[x86] Allow folding unaligned memory operands into pcmp[ei]str*

instructions.

These have special permission according to the x86 manual to read
unaligned memory, and this folding is done by ICC and GCC as well.

This corrects one of the issues identified in PR37246.

llvm-svn: 330896

diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index 11ca8b0..4301fb3 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -631,10 +631,10 @@ X86InstrInfo::X86InstrInfo(X86Subtarget &STI)
     { X86::PABSBrr,         X86::PABSBrm,             TB_ALIGN_16 },
     { X86::PABSDrr,         X86::PABSDrm,             TB_ALIGN_16 },
     { X86::PABSWrr,         X86::PABSWrm,             TB_ALIGN_16 },
-    { X86::PCMPESTRIrr,     X86::PCMPESTRIrm,         TB_ALIGN_16 },
-    { X86::PCMPESTRM128rr,  X86::PCMPESTRM128rm,      TB_ALIGN_16 },
-    { X86::PCMPISTRIrr,     X86::PCMPISTRIrm,         TB_ALIGN_16 },
-    { X86::PCMPISTRM128rr,  X86::PCMPISTRM128rm,      TB_ALIGN_16 },
+    { X86::PCMPESTRIrr,     X86::PCMPESTRIrm,         0 },
+    { X86::PCMPESTRM128rr,  X86::PCMPESTRM128rm,      0 },
+    { X86::PCMPISTRIrr,     X86::PCMPISTRIrm,         0 },
+    { X86::PCMPISTRM128rr,  X86::PCMPISTRM128rm,      0 },
     { X86::PHMINPOSUWrr,    X86::PHMINPOSUWrm,        TB_ALIGN_16 },
     { X86::PMOVSXBDrr,      X86::PMOVSXBDrm,          TB_NO_REVERSE },
     { X86::PMOVSXBQrr,      X86::PMOVSXBQrm,          TB_NO_REVERSE },

diff --git a/llvm/test/CodeGen/X86/sse42-intrinsics-x86.ll b/llvm/test/CodeGen/X86/sse42-intrinsics-x86.ll
index 400a78f..56a7f9e 100644 (file)
--- a/llvm/test/CodeGen/X86/sse42-intrinsics-x86.ll
+++ b/llvm/test/CodeGen/X86/sse42-intrinsics-x86.ll
@@ -437,7 +437,7 @@ define <16 x i8> @test_x86_sse42_pcmpistrm128_load(<16 x i8> %a0, <16 x i8>* %a1
 ; VCHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax ## encoding: [0x8b,0x44,0x24,0x04]
 ; VCHECK-NEXT:    vpcmpistrm $7, (%eax), %xmm0 ## encoding: [0xc4,0xe3,0x79,0x62,0x00,0x07]
 ; VCHECK-NEXT:    retl ## encoding: [0xc3]
-  %1 = load <16 x i8>, <16 x i8>* %a1
+  %1 = load <16 x i8>, <16 x i8>* %a1, align 1
   %res = call <16 x i8> @llvm.x86.sse42.pcmpistrm128(<16 x i8> %a0, <16 x i8> %1, i8 7) ; <<16 x i8>> [#uses=1]
   ret <16 x i8> %res
 }

