+; X64-NEXT:    imulq %rbx, %rsi
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rbx, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
-; X64-NEXT:    imulq %r8, %rcx
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rsi, %rdx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r11 # 8-byte Reload
+; X64-NEXT:    imulq %r11, %rcx
 ; X64-NEXT:    addq %rdx, %rcx
-; X64-NEXT:    movq %rcx, %r11
+; X64-NEXT:    movq %rcx, %rsi
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
 ; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    imulq %r14, %rcx
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    imulq %rbp, %rcx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %r13
 ; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r15 # 8-byte Reload
-; X64-NEXT:    imulq %rbx, %r15
-; X64-NEXT:    addq %rdx, %r15
-; X64-NEXT:    addq %rsi, %r9
-; X64-NEXT:    adcq %r11, %r15
-; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    movq %rbx, %rcx
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
+; X64-NEXT:    imulq %rbp, %rax
+; X64-NEXT:    addq %rdx, %rax
+; X64-NEXT:    addq %r10, %r13
+; X64-NEXT:    adcq %rsi, %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %rbx
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %r10, %rsi
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    sbbq %rsi, %rsi
-; X64-NEXT:    imulq %r8, %rcx
-; X64-NEXT:    imulq %r12, %rbx
-; X64-NEXT:    addq %r11, %rcx
-; X64-NEXT:    addq %rcx, %rbx
-; X64-NEXT:    andl $1, %esi
-; X64-NEXT:    addq %r10, %rdx
-; X64-NEXT:    adcq %rsi, %rbx
-; X64-NEXT:    addq %r9, %rdx
-; X64-NEXT:    adcq %r15, %rbx
-; X64-NEXT:    movq %rbx, %rbp
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    adcq %rdi, %rdx
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
+; X64-NEXT:    movq %rbp, %r10
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %r15
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rcx, %r10
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    addq %rdi, %r15
+; X64-NEXT:    sbbq %rcx, %rcx
+; X64-NEXT:    andl $1, %ecx
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    addq %r15, %rax
+; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    addq %r13, %rax
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    adcq %r14, %rbp
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
+; X64-NEXT:    adcq %r12, %r10
+; X64-NEXT:    adcq %r9, %rax
+; X64-NEXT:    adcq %r8, %rdx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %r8 # 8-byte Reload
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    adcq %r8, %rdi
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq %rdi, %r9
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    adcq %rdi, %rsi
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    adcq %r13, %rsi
-; X64-NEXT:    addq %rcx, %rbx
-; X64-NEXT:    movq %rbx, %r12
-; X64-NEXT:    adcq %r9, %rsi
-; X64-NEXT:    movq %rsi, %r15
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    adcq (%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    movq (%rsp), %rbx # 8-byte Reload
 ; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %r12 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r15 # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, %r10
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
+; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    movq %rcx, %r8
+; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
+; X64-NEXT:    movq %rdi, %r9
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    movq %rbx, %r14
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %r11 # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
-; X64-NEXT:    addq %r8, %rdi
 ; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, %r9
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq %rsi, %r8
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    movq %rsi, 32(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, (%rcx)
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    movq %rsi, 40(%rcx)
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    movq %rsi, 48(%rcx)
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    movq %rsi, 56(%rcx)
-; X64-NEXT:    movq %r9, 8(%rcx)
-; X64-NEXT:    movq %rbx, 16(%rcx)
-; X64-NEXT:    movq %r12, 64(%rcx)
-; X64-NEXT:    movq %r15, 72(%rcx)
-; X64-NEXT:    movq %r8, 24(%rcx)
-; X64-NEXT:    movq %r10, 80(%rcx)
-; X64-NEXT:    movq %r14, 88(%rcx)
-; X64-NEXT:    movq %r11, 96(%rcx)
-; X64-NEXT:    movq %rax, 104(%rcx)
-; X64-NEXT:    movq %rdx, 112(%rcx)
-; X64-NEXT:    movq %rbp, 120(%rcx)
-; X64-NEXT:    addq $496, %rsp # imm = 0x1F0
