Frank Winter via llvm-dev
2016-Jun-29 19:41 UTC
[llvm-dev] avx512 JIT backend generates wrong code on <4 x float>
Hi! When compiling the attached module with the JIT engine on an Intel KNL I see wrong code getting emitted. I attach a complete exploit program which shows the bug in LLVM 3.8. It loads and JIT compiles the module and prints the assembler. I stumbled on this since the result of an actual calculation was wrong. So, it's not only the text version of the assembler also the machine assembler is wrong. When I execute the exploit program on an Intel KNL the following output is produced: CPU name = knl -sse4a,-avx512bw,cx16,-tbm,xsave,-fma4,-avx512vl,prfchw,bmi2,adx,-xsavec,fsgsbase,avx,avx512cd,avx512pf,-rtm,popcnt,fma,bmi,aes,rdrnd,-xsaves,sse4.1,sse4.2,avx2,avx512er,sse,lzcnt,pclmul,avx512f,f16c,ssse3,mmx,-pku,cmov,-xop,rdseed,movbe,-hle,xsaveopt,-sha,sse2,sse3,-avx512dq, Assembly: .text .file "module_KFxOBX_i4_after.ll" .globl adjmul .align 16, 0x90 .type adjmul, at function adjmul: .cfi_startproc leaq (%rdi,%r8), %rdx addq %rsi, %r8 testb $1, %cl cmoveq %rdi, %rdx cmoveq %rsi, %r8 movq %rdx, %rax sarq $63, %rax shrq $62, %rax addq %rdx, %rax sarq $2, %rax movq %r8, %rcx sarq $63, %rcx shrq $62, %rcx addq %r8, %rcx sarq $2, %rcx movq %rax, %rdx shlq $5, %rdx leaq 16(%r9,%rdx), %rsi orq $16, %rdx movq 16(%rsp), %rdi addq %rdx, %rdi addq 8(%rsp), %rdx .align 16, 0x90 .LBB0_1: vmovaps -16(%rdx), %xmm0 vmovaps (%rdx), %xmm1 vmovaps -16(%rdi), %xmm2 vmovaps (%rdi), %xmm3 vmulps %xmm3, %xmm1, %xmm4 vmulps %xmm2, %xmm1, %xmm1 vfmadd213ss %xmm4, %xmm0, %xmm2 vfmsub213ss %xmm1, %xmm0, %xmm3 vmovaps %xmm2, -16(%rsi) vmovaps %xmm3, (%rsi) addq $1, %rax addq $32, %rsi addq $32, %rdi addq $32, %rdx cmpq %rcx, %rax jl .LBB0_1 retq .Lfunc_end0: .size adjmul, .Lfunc_end0-adjmul .cfi_endproc .section ".note.GNU-stack","", at progbits end assembly! The instructions 'vfmadd213ss' are 'Fused Multiply-Add of Scalar Single-Precision Floating-Point'. Those should be SIMD vector instructions. Note that the KNL has 16 wide float SIMD, while the exploit module uses only 4. However, the backend should be able to handle this. Unless I receive further ideas I will file an official bug report. Frank -------------- next part -------------- LLVMPATH=/home/fwinter/toolchain/install/llvm-3.8-recent LLVMCONF=$(LLVMPATH)/bin/llvm-config CXXFLAGS=$(shell $(LLVMCONF) --cxxflags) LIBS=$(shell $(LLVMCONF) --libs core mcjit native scalaropts vectorize irreader linker) -ldl LDFLAGS=$(shell $(LLVMCONF) --ldflags) CXX=g++ OBJS=main.o TARGET=main all: $(TARGET) main: $(OBJS) $(CXX) -g -o $@ $^ $(LIBS) $(LDFLAGS) %.o: %.cc $(CXX) $(CXXFLAGS) -c -g $< clean: rm -rf $(TARGET) $(OBJS) *~ -------------- next part -------------- target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" define void @adjmul(i64 %lo, i64 %hi, i64 %myId, i1 %ordered, i64 %start, float* noalias align 64 %arg0, float* noalias align 64 %arg1, float* noalias align 64 %arg2) { entrypoint: %0 = add nsw i64 %lo, %start %1 = add nsw i64 %hi, %start %2 = select i1 %ordered, i64 %0, i64 %lo %3 = select i1 %ordered, i64 %1, i64 %hi %4 = sdiv i64 %2, 4 %5 = sdiv i64 %3, 4 br label %L5 L5: ; preds = %L5, %entrypoint %6 = phi i64 [ %27, %L5 ], [ %4, %entrypoint ] %7 = shl i64 %6, 3 %8 = or i64 %7, 4 %9 = getelementptr float, float* %arg1, i64 %7 %10 = bitcast float* %9 to <4 x float>* %wide.load = load <4 x float>, <4 x float>* %10, align 16 %11 = getelementptr float, float* %arg1, i64 %8 %12 = bitcast float* %11 to <4 x float>* %wide.load5 = load <4 x float>, <4 x float>* %12, align 16 %13 = getelementptr float, float* %arg2, i64 %7 %14 = bitcast float* %13 to <4 x float>* %wide.load6 = load <4 x float>, <4 x float>* %14, align 16 %15 = getelementptr