Argyrios Kyrtzidis
2010-Aug-31 18:18 UTC
[LLVMdev] "equivalent" .ll files diverge after optimizations are applied
Hi, I've attached 2 .ll files which are supposed to be equivalent but 'unopt-fail.ll' causes a crash in webkit's test suite while 'unopt-pass.ll' does not. I can't give more details about the crash, when I run the crashing test it in isolation it passes, when I run the full suite it crashes; it boggles the mind. Below I provide the optimized asm that is produced from each file. Could you give a hint on what is the problem ? I also attached 't.cpp' which approximates the source that the .ll files came from. -Argiris -------------- next part -------------- A non-text attachment was scrubbed... Name: unopt-fail.ll Type: application/octet-stream Size: 15313 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100831/9782fea6/attachment.obj> -------------- next part -------------- A non-text attachment was scrubbed... Name: unopt-pass.ll Type: application/octet-stream Size: 17832 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100831/9782fea6/attachment-0001.obj> -------------- next part -------------- A non-text attachment was scrubbed... Name: t.cpp Type: application/octet-stream Size: 447 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100831/9782fea6/attachment-0002.obj> -------------- next part -------------- $ opt -std-compile-opts unopt-pass.ll -o - | llc -o - .section __TEXT,__text,regular,pure_instructions .globl __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE .align 4, 0x90 __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE: ## @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE ## BB#0: subq $24, %rsp movq %rsi, %rdx movl $0, 16(%rsp) movl $0, 20(%rsp) movl $0, 8(%rsp) movl $0, 12(%rsp) movq 8(%rdi), %rsi leaq 16(%rsp), %rcx leaq 8(%rsp), %r8 callq __ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_ movss 8(%rsp), %xmm1 movss 12(%rsp), %xmm0 subss 20(%rsp), %xmm0 subss 16(%rsp), %xmm1 ## kill: XMM1<def> XMM1<kill> XMM1<def> insertps $16, %xmm0, %xmm1 ## xmm1 = xmm1[0],xmm0[0],xmm1[2,3] movq 16(%rsp), %xmm0 addq $24, %rsp ret $ opt -std-compile-opts unopt-fail.ll -o - | llc -o - .section __TEXT,__text,regular,pure_instructions .globl __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE .align 4, 0x90 __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE: ## @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE ## BB#0: subq $24, %rsp movq %rsi, %rdx movl $0, 16(%rsp) movl $0, 20(%rsp) movl $0, 8(%rsp) movl $0, 12(%rsp) movq 8(%rdi), %rsi leaq 16(%rsp), %rcx leaq 8(%rsp), %r8 callq __ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_ movss 8(%rsp), %xmm0 movss 12(%rsp), %xmm1 subss 20(%rsp), %xmm1 subss 16(%rsp), %xmm0 insertps $16, %xmm1, %xmm0 ## xmm0 = xmm0[0],xmm1[0],xmm0[2,3] movd %xmm0, %rax movd %rax, %mm0 movq2dq %mm0, %xmm1 movq2dq %mm0, %xmm2 punpcklqdq %xmm2, %xmm1 ## xmm1 = xmm1[0],xmm2[0] movq 16(%rsp), %rax movd %rax, %mm0 movq2dq %mm0, %xmm0 punpcklqdq %xmm2, %xmm0 ## xmm0 = xmm0[0],xmm2[0] addq $24, %rsp ret
Dale Johannesen
2010-Aug-31 18:51 UTC
[LLVMdev] "equivalent" .ll files diverge after optimizations are applied