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 8(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 16(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 24(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 32(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 40(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 48(%rcx)
+; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, 56(%rcx)
+; X64-NEXT:    movq %r8, 64(%rcx)
+; X64-NEXT:    movq %r9, 72(%rcx)
+; X64-NEXT:    movq %rbx, 80(%rcx)
+; X64-NEXT:    movq %rsi, 88(%rcx)
+; X64-NEXT:    movq %rbp, 96(%rcx)
+; X64-NEXT:    movq %r10, 104(%rcx)
+; X64-NEXT:    movq %rax, 112(%rcx)
+; X64-NEXT:    movq %rdx, 120(%rcx)
+; X64-NEXT:    addq $360, %rsp # imm = 0x168
 ; X64-NEXT:    popq %rbx
 ; X64-NEXT:    popq %r12
 ; X64-NEXT:    popq %r13
diff --git a/llvm/test/CodeGen/X86/mul-i256.ll b/llvm/test/CodeGen/X86/mul-i256.ll
index 8b8b10aa1790..8f207b8dd086 100644
--- a/llvm/test/CodeGen/X86/mul-i256.ll
+++ b/llvm/test/CodeGen/X86/mul-i256.ll
@@ -15,12 +15,17 @@ entry:
 ; There is a lot of inter-register motion, and so matching the instruction
 ; sequence will be fragile. There should be 6 underlying multiplications.
 ; CHECK: imulq
+; CHECK: mulq
 ; CHECK: imulq
 ; CHECK: imulq
+; CHECK: mulq
 ; CHECK: imulq
-; CHECK: imulq
-; CHECK: imulq
+; CHECK: mulq
+; CHECK: mulq
+; CHECK: mulq
+; CHECK: mulq
 ; CHECK-NOT: imulq
+; CHECK-NOT: mulq
 ; CHECK: retq
 
 attributes #0 = { norecurse nounwind uwtable "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" }
diff --git a/llvm/test/CodeGen/X86/mul-i512.ll b/llvm/test/CodeGen/X86/mul-i512.ll
index 7b17be516177..f9cea6d2ba32 100644
--- a/llvm/test/CodeGen/X86/mul-i512.ll
+++ b/llvm/test/CodeGen/X86/mul-i512.ll
@@ -11,193 +11,145 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    andl $-8, %esp
-; X32-NEXT:    subl $632, %esp # imm = 0x278
-; X32-NEXT:    movl 12(%ebp), %ecx
-; X32-NEXT:    movl 8(%ebp), %ebx
-; X32-NEXT:    movl (%ebx), %edi
-; X32-NEXT:    movl 4(%ebx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 8(%ebx), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 12(%ebx), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 8(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 12(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl (%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 4(%ecx), %ecx
+; X32-NEXT:    subl $656, %esp # imm = 0x290
+; X32-NEXT:    movl 8(%ebp), %eax
+; X32-NEXT:    movl 48(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ecx
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edx
-; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    calll __multi3
-; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    calll __multi3
-; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %eax
-; X32-NEXT:    calll __multi3
-; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    seto %al
-; X32-NEXT:    lahf
-; X32-NEXT:    movl %eax, %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    movl 52(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 60(%ebx), %ecx
+; X32-NEXT:    movl 56(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 56(%ebx), %ecx
+; X32-NEXT:    movl 60(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 52(%ebx), %ecx
+; X32-NEXT:    movl 40(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 48(%ebx), %ecx
+; X32-NEXT:    movl 44(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 44(%ebx), %ecx
+; X32-NEXT:    movl 32(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 40(%ebx), %ecx
+; X32-NEXT:    movl 36(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 36(%ebx), %ecx
+; X32-NEXT:    movl (%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 32(%ebx), %ecx
+; X32-NEXT:    movl 4(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 28(%ebx), %ecx
+; X32-NEXT:    movl 16(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 24(%ebx), %ecx
+; X32-NEXT:    movl 20(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 20(%ebx), %ecx
+; X32-NEXT:    movl 8(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 16(%ebx), %ecx
+; X32-NEXT:    movl 12(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %eax, %eax
-; X32-NEXT:    addb $127, %al
-; X32-NEXT:    sahf
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    movl 24(%eax), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl 28(%eax), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 12(%ebp), %ecx
-; X32-NEXT:    movl 28(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 24(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 20(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 16(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 44(%ecx), %edi
-; X32-NEXT:    movl 40(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 36(%ecx), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl 32(%ecx), %eax
+; X32-NEXT:    movl 12(%ebp), %eax