diff --git a/llvm/test/CodeGen/X86/sse42.ll b/llvm/test/CodeGen/X86/sse42.ll
index 1e197dc..c53511f 100644 (file)
--- a/llvm/test/CodeGen/X86/sse42.ll
+++ b/llvm/test/CodeGen/X86/sse42.ll
@@ -123,19 +123,18 @@ define i1 @pcmpestri_mem_eq_i8(i8* %lhs_ptr, i32 %lhs_len, i8* %rhs_ptr, i32 %rh
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movdqu (%esi), %xmm0
-; X32-NEXT:    movdqu (%ecx), %xmm1
-; X32-NEXT:    pcmpestri $24, %xmm1, %xmm0
+; X32-NEXT:    pcmpestri $24, (%ecx), %xmm0
 ; X32-NEXT:    setae %al
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpestri_mem_eq_i8:
 ; X64:       # %bb.0: # %entry
+; X64-NEXT:    movq %rdx, %r8
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rdx), %xmm1
 ; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl %ecx, %edx
-; X64-NEXT:    pcmpestri $24, %xmm1, %xmm0
+; X64-NEXT:    pcmpestri $24, (%r8), %xmm0
 ; X64-NEXT:    setae %al
 ; X64-NEXT:    retq
 entry:
@@ -157,19 +156,18 @@ define i32 @pcmpestri_mem_idx_i8(i8* %lhs_ptr, i32 %lhs_len, i8* %rhs_ptr, i32 %
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movdqu (%esi), %xmm0
-; X32-NEXT:    movdqu (%ecx), %xmm1
-; X32-NEXT:    pcmpestri $24, %xmm1, %xmm0
+; X32-NEXT:    pcmpestri $24, (%ecx), %xmm0
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpestri_mem_idx_i8:
 ; X64:       # %bb.0: # %entry
+; X64-NEXT:    movq %rdx, %r8
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rdx), %xmm1
 ; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl %ecx, %edx
-; X64-NEXT:    pcmpestri $24, %xmm1, %xmm0
+; X64-NEXT:    pcmpestri $24, (%r8), %xmm0
 ; X64-NEXT:    movl %ecx, %eax
 ; X64-NEXT:    retq
 entry:
@@ -380,19 +378,18 @@ define i1 @pcmpestri_mem_eq_i16(i16* %lhs_ptr, i32 %lhs_len, i16* %rhs_ptr, i32
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movdqu (%esi), %xmm0
-; X32-NEXT:    movdqu (%ecx), %xmm1
-; X32-NEXT:    pcmpestri $25, %xmm1, %xmm0
+; X32-NEXT:    pcmpestri $25, (%ecx), %xmm0
 ; X32-NEXT:    setae %al
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpestri_mem_eq_i16:
 ; X64:       # %bb.0: # %entry
+; X64-NEXT:    movq %rdx, %r8
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rdx), %xmm1
 ; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl %ecx, %edx
-; X64-NEXT:    pcmpestri $25, %xmm1, %xmm0
+; X64-NEXT:    pcmpestri $25, (%r8), %xmm0
 ; X64-NEXT:    setae %al
 ; X64-NEXT:    retq
 entry:
@@ -416,19 +413,18 @@ define i32 @pcmpestri_mem_idx_i16(i16* %lhs_ptr, i32 %lhs_len, i16* %rhs_ptr, i3
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movdqu (%esi), %xmm0
-; X32-NEXT:    movdqu (%ecx), %xmm1
-; X32-NEXT:    pcmpestri $25, %xmm1, %xmm0
+; X32-NEXT:    pcmpestri $25, (%ecx), %xmm0
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpestri_mem_idx_i16:
 ; X64:       # %bb.0: # %entry
+; X64-NEXT:    movq %rdx, %r8
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rdx), %xmm1
 ; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movl %ecx, %edx
-; X64-NEXT:    pcmpestri $25, %xmm1, %xmm0
+; X64-NEXT:    pcmpestri $25, (%r8), %xmm0
 ; X64-NEXT:    movl %ecx, %eax
 ; X64-NEXT:    retq
 entry:
@@ -622,16 +618,14 @@ define i1 @pcmpistri_mem_eq_i8(i8* %lhs_ptr, i8* %rhs_ptr) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movdqu (%ecx), %xmm0
-; X32-NEXT:    movdqu (%eax), %xmm1
-; X32-NEXT:    pcmpistri $24, %xmm1, %xmm0
+; X32-NEXT:    pcmpistri $24, (%eax), %xmm0
 ; X32-NEXT:    setae %al
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpistri_mem_eq_i8:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rsi), %xmm1
-; X64-NEXT:    pcmpistri $24, %xmm1, %xmm0
+; X64-NEXT:    pcmpistri $24, (%rsi), %xmm0
 ; X64-NEXT:    setae %al
 ; X64-NEXT:    retq
 entry:
@@ -650,16 +644,14 @@ define i32 @pcmpistri_mem_idx_i8(i8* %lhs_ptr, i8* %rhs_ptr) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movdqu (%ecx), %xmm0
-; X32-NEXT:    movdqu (%eax), %xmm1
-; X32-NEXT:    pcmpistri $24, %xmm1, %xmm0
+; X32-NEXT:    pcmpistri $24, (%eax), %xmm0
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpistri_mem_idx_i8:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rsi), %xmm1
-; X64-NEXT:    pcmpistri $24, %xmm1, %xmm0
+; X64-NEXT:    pcmpistri $24, (%rsi), %xmm0
 ; X64-NEXT:    movl %ecx, %eax
 ; X64-NEXT:    retq
 entry:
@@ -849,16 +841,14 @@ define i1 @pcmpistri_mem_eq_i16(i16* %lhs_ptr, i16* %rhs_ptr) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movdqu (%ecx), %xmm0
-; X32-NEXT:    movdqu (%eax), %xmm1
-; X32-NEXT:    pcmpistri $25, %xmm1, %xmm0
+; X32-NEXT:    pcmpistri $25, (%eax), %xmm0
 ; X32-NEXT:    setae %al
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpistri_mem_eq_i16:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rsi), %xmm1
-; X64-NEXT:    pcmpistri $25, %xmm1, %xmm0
+; X64-NEXT:    pcmpistri $25, (%rsi), %xmm0
 ; X64-NEXT:    setae %al
 ; X64-NEXT:    retq
 entry:
@@ -879,16 +869,14 @@ define i32 @pcmpistri_mem_idx_i16(i16* %lhs_ptr, i16* %rhs_ptr) nounwind {
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movdqu (%ecx), %xmm0
-; X32-NEXT:    movdqu (%eax), %xmm1
-; X32-NEXT:    pcmpistri $25, %xmm1, %xmm0
+; X32-NEXT:    pcmpistri $25, (%eax), %xmm0
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: pcmpistri_mem_idx_i16:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movdqu (%rdi), %xmm0
-; X64-NEXT:    movdqu (%rsi), %xmm1
-; X64-NEXT:    pcmpistri $25, %xmm1, %xmm0
+; X64-NEXT:    pcmpistri $25, (%rsi), %xmm0
 ; X64-NEXT:    movl %ecx, %eax
 ; X64-NEXT:    retq
 entry: