[WebAssembly] Narrowing and widening SIMD ops

Summary:
Implements target-specific LLVM intrinsics and clang builtins for
these new SIMD operations, as described at https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md#integer-to-integer-narrowing.

Reviewers: aheejin

Subscribers: dschuff, sbc100, jgravelle-google, hiraditya, sunfish, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D67425

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@371906 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 147 insertions, 0 deletions

diff --git a/include/clang/Basic/BuiltinsWebAssembly.def b/include/clang/Basic/BuiltinsWebAssembly.def
index 37f945a319..f825040d61 100644
--- a/include/clang/Basic/BuiltinsWebAssembly.def
+++ b/include/clang/Basic/BuiltinsWebAssembly.def
@@ -118,5 +118,19 @@ TARGET_BUILTIN(__builtin_wasm_trunc_saturate_u_i32x4_f32x4, "V4iV4f", "nc", "sim
 TARGET_BUILTIN(__builtin_wasm_trunc_saturate_s_i64x2_f64x2, "V2LLiV2d", "nc", "unimplemented-simd128")
 TARGET_BUILTIN(__builtin_wasm_trunc_saturate_u_i64x2_f64x2, "V2LLiV2d", "nc", "unimplemented-simd128")
 
+TARGET_BUILTIN(__builtin_wasm_narrow_s_i8x16_i16x8, "V16cV8sV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_narrow_u_i8x16_i16x8, "V16cV8sV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_narrow_s_i16x8_i32x4, "V8sV4iV4i", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_narrow_u_i16x8_i32x4, "V8sV4iV4i", "nc", "simd128")
+
+TARGET_BUILTIN(__builtin_wasm_widen_low_s_i16x8_i8x16, "V8sV16c", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_high_s_i16x8_i8x16, "V8sV16c", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_low_u_i16x8_i8x16, "V8sV16c", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_high_u_i16x8_i8x16, "V8sV16c", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_low_s_i32x4_i16x8, "V4iV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_high_s_i32x4_i16x8, "V4iV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_low_u_i32x4_i16x8, "V4iV8s", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_high_u_i32x4_i16x8, "V4iV8s", "nc", "simd128")
+
 #undef BUILTIN
 #undef TARGET_BUILTIN
diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 1b2468d336..f40ced06e2 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -14196,6 +14196,63 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
     Function *Callee = CGM.getIntrinsic(IntNo, A->getType());
     return Builder.CreateCall(Callee, {A, B, C});
   }
+  case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8:
+  case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8:
+  case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4:
+  case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4: {
+    Value *Low = EmitScalarExpr(E->getArg(0));
+    Value *High = EmitScalarExpr(E->getArg(1));
+    unsigned IntNo;
+    switch (BuiltinID) {
+    case WebAssembly::BI__builtin_wasm_narrow_s_i8x16_i16x8:
+    case WebAssembly::BI__builtin_wasm_narrow_s_i16x8_i32x4:
+      IntNo = Intrinsic::wasm_narrow_signed;
+      break;
+    case WebAssembly::BI__builtin_wasm_narrow_u_i8x16_i16x8:
+    case WebAssembly::BI__builtin_wasm_narrow_u_i16x8_i32x4:
+      IntNo = Intrinsic::wasm_narrow_unsigned;
+      break;
+    default:
+      llvm_unreachable("unexpected builtin ID");
+    }
+    Function *Callee =
+        CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Low->getType()});
+    return Builder.CreateCall(Callee, {Low, High});
+  }
+  case WebAssembly::BI__builtin_wasm_widen_low_s_i16x8_i8x16:
+  case WebAssembly::BI__builtin_wasm_widen_high_s_i16x8_i8x16:
+  case WebAssembly::BI__builtin_wasm_widen_low_u_i16x8_i8x16:
+  case WebAssembly::BI__builtin_wasm_widen_high_u_i16x8_i8x16:
+  case WebAssembly::BI__builtin_wasm_widen_low_s_i32x4_i16x8:
+  case WebAssembly::BI__builtin_wasm_widen_high_s_i32x4_i16x8:
+  case WebAssembly::BI__builtin_wasm_widen_low_u_i32x4_i16x8:
+  case WebAssembly::BI__builtin_wasm_widen_high_u_i32x4_i16x8: {
+    Value *Vec = EmitScalarExpr(E->getArg(0));
+    unsigned IntNo;
+    switch (BuiltinID) {
+    case WebAssembly::BI__builtin_wasm_widen_low_s_i16x8_i8x16:
+    case WebAssembly::BI__builtin_wasm_widen_low_s_i32x4_i16x8:
+      IntNo = Intrinsic::wasm_widen_low_signed;
+      break;
+    case WebAssembly::BI__builtin_wasm_widen_high_s_i16x8_i8x16:
+    case WebAssembly::BI__builtin_wasm_widen_high_s_i32x4_i16x8:
+      IntNo = Intrinsic::wasm_widen_high_signed;
+      break;
+    case WebAssembly::BI__builtin_wasm_widen_low_u_i16x8_i8x16:
+    case WebAssembly::BI__builtin_wasm_widen_low_u_i32x4_i16x8:
+      IntNo = Intrinsic::wasm_widen_low_unsigned;
+      break;
+    case WebAssembly::BI__builtin_wasm_widen_high_u_i16x8_i8x16:
+    case WebAssembly::BI__builtin_wasm_widen_high_u_i32x4_i16x8:
+      IntNo = Intrinsic::wasm_widen_high_unsigned;
+      break;
+    default:
+      llvm_unreachable("unexpected builtin ID");
+    }
+    Function *Callee =
+        CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Vec->getType()});
+    return Builder.CreateCall(Callee, Vec);
+  }
   default:
     return nullptr;
   }
diff --git a/test/CodeGen/builtins-wasm.c b/test/CodeGen/builtins-wasm.c
index ee08983d9a..ca407ebc0e 100644
--- a/test/CodeGen/builtins-wasm.c
+++ b/test/CodeGen/builtins-wasm.c
@@ -463,3 +463,79 @@ i64x2 trunc_saturate_u_i64x2_f64x2(f64x2 f) {
   // WEBASSEMBLY: call <2 x i64> @llvm.wasm.trunc.saturate.unsigned.v2i64.v2f64(<2 x double> %f)
   // WEBASSEMBLY-NEXT: ret
 }
+
+i8x16 narrow_s_i8x16_i16x8(i16x8 low, i16x8 high) {
+  return __builtin_wasm_narrow_s_i8x16_i16x8(low, high);
+  // WEBASSEMBLY: call <16 x i8> @llvm.wasm.narrow.signed.v16i8.v8i16(
+  // WEBASSEMBLY-SAME: <8 x i16> %low, <8 x i16> %high)
+  // WEBASSEMBLY: ret
+}
+
+i8x16 narrow_u_i8x16_i16x8(i16x8 low, i16x8 high) {
+  return __builtin_wasm_narrow_u_i8x16_i16x8(low, high);
+  // WEBASSEMBLY: call <16 x i8> @llvm.wasm.narrow.unsigned.v16i8.v8i16(
+  // WEBASSEMBLY-SAME: <8 x i16> %low, <8 x i16> %high)
+  // WEBASSEMBLY: ret
+}
+
+i16x8 narrow_s_i16x8_i32x4(i32x4 low, i32x4 high) {
+  return __builtin_wasm_narrow_s_i16x8_i32x4(low, high);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.narrow.signed.v8i16.v4i32(
+  // WEBASSEMBLY-SAME: <4 x i32> %low, <4 x i32> %high)
+  // WEBASSEMBLY: ret
+}
+
+i16x8 narrow_u_i16x8_i32x4(i32x4 low, i32x4 high) {
+  return __builtin_wasm_narrow_u_i16x8_i32x4(low, high);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.narrow.unsigned.v8i16.v4i32(
+  // WEBASSEMBLY-SAME: <4 x i32> %low, <4 x i32> %high)
+  // WEBASSEMBLY: ret
+}
+
+i16x8 widen_low_s_i16x8_i8x16(i8x16 v) {
+  return __builtin_wasm_widen_low_s_i16x8_i8x16(v);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.widen.low.signed.v8i16.v16i8(<16 x i8> %v)
+  // WEBASSEMBLY: ret
+}
+
+i16x8 widen_high_s_i16x8_i8x16(i8x16 v) {
+  return __builtin_wasm_widen_high_s_i16x8_i8x16(v);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.widen.high.signed.v8i16.v16i8(<16 x i8> %v)
+  // WEBASSEMBLY: ret
+}
+
+i16x8 widen_low_u_i16x8_i8x16(i8x16 v) {
+  return __builtin_wasm_widen_low_u_i16x8_i8x16(v);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.widen.low.unsigned.v8i16.v16i8(<16 x i8> %v)
+  // WEBASSEMBLY: ret
+}
+
+i16x8 widen_high_u_i16x8_i8x16(i8x16 v) {
+  return __builtin_wasm_widen_high_u_i16x8_i8x16(v);
+  // WEBASSEMBLY: call <8 x i16> @llvm.wasm.widen.high.unsigned.v8i16.v16i8(<16 x i8> %v)
+  // WEBASSEMBLY: ret
+}
+
+i32x4 widen_low_s_i32x4_i16x8(i16x8 v) {
+  return __builtin_wasm_widen_low_s_i32x4_i16x8(v);
+  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.widen.low.signed.v4i32.v8i16(<8 x i16> %v)
+  // WEBASSEMBLY: ret
+}
+
+i32x4 widen_high_s_i32x4_i16x8(i16x8 v) {
+  return __builtin_wasm_widen_high_s_i32x4_i16x8(v);
+  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.widen.high.signed.v4i32.v8i16(<8 x i16> %v)
+  // WEBASSEMBLY: ret
+}
+
+i32x4 widen_low_u_i32x4_i16x8(i16x8 v) {
+  return __builtin_wasm_widen_low_u_i32x4_i16x8(v);
+  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.widen.low.unsigned.v4i32.v8i16(<8 x i16> %v)
+  // WEBASSEMBLY: ret
+}
+
+i32x4 widen_high_u_i32x4_i16x8(i16x8 v) {
+  return __builtin_wasm_widen_high_u_i32x4_i16x8(v);
+  // WEBASSEMBLY: call <4 x i32> @llvm.wasm.widen.high.unsigned.v4i32.v8i16(<8 x i16> %v)
+  // WEBASSEMBLY: ret
+}