+; X32-NEXT:    movl 48(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 52(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 56(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 60(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 32(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 36(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 40(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 44(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl (%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 4(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 8(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 12(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl 16(%eax), %esi
+; X32-NEXT:    movl 20(%eax), %edi
+; X32-NEXT:    movl 24(%eax), %ebx
+; X32-NEXT:    movl 28(%eax), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl 60(%ecx)
-; X32-NEXT:    pushl 56(%ecx)
-; X32-NEXT:    pushl 52(%ecx)
-; X32-NEXT:    pushl 48(%ecx)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %eax
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    pushl %ecx
-; X32-NEXT:    movl %ecx, %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl %esi, %ebx
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -205,7 +157,9 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
@@ -215,133 +169,167 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl %esi, %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
@@ -351,50 +339,50 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -402,7 +390,7 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
@@ -413,42 +401,40 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %ebx
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    pushl %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
@@ -459,110 +445,92 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %esi
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    pushl %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
 ; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    pushl %edi
-; X32-NEXT:    pushl $0
-; X32-NEXT:    pushl $0
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
 ; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl $0
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    pushl %esi
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll __multi3
 ; X32-NEXT:    addl $32, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl $0, %edx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    sbbl %esi, %esi
-; X32-NEXT:    andl $1, %esi
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    movl $0, %edi
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    sbbl %edx, %edx
+; X32-NEXT:    andl $1, %edx
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    addl %edi, %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    addl %edx, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl %edx, %edi
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    addl %eax, %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, %ecx
@@ -589,29 +557,88 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    addl %esi, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    addl %edi, %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %eax, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl %ecx, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl %edx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edx
-; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    movl $0, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    sbbl %ebx, %ebx
+; X32-NEXT:    andl $1, %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl %edi, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl %ebx, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -622,18 +649,21 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl %edx, %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    movl %esi, %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edi, %esi
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
@@ -646,53 +676,65 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    sbbl %eax, %eax
 ; X32-NEXT:    andl $1, %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl $0, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    sbbl %ebx, %ebx
-; X32-NEXT:    andl $1, %ebx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl $0, %edx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    sbbl %esi, %esi
+; X32-NEXT:    andl $1, %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %eax, %edi
+; X32-NEXT:    adcl %ecx, %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl %edx, {{[0-9]+}}(%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[0-9]+}}(%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