Using MM registers is wrong unless the user has specifically asked for it, which doesn't seem to be the case here. In the awesome MMX architecture, touching an MM register makes subsequent x87 operations fail unless an EMMS instruction is issued first; none of the compilers here are smart enough to insert EMMS instructions in the right places, so the only safe thing is not to use these registers. There is no x87 instruction shown here, but you've probably got one in the full test suite and not in the test by itself, which fits your data. Why this is happening is not immediately clear. It looks like the successful code is doing an aggregate copy field-by-field while the failing code has lowered this to a memcpy. I would certainly expect the memcpy expansion to be smart enough to avoid using MM registers, though; that's a serious bug if it isn't. movd %xmm0, %rax movd %rax, %mm0 movq2dq %mm0, %xmm1 movq2dq %mm0, %xmm2 punpcklqdq %xmm2, %xmm1 ## xmm1 = xmm1[0],xmm2[0] movq 16(%rsp), %rax movd %rax, %mm0 movq2dq %mm0, %xmm0 punpcklqdq %xmm2, %xmm0 ## xmm0 = xmm0[0],xmm2[0] On Aug 31, 2010, at 11:18 AMPDT, Argyrios Kyrtzidis wrote:> Hi, > > I've attached 2 .ll files which are supposed to be equivalent but > 'unopt-fail.ll' causes a crash in webkit's test suite while 'unopt- > pass.ll' does not. I can't give more details about the crash, when I > run the crashing test it in isolation it passes, when I run the full > suite it crashes; it boggles the mind. > > Below I provide the optimized asm that is produced from each file. > Could you give a hint on what is the problem ? > I also attached 't.cpp' which approximates the source that the .ll > files came from. > > -Argiris > >-------------- next part -------------- A non-text attachment was scrubbed... Name: unopt-fail.ll Type: application/octet-stream Size: 15313 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100831/b4aa9ded/attachment.obj> -------------- next part -------------- A non-text attachment was scrubbed... Name: unopt-pass.ll Type: application/octet-stream Size: 17832 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100831/b4aa9ded/attachment-0001.obj> -------------- next part -------------- A non-text attachment was scrubbed... Name: t.cpp Type: application/octet-stream Size: 447 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100831/b4aa9ded/attachment-0002.obj> -------------- next part --------------> > > > $ opt -std-compile-opts unopt-pass.ll -o - | llc -o - > > .section __TEXT,__text,regular,pure_instructions > .globl > __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE > .align 4, 0x90 > __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE: > ## @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE > ## BB#0: > subq $24, %rsp > movq %rsi, %rdx > movl $0, 16(%rsp) > movl $0, 20(%rsp) > movl $0, 8(%rsp) > movl $0, 12(%rsp) > movq 8(%rdi), %rsi > leaq 16(%rsp), %rcx > leaq 8(%rsp), %r8 > callq > __ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_ > movss 8(%rsp), %xmm1 > movss 12(%rsp), %xmm0 > subss 20(%rsp), %xmm0 > subss 16(%rsp), %xmm1 > ## kill: XMM1<def> XMM1<kill> > XMM1<def> > insertps $16, %xmm0, %xmm1 ## xmm1 = xmm1[0],xmm0[0],xmm1[2,3] > movq 16(%rsp), %xmm0 > addq $24, %rsp > ret > > > $ opt -std-compile-opts unopt-fail.ll -o - | llc -o - > > .section __TEXT,__text,regular,pure_instructions > .globl > __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE > .align 4, 0x90 > __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE: > ## @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE > ## BB#0: > subq $24, %rsp > movq %rsi, %rdx > movl $0, 16(%rsp) > movl $0, 20(%rsp) > movl $0, 8(%rsp) > movl $0, 12(%rsp) > movq 8(%rdi), %rsi > leaq 16(%rsp), %rcx > leaq 8(%rsp), %r8 > callq > __ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_ > movss 8(%rsp), %xmm0 > movss 12(%rsp), %xmm1 > subss 20(%rsp), %xmm1 > subss 16(%rsp), %xmm0 > insertps $16, %xmm1, %xmm0 ## xmm0 = xmm0[0],xmm1[0],xmm0[2,3] > movd %xmm0, %rax > movd %rax, %mm0 > movq2dq %mm0, %xmm1 > movq2dq %mm0, %xmm2 > punpcklqdq %xmm2, %xmm1 ## xmm1 = xmm1[0],xmm2[0] > movq 16(%rsp), %rax > movd %rax, %mm0 > movq2dq %mm0, %xmm0 > punpcklqdq %xmm2, %xmm0 ## xmm0 = xmm0[0],xmm2[0] > addq $24, %rsp > ret > > _______________________________________________ > LLVM Developers mailing list > LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
