Fix alignment on Win32 and Solaris Sparc (PR #709)

These fixes were interesting in a morbid sort of way. I thought the FixedSizeAllocatorWithCleanup specializations faithfully reproduced semantics but I was wrong on Win32 and Sparc. Also see Commit e054d36dc88d.

It seems there was another requirement or dependency that we missed, but it was not readily apparent. If I am parsing results correctly (which I may not be), it appears the bit twiddling using 8 byte alignment had more influence on alignment than I originally thought based on use of CRYPTOPP_BOOL_ALIGN16 and T_Align16. Or maybe the alignment attributes specified by CRYPTOPP_ALIGN_DATA are not being honored like they should for stack allocations.

This check-in avoids some uses of x86 movdqa (aligned) in favor of movdqu (unaligned). The uses were concentrated on memory operands which were 8-byte aligned instead of 16-byte aligned. It is not clear to me how the specializations lost 8-bytes of alignment. The check-in also enlists CRYPTOPP_ASSERT to tell us when there's a problem so we don't need to go hunting for bugs.

1 files changed, 29 insertions, 27 deletions

diff --git a/panama.cpp b/panama.cpp
index 19b5fcca..351165ab 100644
--- a/panama.cpp
+++ b/panama.cpp
@@ -93,10 +93,10 @@ void CRYPTOPP_NOINLINE Panama_SSE2_Pull(size_t count, word32 *state, word32 *z,
 		AS_PUSH_IF86(	cx)

 	#endif

 

-	AS2(	movdqa	xmm0, XMMWORD_PTR [AS_REG_2+0*16])

-	AS2(	movdqa	xmm1, XMMWORD_PTR [AS_REG_2+1*16])

-	AS2(	movdqa	xmm2, XMMWORD_PTR [AS_REG_2+2*16])

-	AS2(	movdqa	xmm3, XMMWORD_PTR [AS_REG_2+3*16])

+	AS2(	movdqu	xmm0, XMMWORD_PTR [AS_REG_2+0*16])

+	AS2(	movdqu	xmm1, XMMWORD_PTR [AS_REG_2+1*16])

+	AS2(	movdqu	xmm2, XMMWORD_PTR [AS_REG_2+2*16])

+	AS2(	movdqu	xmm3, XMMWORD_PTR [AS_REG_2+3*16])

 	AS2(	mov		eax, dword ptr [AS_REG_2+4*16])

 

 	ASL(4)

@@ -184,8 +184,8 @@ void CRYPTOPP_NOINLINE Panama_SSE2_Pull(size_t count, word32 *state, word32 *z,
 	ASL(1)

 	AS2(	test	AS_REG_3, 15)

 	ASJ(	jnz,	3, f)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_3], xmm4)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_3+16], xmm6)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_3], xmm4)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_3+16], xmm6)

 	AS2(	add		AS_REG_3, 32)

 	ASJ(	jmp,	0, f)

 	ASL(3)

@@ -200,24 +200,26 @@ void CRYPTOPP_NOINLINE Panama_SSE2_Pull(size_t count, word32 *state, word32 *z,
 	AS2(	lea		AS_REG_7, [AS_REG_6 + (32-24)*32])

 	AS2(	and		AS_REG_7, 31*32)

 

-	AS2(	movdqa	xmm0, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+0*8])

+	AS2(	movdqu	xmm0, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+0*8])

 	AS2(	pxor	xmm3, xmm0)

 	ASS(	pshufd	xmm0, xmm0, 2, 3, 0, 1)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+0*8], xmm3)

-	AS2(	pxor	xmm0, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+2*8])

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+2*8], xmm0)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+0*8], xmm3)

+	AS2(	movdqu	xmm5, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+2*8])

+	AS2(	pxor	xmm0, xmm5)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+2*8], xmm0)

 

-	AS2(	movdqa	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+2*8])

+	AS2(	movdqu	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+2*8])

 	AS2(	pxor	xmm1, xmm4)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+2*8], xmm1)

-	AS2(	pxor	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+0*8])

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+0*8], xmm4)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+2*8], xmm1)

+	AS2(	movdqu	xmm5, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+0*8])

+	AS2(	pxor	xmm4, xmm5)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+0*8], xmm4)

 

 	// theta

-	AS2(	movdqa	xmm3, XMMWORD_PTR [AS_REG_2+3*16])

-	AS2(	movdqa	xmm2, XMMWORD_PTR [AS_REG_2+2*16])

-	AS2(	movdqa	xmm1, XMMWORD_PTR [AS_REG_2+1*16])

-	AS2(	movdqa	xmm0, XMMWORD_PTR [AS_REG_2+0*16])

+	AS2(	movdqu	xmm3, XMMWORD_PTR [AS_REG_2+3*16])

+	AS2(	movdqu	xmm2, XMMWORD_PTR [AS_REG_2+2*16])

+	AS2(	movdqu	xmm1, XMMWORD_PTR [AS_REG_2+1*16])

+	AS2(	movdqu	xmm0, XMMWORD_PTR [AS_REG_2+0*16])

 

 #if CRYPTOPP_SSSE3_ASM_AVAILABLE

 	AS2(	test	AS_REG_6, 1)

@@ -271,16 +273,16 @@ void CRYPTOPP_NOINLINE Panama_SSE2_Pull(size_t count, word32 *state, word32 *z,
 	AS2(	lea		AS_REG_7, [AS_REG_6 + 16*32])

 	AS2(	and		AS_REG_7, 31*32)

 

-	AS2(	movdqa	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+0*16])

-	AS2(	movdqa	xmm5, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+0*16])

+	AS2(	movdqu	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+0*16])

+	AS2(	movdqu	xmm5, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+0*16])

 	AS2(	movdqa	xmm6, xmm4)

 	AS2(	punpcklqdq	xmm4, xmm5)

 	AS2(	punpckhqdq	xmm6, xmm5)

 	AS2(	pxor	xmm3, xmm4)

 	AS2(	pxor	xmm2, xmm6)

 

-	AS2(	movdqa	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+1*16])

-	AS2(	movdqa	xmm5, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+1*16])

+	AS2(	movdqu	xmm4, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_1+1*16])

+	AS2(	movdqu	xmm5, XMMWORD_PTR [AS_REG_2+20*4+AS_REG_7+1*16])

 	AS2(	movdqa	xmm6, xmm4)

 	AS2(	punpcklqdq	xmm4, xmm5)

 	AS2(	punpckhqdq	xmm6, xmm5)

@@ -294,10 +296,10 @@ void CRYPTOPP_NOINLINE Panama_SSE2_Pull(size_t count, word32 *state, word32 *z,
 

 	// save state

 	AS2(	mov		[AS_REG_2+4*16], eax)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+3*16], xmm3)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+2*16], xmm2)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+1*16], xmm1)

-	AS2(	movdqa	XMMWORD_PTR [AS_REG_2+0*16], xmm0)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+3*16], xmm3)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+2*16], xmm2)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+1*16], xmm1)

+	AS2(	movdqu	XMMWORD_PTR [AS_REG_2+0*16], xmm0)

 

 	#if CRYPTOPP_BOOL_X86

 		AS2(	add		esp, 4)

@@ -329,7 +331,7 @@ void CRYPTOPP_NOINLINE Panama_SSE2_Pull(size_t count, word32 *state, word32 *z,
 #else

 }

 #endif

-#endif	// #ifdef CRYPTOPP_SSE2_ASM_AVAILABLE

+#endif	// CRYPTOPP_SSE2_ASM_AVAILABLE

 

 #ifndef CRYPTOPP_GENERATE_X64_MASM