@@ -750,112 +792,108 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %ecx, %ebx
+; X32-NEXT:    addl %eax, %esi
+; X32-NEXT:    adcl %ecx, %edi
 ; X32-NEXT:    movl $0, %eax
 ; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    sbbl %edx, %edx
-; X32-NEXT:    andl $1, %edx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    sbbl %ecx, %ecx
+; X32-NEXT:    andl $1, %ecx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    leal {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl {{[0-9]+}}(%esp) # 4-byte Folded Reload
+; X32-NEXT:    pushl %eax
+; X32-NEXT:    calll __multi3
+; X32-NEXT:    addl $32, %esp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    addl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    addl %edi, %ecx
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
+; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    addl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[0-9]+}}(%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[0-9]+}}(%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl 16(%ebp), %eax
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, 16(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, 20(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, 24(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, 28(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, (%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, 4(%eax)
-; X32-NEXT:    movl %ecx, 8(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 32(%eax)
-; X32-NEXT:    movl %edx, 12(%eax)
-; X32-NEXT:    movl %edi, 36(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 40(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 44(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 48(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 52(%eax)
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, 56(%eax)
-; X32-NEXT:    movl %esi, 60(%eax)
+; X32-NEXT:    adcl {{[0-9]+}}(%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, {{[0-9]+}}(%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 16(%ebp), %edi
+; X32-NEXT:    movl %esi, 4(%edi)
+; X32-NEXT:    movl 16(%ebp), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, (%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 8(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 12(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 16(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 20(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 24(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 28(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 32(%esi)
+; X32-NEXT:    movl %ebx, 36(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 40(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 44(%esi)
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, 48(%esi)
+; X32-NEXT:    movl %ecx, 52(%esi)
+; X32-NEXT:    movl %edx, 56(%esi)
+; X32-NEXT:    movl %eax, 60(%esi)
 ; X32-NEXT:    leal -12(%ebp), %esp
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    popl %edi
@@ -871,323 +909,320 @@ define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X64-NEXT:    pushq %r13
 ; X64-NEXT:    pushq %r12
 ; X64-NEXT:    pushq %rbx
-; X64-NEXT:    subq $56, %rsp
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 24(%rdi), %rcx
-; X64-NEXT:    movq 16(%rdi), %r15
+; X64-NEXT:    pushq %rax
+; X64-NEXT:    movq %rdx, (%rsp) # 8-byte Spill
+; X64-NEXT:    movq 24(%rdi), %rbp
+; X64-NEXT:    movq 16(%rdi), %rcx
 ; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq (%rsi), %r11
-; X64-NEXT:    movq 8(%rsi), %rbp
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq (%rsi), %rdx
+; X64-NEXT:    movq 8(%rsi), %r8
 ; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r14
-; X64-NEXT:    movq %r14, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rsi, %r9
 ; X64-NEXT:    movq %rdx, %rbx
 ; X64-NEXT:    movq %rax, %rsi
 ; X64-NEXT:    addq %r10, %rsi
 ; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rcx, %r11
+; X64-NEXT:    movq %r11, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rsi, %r9
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rsi, %r12
 ; X64-NEXT:    adcq $0, %rcx
 ; X64-NEXT:    addq %rbx, %rcx
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    movq %r15, %rbx
+; X64-NEXT:    sbbq %rbx, %rbx
+; X64-NEXT:    andl $1, %ebx
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq %rbx, %rsi
+; X64-NEXT:    xorl %ecx, %ecx
+; X64-NEXT:    movq %r9, %rbx
 ; X64-NEXT:    movq %rbx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbx, %rdx
-; X64-NEXT:    imulq %rbp, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %r12
-; X64-NEXT:    movq %r12, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    leaq (%r8,%rdx), %rbp
-; X64-NEXT:    imulq %r11, %r14