Argyrios Kyrtzidis
2010-Aug-31 20:21 UTC
[LLVMdev] "equivalent" .ll files diverge after optimizations are applied
Here's the optimized versions: $ opt -std-compile-opts unopt-pass.ll -o - | llvm-dis -o - [...] define %3 @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE(%"class.WebCore::GraphicsContext"* %this, %"struct.WebCore::FloatRect"* %rect) nounwind ssp align 2 { %roundedOrigin = alloca %"class.WebCore::FloatSize", align 4 ; <%"class.WebCore::FloatSize"*> [#uses=3] %roundedLowerRight = alloca %"class.WebCore::FloatSize", align 4 ; <%"class.WebCore::FloatSize"*> [#uses=3] %1 = getelementptr inbounds %"class.WebCore::FloatSize"* %roundedOrigin, i64 0, i32 0 ; <float*> [#uses=2] store float 0.000000e+00, float* %1, align 4 %2 = getelementptr inbounds %"class.WebCore::FloatSize"* %roundedOrigin, i64 0, i32 1 ; <float*> [#uses=2] store float 0.000000e+00, float* %2, align 4 %3 = getelementptr inbounds %"class.WebCore::FloatSize"* %roundedLowerRight, i64 0, i32 0 ; <float*> [#uses=2] store float 0.000000e+00, float* %3, align 4 %4 = getelementptr inbounds %"class.WebCore::FloatSize"* %roundedLowerRight, i64 0, i32 1 ; <float*> [#uses=2] store float 0.000000e+00, float* %4, align 4 %5 = getelementptr inbounds %"class.WebCore::GraphicsContext"* %this, i64 0, i32 1 ; <%"class.WebCore::GraphicsContextPlatformPrivate"**> [#uses=1] %6 = load %"class.WebCore::GraphicsContextPlatformPrivate"** %5, align 8 ; <%"class.WebCore::GraphicsContextPlatformPrivate"*> [#uses=1] call void @_ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_(%"class.WebCore::GraphicsContext"* %this, %"class.WebCore::GraphicsContextPlatformPrivate"* %6, %"struct.WebCore::FloatRect"* %rect, %"class.WebCore::FloatSize"* %roundedOrigin, %"class.WebCore::FloatSize"* %roundedLowerRight) nounwind %7 = load float* %3, align 4 ; <float> [#uses=1] %8 = load float* %1, align 4 ; <float> [#uses=2] %9 = fsub float %7, %8 ; <float> [#uses=1] %10 = load float* %4, align 4 ; <float> [#uses=1] %11 = load float* %2, align 4 ; <float> [#uses=2] %12 = fsub float %10, %11 ; <float> [#uses=1] %13 = insertelement <2 x float> undef, float %8, i32 0 ; <<2 x float>> [#uses=1] %14 = insertelement <2 x float> %13, float %11, i32 1 ; <<2 x float>> [#uses=1] %tmp8 = insertvalue %3 undef, <2 x float> %14, 0 ; <%3> [#uses=1] %15 = insertelement <2 x float> undef, float %9, i32 0 ; <<2 x float>> [#uses=1] %16 = insertelement <2 x float> %15, float %12, i32 1 ; <<2 x float>> [#uses=1] %tmp12 = insertvalue %3 %tmp8, <2 x float> %16, 1 ; <%3> [#uses=1] ret %3 %tmp12 } $ opt -std-compile-opts unopt-fail.ll -o - | llvm-dis -o - [...] define %3 @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE(%"class.WebCore::GraphicsContext"* %this, %"struct.WebCore::FloatRect"* %rect) nounwind ssp align 2 { %roundedOrigin = alloca i64, align 8 ; <i64*> [#uses=3] %tmpcast = bitcast i64* %roundedOrigin to %"class.WebCore::FloatSize"* ; <%"class.WebCore::FloatSize"*> [#uses=2] %roundedLowerRight = alloca %"class.WebCore::FloatSize", align 4 ; <%"class.WebCore::FloatSize"*> [#uses=3] %1 = bitcast i64* %roundedOrigin to float* ; <float*> [#uses=2] store float 0.000000e+00, float* %1, align 8 %2 = getelementptr inbounds %"class.WebCore::FloatSize"* %tmpcast, i64 