float, float* %arg2, i64 %8 %16 = bitcast float* %15 to <4 x float>* %wide.load7 = load <4 x float>, <4 x float>* %16, align 16 %17 = fmul <4 x float> %wide.load, %wide.load6 %18 = fmul <4 x float> %wide.load5, %wide.load7 %19 = fadd <4 x float> %17, %18 %20 = fmul <4 x float> %wide.load, %wide.load7 %21 = fmul <4 x float> %wide.load5, %wide.load6 %22 = fsub <4 x float> %20, %21 %23 = getelementptr float, float* %arg0, i64 %7 %24 = bitcast float* %23 to <4 x float>* store <4 x float> %19, <4 x float>* %24, align 16 %25 = getelementptr float, float* %arg0, i64 %8 %26 = bitcast float* %25 to <4 x float>* store <4 x float> %22, <4 x float>* %26, align 16 %27 = add nsw i64 %6, 1 %28 = icmp slt i64 %27, %5 br i1 %28, label %L5, label %L6 L6: ; preds = %L5 ret void } -------------- next part -------------- A non-text attachment was scrubbed... Name: main.cc Type: text/x-c++src Size: 5059 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20160629/8008aa24/attachment.cc>
Hal Finkel via llvm-dev
2016-Jun-29 19:48 UTC
[llvm-dev] avx512 JIT backend generates wrong code on <4 x float>
Hi Frank, I recommend trying trunk LLVM. AVX-512 development has been very active recently. -Hal ----- Original Message -----> From: "Frank Winter via llvm-dev" <llvm-dev at lists.llvm.org> > To: "LLVM Dev" <llvm-dev at lists.llvm.org> > Sent: Wednesday, June 29, 2016 2:41:39 PM > Subject: [llvm-dev] avx512 JIT backend generates wrong code on <4 x float> > > Hi! > > When compiling the attached module with the JIT engine on an Intel > KNL I > see wrong code getting emitted. I attach a complete exploit program > which shows the bug in LLVM 3.8. It loads and JIT compiles the module > and prints the assembler. I stumbled on this since the result of an > actual calculation was wrong. So, it's not only the text version of > the > assembler also the machine assembler is wrong. > > When I execute the exploit program on an Intel KNL the following > output > is produced: > > CPU name = knl > -sse4a,-avx512bw,cx16,-tbm,xsave,-fma4,-avx512vl,prfchw,bmi2,adx,-xsavec,fsgsbase,avx,avx512cd,avx512pf,-rtm,popcnt,fma,bmi,aes,rdrnd,-xsaves,sse4.1,sse4.2,avx2,avx512er,sse,lzcnt,pclmul,avx512f,f16c,ssse3,mmx,-pku,cmov,-xop,rdseed,movbe,-hle,xsaveopt,-sha,sse2,sse3,-avx512dq, > Assembly: > .text > .file "module_KFxOBX_i4_after.ll" > .globl adjmul > .align 16, 0x90 > .type adjmul, at function > adjmul: > .cfi_startproc > leaq (%rdi,%r8), %rdx > addq %rsi, %r8 > testb $1, %cl > cmoveq %rdi, %rdx > cmoveq %rsi, %r8 > movq %rdx, %rax > sarq $63, %rax > shrq $62, %rax > addq %rdx, %rax > sarq $2, %rax > movq %r8, %rcx > sarq $63, %rcx > shrq $62, %rcx > addq %r8, %rcx > sarq $2, %rcx > movq %rax, %rdx > shlq $5, %rdx > leaq 16(%r9,%rdx), %rsi > orq $16, %rdx > movq 16(%rsp), %rdi > addq %rdx, %rdi > addq 8(%rsp), %rdx > .align 16, 0x90 > .LBB0_1: > vmovaps -16(%rdx), %xmm0 > vmovaps (%rdx), %xmm1 > vmovaps -16(%rdi), %xmm2 > vmovaps (%rdi), %xmm3 > vmulps %xmm3, %xmm1, %xmm4 > vmulps %xmm2, %xmm1, %xmm1 > vfmadd213ss %xmm4, %xmm0, %xmm2 > vfmsub213ss %xmm1, %xmm0, %xmm3 > vmovaps %xmm2, -16(%rsi) > vmovaps %xmm3, (%rsi) > addq $1, %rax > addq $32, %rsi > addq $32, %rdi > addq $32, %rdx > cmpq %rcx, %rax > jl .LBB0_1 > retq > .Lfunc_end0: > .size adjmul, .Lfunc_end0-adjmul > .cfi_endproc > > > .section ".note.GNU-stack","", at progbits > > end assembly! > > > The instructions 'vfmadd213ss' are 'Fused Multiply-Add of Scalar > Single-Precision Floating-Point'. Those should be SIMD vector > instructions. Note that the KNL has 16 wide float SIMD, while the > exploit module uses only 4. However, the backend should be able to > handle this. > > Unless I receive further ideas I will file an official bug report. > > Frank > > > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev >-- Hal Finkel Assistant Computational Scientist Leadership Computing Facility Argonne National Laboratory