-; X64-NEXT:    movq %r14, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r14, %rbp
-; X64-NEXT:    addq %r10, %rcx
-; X64-NEXT:    adcq %rax, %rbp
-; X64-NEXT:    xorl %r14d, %r14d
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq %r10, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq %rsi, {{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rsi, %rbx
-; X64-NEXT:    adcq %r10, %r14
-; X64-NEXT:    addq %rcx, %rbx
-; X64-NEXT:    adcq %rbp, %r14
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r13
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    movq %r15, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    addq %r10, %r15
+; X64-NEXT:    adcq %r13, %r9
+; X64-NEXT:    addq %rbp, %r15
+; X64-NEXT:    adcq %rsi, %r9
 ; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq (%rdi), %rax
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    movq 8(%rdi), %rax
 ; X64-NEXT:    movq %rax, %rcx
 ; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %r11, %rsi
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    addq %rbp, %rbx
+; X64-NEXT:    sbbq %rdi, %rdi
+; X64-NEXT:    andl $1, %edi
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq 8(%rdi), %rax
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    movq %rdx, (%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %r12
+; X64-NEXT:    addq %rbx, %rbp
+; X64-NEXT:    adcq %rdi, %rsi
+; X64-NEXT:    movq %r14, %rcx
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    xorl %edx, %edx
+; X64-NEXT:    mulq %rdx
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    addq %r11, %r10
+; X64-NEXT:    adcq %r14, %r13
+; X64-NEXT:    addq %rbp, %r10
+; X64-NEXT:    adcq %rsi, %r13
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r10 # 8-byte Folded Reload
+; X64-NEXT:    adcq %r12, %r13
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    adcq $0, %r9
+; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    movq 16(%rsi), %r8
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rcx, %r9
+; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rdi, %rbp
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq 24(%rsi), %rdi
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rsi
 ; X64-NEXT:    addq %rbp, %rax
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    addq %rbx, %rsi
+; X64-NEXT:    sbbq %rbp, %rbp
+; X64-NEXT:    andl $1, %ebp
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %rsi, %r9
+; X64-NEXT:    adcq %rbp, %rbx
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    xorl %ecx, %ecx
+; X64-NEXT:    mulq %rcx
 ; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rsi, %r8
-; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rax, %r9
-; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rbp, %r11
+; X64-NEXT:    adcq %rdx, %r14
+; X64-NEXT:    addq %r9, %r11
+; X64-NEXT:    adcq %rbx, %r14
+; X64-NEXT:    addq %r10, %r12
+; X64-NEXT:    movq %r12, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r13, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
+; X64-NEXT:    adcq $0, %r11
 ; X64-NEXT:    adcq $0, %r14
+; X64-NEXT:    addq %r15, %r11
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    movq %rcx, %r13
+; X64-NEXT:    sbbq %r9, %r9
+; X64-NEXT:    andl $1, %r9d
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq 16(%rcx), %r10
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, %r12
-; X64-NEXT:    movq %r12, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %r15, %rdi
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq 24(%rcx), %rsi
-; X64-NEXT:    movq %r13, %rcx
 ; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %r13
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rdi, %r9
-; X64-NEXT:    adcq $0, %r13
-; X64-NEXT:    addq %rbp, %r13
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    movq %rcx, %rdx
-; X64-NEXT:    movq %rcx, %rdi
-; X64-NEXT:    imulq %rsi, %rdx
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %r12, %rbp
-; X64-NEXT:    imulq %r10, %rbp
-; X64-NEXT:    leaq (%r8,%rdx), %rcx
-; X64-NEXT:    addq %rbp, %rcx
-; X64-NEXT:    addq %r15, %r13
-; X64-NEXT:    adcq %rax, %rcx
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    xorl %esi, %esi
-; X64-NEXT:    mulq %rsi
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %r15
 ; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    xorl %esi, %esi
-; X64-NEXT:    movq %rdx, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, {{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rax, %r12
-; X64-NEXT:    adcq %rdx, %r15
-; X64-NEXT:    addq %r13, %r12
-; X64-NEXT:    adcq %rcx, %r15
-; X64-NEXT:    addq %r8, -{{[0-9]+}}(%rsp) # 8-byte Folded Spill
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movq %r9, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r12
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    adcq %r14, %r15
-; X64-NEXT:    movq %r15, %r8
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rsi, %r13
-; X64-NEXT:    sbbq %r15, %r15
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq -{{[0-9]+}}(%rsp) # 8-byte Folded Reload
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
 ; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    movq %rdi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %r15, %rbx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    addq %rbp, %r14
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r14 # 8-byte Folded Reload
-; X64-NEXT:    andl $1, %r15d
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    adcq %rbx, %r14
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    addq (%rsp), %rbx # 8-byte Folded Reload
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rsi, %rcx
 ; X64-NEXT:    sbbq %rsi, %rsi
-; X64-NEXT:    movq %rdi, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    imulq %rdi, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    imulq %r11, %rbp
-; X64-NEXT:    addq %rdx, %rbp
 ; X64-NEXT:    andl $1, %esi
-; X64-NEXT:    movq %rbx, %rdx
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %rsi, %rbp
-; X64-NEXT:    movq %rbp, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    adcq %rbp, %r14
-; X64-NEXT:    addq %rcx, %r12
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %r8 # 8-byte Folded Reload
-; X64-NEXT:    movq %r8, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r13, %rax
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r15, %r14
+; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rsi # 8-byte Reload
+; X64-NEXT:    addq %rbp, %rsi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
+; X64-NEXT:    addq %rax, %rsi
+; X64-NEXT:    adcq %rdx, %rcx
+; X64-NEXT:    addq %r11, %r12
+; X64-NEXT:    movq %r12, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r14, %rbx
+; X64-NEXT:    movq %rbx, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r13, %rsi
+; X64-NEXT:    movq %rsi, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    adcq %r9, %rcx
+; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    movq 32(%rcx), %rsi
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    imulq %rsi, %rbp
+; X64-NEXT:    imulq %rsi, %rdi
 ; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rbp, %rdx
-; X64-NEXT:    movq 40(%rcx), %r8
-; X64-NEXT:    imulq %r8, %r10
-; X64-NEXT:    addq %rdx, %r10
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    addq %rdi, %rdx
+; X64-NEXT:    movq 40(%rcx), %r9
+; X64-NEXT:    imulq %r9, %r8
+; X64-NEXT:    addq %rdx, %r8
 ; X64-NEXT:    movq 48(%rcx), %rax
-; X64-NEXT:    movq %rcx, %rbp
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    imulq %rdi, %rcx
-; X64-NEXT:    movq %rdi, %r9
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq 56(%rbp), %rdi
-; X64-NEXT:    imulq %r11, %rdi
-; X64-NEXT:    addq %rdx, %rdi
-; X64-NEXT:    addq %rbx, %r15
-; X64-NEXT:    adcq %r10, %rdi
-; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    movq %rcx, %rbx
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    imulq %rcx, %rdi
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rdi, %rdx
+; X64-NEXT:    movq 56(%rbx), %rbx
+; X64-NEXT:    imulq %rbp, %rbx
+; X64-NEXT:    addq %rdx, %rbx
+; X64-NEXT:    addq %r11, %r12
+; X64-NEXT:    adcq %r8, %rbx
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, %r8
 ; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq %r10, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %r13, %rcx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rdi, %rbp
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    addq %rbp, %r11
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    addq %rsi, %rdi
+; X64-NEXT:    sbbq %rsi, %rsi
+; X64-NEXT:    andl $1, %esi
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    addq %rdi, %r15
+; X64-NEXT:    adcq %rsi, %r14
+; X64-NEXT:    addq %r12, %r15
+; X64-NEXT:    adcq %rbx, %r14
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rdx # 8-byte Reload
+; X64-NEXT:    movq 56(%rdx), %rcx
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
+; X64-NEXT:    imulq %r8, %rcx
+; X64-NEXT:    movq 48(%rdx), %rbp
 ; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    addq %rbp, %rbx
-; X64-NEXT:    sbbq %rax, %rax
-; X64-NEXT:    imulq %r8, %r11
-; X64-NEXT:    movq %r9, %rcx
-; X64-NEXT:    imulq %rsi, %rcx
-; X64-NEXT:    addq %r10, %r11
-; X64-NEXT:    addq %r11, %rcx
-; X64-NEXT:    andl $1, %eax
-; X64-NEXT:    addq %r13, %rbx
-; X64-NEXT:    adcq %rax, %rcx
-; X64-NEXT:    addq %r15, %rbx
-; X64-NEXT:    adcq %rdi, %rcx
-; X64-NEXT:    movq %rcx, -{{[0-9]+}}(%rsp) # 8-byte Spill