0, i32 1 ; <float*> [#uses=2] store float 0.000000e+00, float* %2, align 4 %3 = getelementptr inbounds %"class.WebCore::FloatSize"* %roundedLowerRight, i64 0, i32 0 ; <float*> [#uses=2] store float 0.000000e+00, float* %3, align 4 %4 = getelementptr inbounds %"class.WebCore::FloatSize"* %roundedLowerRight, i64 0, i32 1 ; <float*> [#uses=2] store float 0.000000e+00, float* %4, align 4 %5 = getelementptr inbounds %"class.WebCore::GraphicsContext"* %this, i64 0, i32 1 ; <%"class.WebCore::GraphicsContextPlatformPrivate"**> [#uses=1] %6 = load %"class.WebCore::GraphicsContextPlatformPrivate"** %5, align 8 ; <%"class.WebCore::GraphicsContextPlatformPrivate"*> [#uses=1] call void @_ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_(%"class.WebCore::GraphicsContext"* %this, %"class.WebCore::GraphicsContextPlatformPrivate"* %6, %"struct.WebCore::FloatRect"* %rect, %"class.WebCore::FloatSize"* %tmpcast, %"class.WebCore::FloatSize"* %roundedLowerRight) nounwind %7 = load float* %3, align 4 ; <float> [#uses=1] %8 = load float* %1, align 8 ; <float> [#uses=1] %9 = fsub float %7, %8 ; <float> [#uses=1] %10 = load float* %4, align 4 ; <float> [#uses=1] %11 = load float* %2, align 4 ; <float> [#uses=1] %12 = fsub float %10, %11 ; <float> [#uses=1] %tmp3.i = insertelement <2 x float> undef, float %9, i32 0 ; <<2 x float>> [#uses=1] %tmp1.i = insertelement <2 x float> %tmp3.i, float %12, i32 1 ; <<2 x float>> [#uses=1] %tmp = bitcast <2 x float> %tmp1.i to i64 ; <i64> [#uses=1] %tmp.i.i = load i64* %roundedOrigin, align 8 ; <i64> [#uses=1] %tmp9 = insertelement <1 x i64> undef, i64 %tmp.i.i, i32 0 ; <<1 x i64>> [#uses=1] %tmp6 = insertelement <1 x i64> undef, i64 %tmp, i32 0 ; <<1 x i64>> [#uses=1] %tmp11 = bitcast <1 x i64> %tmp9 to <2 x float> ; <<2 x float>> [#uses=1] %insert = insertvalue %3 undef, <2 x float> %tmp11, 0 ; <%3> [#uses=1] %tmp8 = bitcast <1 x i64> %tmp6 to <2 x float> ; <<2 x float>> [#uses=1] %insert4 = insertvalue %3 %insert, <2 x float> %tmp8, 1 ; <%3> [#uses=1] ret %3 %insert4 } Just to be clear, are you saying that the fact that, after using llc on the second IR, the produced asm is using MM registers, indicates a bug ? -Argiris On Aug 31, 2010, at 7:51 PM, Dale Johannesen wrote:> Using MM registers is wrong unless the user has specifically asked for it, which doesn't seem to be the case here. > In the awesome MMX architecture, touching an MM register makes subsequent x87 operations fail unless an EMMS instruction is issued first; none of the compilers here are smart enough to insert EMMS instructions in the right places, so the only safe thing is not to use these registers. There is no x87 instruction shown here, but you've probably got one in the full test suite and not in the test by itself, which fits your data. > > Why this is happening is not immediately clear. It looks like the successful code is doing an aggregate copy field-by-field while the failing code has lowered this to a memcpy. I would certainly expect the memcpy expansion to be smart enough to avoid using MM registers, though; that's a serious bug if it isn't. > > movd %xmm0, %rax > movd %rax, %mm0 > movq2dq %mm0, %xmm1 > movq2dq %mm0, %xmm2 > punpcklqdq %xmm2, %xmm1 ## xmm1 = xmm1[0],xmm2[0] > movq 16(%rsp), %rax > movd %rax, %mm0 > movq2dq %mm0, %xmm0 > punpcklqdq %xmm2, %xmm0 ## xmm0 = xmm0[0],xmm2[0] > > > On Aug 31, 2010, at 11:18 AMPDT, Argyrios Kyrtzidis wrote: > >> Hi, >> >> I've attached 2 .ll files which are supposed to be equivalent but 'unopt-fail.ll' causes a crash in webkit's test suite while 'unopt-pass.ll' does not. I can't give more details