Frank Winter via llvm-dev
2016-Jun-30 16:49 UTC
[llvm-dev] avx512 JIT backend generates wrong code on <4 x float>
Hi Hal! Thanks, but unfortunately it didn't help. The exact same assembler instructions are generated for both 3.8 (yesterday) and trunk (from today). So, this really looks like a bug. Best, Frank On 06/29/2016 03:48 PM, Hal Finkel wrote:> Hi Frank, > > I recommend trying trunk LLVM. AVX-512 development has been very active recently. > > -Hal > > ----- Original Message ----- >> From: "Frank Winter via llvm-dev" <llvm-dev at lists.llvm.org> >> To: "LLVM Dev" <llvm-dev at lists.llvm.org> >> Sent: Wednesday, June 29, 2016 2:41:39 PM >> Subject: [llvm-dev] avx512 JIT backend generates wrong code on <4 x float> >> >> Hi! >> >> When compiling the attached module with the JIT engine on an Intel >> KNL I >> see wrong code getting emitted. I attach a complete exploit program >> which shows the bug in LLVM 3.8. It loads and JIT compiles the module >> and prints the assembler. I stumbled on this since the result of an >> actual calculation was wrong. So, it's not only the text version of >> the >> assembler also the machine assembler is wrong. >> >> When I execute the exploit program on an Intel KNL the following >> output >> is produced: >> >> CPU name = knl >> -sse4a,-avx512bw,cx16,-tbm,xsave,-fma4,-avx512vl,prfchw,bmi2,adx,-xsavec,fsgsbase,avx,avx512cd,avx512pf,-rtm,popcnt,fma,bmi,aes,rdrnd,-xsaves,sse4.1,sse4.2,avx2,avx512er,sse,lzcnt,pclmul,avx512f,f16c,ssse3,mmx,-pku,cmov,-xop,rdseed,movbe,-hle,xsaveopt,-sha,sse2,sse3,-avx512dq, >> Assembly: >> .text >> .file "module_KFxOBX_i4_after.ll" >> .globl adjmul >> .align 16, 0x90 >> .type adjmul, at function >> adjmul: >> .cfi_startproc >> leaq (%rdi,%r8), %rdx >> addq %rsi, %r8 >> testb $1, %cl >> cmoveq %rdi, %rdx >> cmoveq %rsi, %r8 >> movq %rdx, %rax >> sarq $63, %rax >> shrq $62, %rax >> addq %rdx, %rax >> sarq $2, %rax >> movq %r8, %rcx >> sarq $63, %rcx >> shrq $62, %rcx >> addq %r8, %rcx >> sarq $2, %rcx >> movq %rax, %rdx >> shlq $5, %rdx >> leaq 16(%r9,%rdx), %rsi >> orq $16, %rdx >> movq 16(%rsp), %rdi >> addq %rdx, %rdi >> addq 8(%rsp), %rdx >> .align 16, 0x90 >> .LBB0_1: >> vmovaps -16(%rdx), %xmm0 >> vmovaps (%rdx), %xmm1 >> vmovaps -16(%rdi), %xmm2 >> vmovaps (%rdi), %xmm3 >> vmulps %xmm3, %xmm1, %xmm4 >> vmulps %xmm2, %xmm1, %xmm1 >> vfmadd213ss %xmm4, %xmm0, %xmm2 >> vfmsub213ss %xmm1, %xmm0, %xmm3 >> vmovaps %xmm2, -16(%rsi) >> vmovaps %xmm3, (%rsi) >> addq $1, %rax >> addq $32, %rsi >> addq $32, %rdi >> addq $32, %rdx >> cmpq %rcx, %rax >> jl .LBB0_1 >> retq >> .Lfunc_end0: >> .size adjmul, .Lfunc_end0-adjmul >> .cfi_endproc >> >> >> .section ".note.GNU-stack","", at progbits >> >> end assembly! >> >> >> The instructions 'vfmadd213ss' are 'Fused Multiply-Add of Scalar >> Single-Precision Floating-Point'. Those should be SIMD vector >> instructions. Note that the KNL has 16 wide float SIMD, while the >> exploit module uses only 4. However, the backend should be able to >> handle this. >> >> Unless I receive further ideas I will file an official bug report. >> >> Frank >> >> >> _______________________________________________ >> LLVM Developers mailing list >> llvm-dev at lists.llvm.org >> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev >>
Apparently Analagous Threads
- avx512 JIT backend generates wrong code on <4 x float>
- avx512 JIT backend generates wrong code on <4 x float>
- AVX512 instruction generated when JIT compiling for an avx2 architecture
- AVX512 instruction generated when JIT compiling for an avx2 architecture
- [LLVMdev] SIMD for sdiv <2 x i64>