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq 56(%rcx), %rsi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %rcx, %rdx
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r10 # 8-byte Reload
-; X64-NEXT:    imulq %r10, %rsi
-; X64-NEXT:    movq 48(%rcx), %rdi
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rsi, %rdx
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r8 # 8-byte Reload
-; X64-NEXT:    imulq %r8, %rdi
-; X64-NEXT:    addq %rdx, %rdi
-; X64-NEXT:    movq 32(%rcx), %rsi
-; X64-NEXT:    movq 40(%rcx), %rbp
+; X64-NEXT:    imulq %r10, %rbp
+; X64-NEXT:    addq %rdx, %rbp
+; X64-NEXT:    movq 32(%rbx), %rdi
+; X64-NEXT:    movq 40(%rbx), %r12
 ; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rax # 8-byte Reload
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    imulq %rbp, %rcx
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    imulq %r12, %rcx
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rax, %r9
 ; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rsi, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    addq %r9, %r15
-; X64-NEXT:    adcq %rdi, %rax
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %r13 # 8-byte Reload
+; X64-NEXT:    imulq %rdi, %r13
+; X64-NEXT:    addq %rdx, %r13
+; X64-NEXT:    addq %rsi, %r9
+; X64-NEXT:    adcq %rbp, %r13
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    movq %r12, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %r13
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rbp, %rdi
+; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    addq %rbx, %rcx
+; X64-NEXT:    sbbq %rbp, %rbp
+; X64-NEXT:    andl $1, %ebp
+; X64-NEXT:    movq %r12, %rax
 ; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %r9, %rcx
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r8
 ; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    addq %rdi, %rdx
-; X64-NEXT:    sbbq %rcx, %rcx
-; X64-NEXT:    imulq %r8, %rsi
-; X64-NEXT:    imulq %r10, %rbp
-; X64-NEXT:    addq %r13, %rsi
-; X64-NEXT:    addq %rsi, %rbp
-; X64-NEXT:    andl $1, %ecx
-; X64-NEXT:    addq %r9, %rdx
-; X64-NEXT:    adcq %rcx, %rbp
-; X64-NEXT:    addq %r15, %rdx
-; X64-NEXT:    adcq %r11, %rbp
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %r13 # 8-byte Folded Reload
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
-; X64-NEXT:    adcq %rbx, %rdx
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rbp # 8-byte Folded Reload
-; X64-NEXT:    addq %r12, %r13
+; X64-NEXT:    adcq %rbp, %rdx
+; X64-NEXT:    addq %r9, %rax
+; X64-NEXT:    adcq %r13, %rdx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq %r11, %rdi
+; X64-NEXT:    adcq %r15, %rax
+; X64-NEXT:    adcq %r14, %rdx
+; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdi # 8-byte Folded Reload
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rax # 8-byte Folded Reload
 ; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rdx # 8-byte Folded Reload
-; X64-NEXT:    adcq %r14, %rbp
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rsi # 8-byte Reload
-; X64-NEXT:    addq {{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    adcq {{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq -{{[0-9]+}}(%rsp), %rsi # 8-byte Folded Reload
-; X64-NEXT:    movq %rsi, %rdi
-; X64-NEXT:    adcq -{{[0-9]+}}(%rsp), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, %rsi
-; X64-NEXT:    movq {{[0-9]+}}(%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    movq %rbx, 16(%rcx)
-; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbx # 8-byte Reload
-; X64-NEXT:    movq %rbx, 24(%rcx)
-; X64-NEXT:    movq %rdi, (%rcx)
-; X64-NEXT:    movq %r13, 32(%rcx)
-; X64-NEXT:    movq %rsi, 8(%rcx)
-; X64-NEXT:    movq %rax, 40(%rcx)
-; X64-NEXT:    movq %rdx, 48(%rcx)
-; X64-NEXT:    movq %rbp, 56(%rcx)
-; X64-NEXT:    addq $56, %rsp
+; X64-NEXT:    movq (%rsp), %rcx # 8-byte Reload
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, (%rcx)
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, 8(%rcx)
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, 16(%rcx)
+; X64-NEXT:    movq -{{[0-9]+}}(%rsp), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, 24(%rcx)
+; X64-NEXT:    movq %rsi, 32(%rcx)
+; X64-NEXT:    movq %rdi, 40(%rcx)
+; X64-NEXT:    movq %rax, 48(%rcx)
+; X64-NEXT:    movq %rdx, 56(%rcx)
+; X64-NEXT:    addq $8, %rsp
 ; X64-NEXT:    popq %rbx
 ; X64-NEXT:    popq %r12
 ; X64-NEXT:    popq %r13
author	Pawel Bylica <chfast@gmail.com>	2016-11-28 22:45:01 +0000
committer	Pawel Bylica <chfast@gmail.com>	2016-11-28 22:45:01 +0000
commit	cf874a215545dd6b8685b5a28ec0ff3e04d87b3b (patch)
tree	57abba27e9710adf4328f4f2e8bf959819cce4e1
parent	97214e4fb8f12738c3b1cc0d4af6d7f579986068 (diff)
download	llvm-cf874a215545dd6b8685b5a28ec0ff3e04d87b3b.tar.gz