about the crash, when I run the crashing test it in isolation it passes, when I run the full suite it crashes; it boggles the mind. >> >> Below I provide the optimized asm that is produced from each file. Could you give a hint on what is the problem ? >> I also attached 't.cpp' which approximates the source that the .ll files came from. >> >> -Argiris >> >> > <unopt-fail.ll><unopt-pass.ll><t.cpp> >> >> >> >> >> $ opt -std-compile-opts unopt-pass.ll -o - | llc -o - >> >> .section __TEXT,__text,regular,pure_instructions >> .globl __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE >> .align 4, 0x90 >> __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE: ## @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE >> ## BB#0: >> subq $24, %rsp >> movq %rsi, %rdx >> movl $0, 16(%rsp) >> movl $0, 20(%rsp) >> movl $0, 8(%rsp) >> movl $0, 12(%rsp) >> movq 8(%rdi), %rsi >> leaq 16(%rsp), %rcx >> leaq 8(%rsp), %r8 >> callq __ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_ >> movss 8(%rsp), %xmm1 >> movss 12(%rsp), %xmm0 >> subss 20(%rsp), %xmm0 >> subss 16(%rsp), %xmm1 >> ## kill: XMM1<def> XMM1<kill> XMM1<def> >> insertps $16, %xmm0, %xmm1 ## xmm1 = xmm1[0],xmm0[0],xmm1[2,3] >> movq 16(%rsp), %xmm0 >> addq $24, %rsp >> ret >> >> >> $ opt -std-compile-opts unopt-fail.ll -o - | llc -o - >> >> .section __TEXT,__text,regular,pure_instructions >> .globl __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE >> .align 4, 0x90 >> __ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE: ## @_ZN7WebCore15GraphicsContext19roundToDevicePixelsERKNS_9FloatRectE >> ## BB#0: >> subq $24, %rsp >> movq %rsi, %rdx >> movl $0, 16(%rsp) >> movl $0, 20(%rsp) >> movl $0, 8(%rsp) >> movl $0, 12(%rsp) >> movq 8(%rdi), %rsi >> leaq 16(%rsp), %rcx >> leaq 8(%rsp), %r8 >> callq __ZN7WebCore5mouniEPNS_15GraphicsContextEPNS_30GraphicsContextPlatformPrivateERKNS_9FloatRectERNS_10FloatPointES8_ >> movss 8(%rsp), %xmm0 >> movss 12(%rsp), %xmm1 >> subss 20(%rsp), %xmm1 >> subss 16(%rsp), %xmm0 >> insertps $16, %xmm1, %xmm0 ## xmm0 = xmm0[0],xmm1[0],xmm0[2,3] >> movd %xmm0, %rax >> movd %rax, %mm0 >> movq2dq %mm0, %xmm1 >> movq2dq %mm0, %xmm2 >> punpcklqdq %xmm2, %xmm1 ## xmm1 = xmm1[0],xmm2[0] >> movq 16(%rsp), %rax >> movd %rax, %mm0 >> movq2dq %mm0, %xmm0 >> punpcklqdq %xmm2, %xmm0 ## xmm0 = xmm0[0],xmm2[0] >> addq $24, %rsp >> ret >> >> _______________________________________________ >> LLVM Developers mailing list >> LLVMdev at cs.uiuc.edu http://llvm.cs.uiuc.edu >> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev >
Chris Lattner
2010-Sep-01 05:19 UTC
[LLVMdev] "equivalent" .ll files diverge after optimizations are applied
On Aug 31, 2010, at 11:18 AM, Argyrios Kyrtzidis wrote:> Hi, > > I've attached 2 .ll files which are supposed to be equivalent but 'unopt-fail.ll' causes a crash in webkit's test suite while 'unopt-pass.ll' does not. I can't give more details about the crash, when I run the crashing test it in isolation it passes, when I run the full suite it crashes; it boggles the mind. > > Below I provide the optimized asm that is produced from each file. Could you give a hint on what is the problem ? > I also attached 't.cpp' which approximates the source that the .ll files came from.Hi Argiris, The real problem here is that the X86 backend is turning datatypes like <1 x i64> into MMX operations, but doesn't do so in a safe way (it's not inserting the requisite EMMS instructions). After discussing this with Dale and Bill, the right fix is to stop mapping generic vectors onto MMX operations. This will define away the existing -disable-mmx flag and make stuff like this impossible. However, this isn't going to happen in the next couple days, certainly not in time for the 2.8 release branch on friday. As such, I checked in a horrible hack in r112696 that prevents SRoA from introducing mmx specific vector types. I'm not aware of a target where those datatypes are actually useful, so this shouldn't be bad. On your testcase, no mmx operations are produced. Please let me know if you see any other MMX stuff being generated. -Chris
Argyrios Kyrtzidis
2010-Sep-01 09:25 UTC
[LLVMdev] "equivalent" .ll files diverge after optimizations are applied
Thanks for the quick response! On 1 Sep 2010, at 06:19, Chris Lattner wrote:> > On Aug 31, 2010, at 11:18 AM, Argyrios Kyrtzidis wrote: > >> Hi, >> >> I've attached 2 .ll files which are supposed to be equivalent but 'unopt-fail.ll' causes a crash in webkit's test suite while 'unopt-pass.ll' does not. I can't give more details about the crash, when I run the crashing test it in isolation it passes, when I run the full suite it crashes; it boggles the mind. >> >> Below I provide the optimized asm that is produced from each file. Could you give a hint on what is the problem ? >> I also attached 't.cpp' which approximates the source that the .ll files came from. > > Hi Argiris, > > The real problem here is that the X86 backend is turning datatypes like <1 x i64> into MMX operations, but doesn't do so in a safe way (it's not inserting the requisite EMMS instructions). After discussing this with Dale and Bill, the right fix is to stop mapping generic vectors onto MMX operations. This will define away the existing -disable-mmx flag and make stuff like this impossible. > > However, this isn't going to happen in the next couple days, certainly not in time for the 2.8 release branch on friday. As such, I checked in a horrible hack in r112696 that prevents SRoA from introducing mmx specific vector types. I'm not aware of a target where those datatypes are actually useful, so this shouldn't be bad. On your testcase, no mmx operations are produced. > > Please let me know if you see any other MMX stuff being generated. > > -Chris
Bob Wilson
2010-Sep-01 16:43 UTC
[LLVMdev] "equivalent" .ll files diverge after optimizations are applied
On Aug 31, 2010, at 10:19 PM, Chris Lattner wrote:> > Hi Argiris, > > The real problem here is that the X86 backend is turning datatypes like <1 x i64> into MMX operations, but doesn't do so in a safe way (it's not inserting the requisite EMMS instructions). After discussing this with Dale and Bill, the right fix is to stop mapping generic vectors onto MMX operations. This will define away the existing -disable-mmx flag and make stuff like this impossible. > > However, this isn't going to happen in the next couple days, certainly not in time for the 2.8 release branch on friday. As such, I checked in a horrible hack in r112696 that prevents SRoA from introducing mmx specific vector types. I'm not aware of a target where those datatypes are actually useful, so this shouldn't be bad. On your testcase, no mmx operations are produced.Those types are all used with NEON. I'll try to do some experiments to measure the impact of this....
Seemingly Similar Threads
- [LLVMdev] "equivalent" .ll files diverge after optimizations are applied
- [LLVMdev] "equivalent" .ll files diverge after optimizations are applied
- [LLVMdev] "equivalent" .ll files diverge after optimizations are applied
- [LLVMdev] Generating movq2dq using IRBuilder
- [cfe-dev] FE_INEXACT being set for an exact conversion from float to unsigned long long