[llvm-dev] unable to emit vectorized code in LLVM IR
Francois Fayard via llvm-dev
llvm-dev at lists.llvm.org
Thu Aug 17 11:51:48 PDT 2017
Try that:
void f(int* a, int* b, int* c, int n) {
for (int i = 0; i < n; ++i) {
c[i] = a[i] + b[i];
}
}
and compile with: clang++ -S -O3 -mavx2 a.cpp -o a.assembly
and look at the a.assembly file. You’ll get something such as:
LBB0_12: ## =>This Inner Loop Header: Depth=1
vmovdqu -96(%rax), %ymm0
vmovdqu -64(%rax), %ymm1
vmovdqu -32(%rax), %ymm2
vmovdqu (%rax), %ymm3
vpaddd -96(%r11), %ymm0, %ymm0
vpaddd -64(%r11), %ymm1, %ymm1
vpaddd -32(%r11), %ymm2, %ymm2
vpaddd (%r11), %ymm3, %ymm3
vmovdqu %ymm0, -96(%rbx)
vmovdqu %ymm1, -64(%rbx)
vmovdqu %ymm2, -32(%rbx)
vmovdqu %ymm3, (%rbx)
subq $-128, %r11
subq $-128, %rax
subq $-128, %rbx
addq $-32, %r9
jne LBB0_12
That’s vectorized code, unrolled by 4. So you get 4 * (32 / 4) = 32 elements processed every loop. The ymm registers shows that you are using 256 bits vector registers as available on avx cpu. With avx512, you would get zmm registers.
François Fayard
> On Aug 17, 2017, at 8:44 PM, Craig Topper via llvm-dev <llvm-dev at lists.llvm.org> wrote:
>
> I assume compiler knows that your only have 2 input values that you just added together 1000 times.
>
> Despite the fact that you stored to a[i] and b[i] here, nothing reads them other than the addition in the same loop iteration. So the compiler easily removed the a and b arrays. Same with 'c', it's not read outside the loop so it doesn't need to exist. So the compiler turned your loop body back into g+= aa + bb; And since the loop is 1000 iterations and aa and bb never change this got further simplified to (aa+bb)*1000.
>
> int main(int argc, char** argv) {
> int a[1000], b[1000], c[1000]; int g=0;
> int aa=atoi(argv[1]), bb=atoi(argv[2]);
> for (int i=0; i<1000; i++) {
> a[i]=aa, b[i]=bb;
> c[i]=a[i] + b[i];
> g+=c[i];
> }
>
> ~Craig
>
> On Thu, Aug 17, 2017 at 11:37 AM, hameeza ahmed <hahmed2305 at gmail.com <mailto:hahmed2305 at gmail.com>> wrote:
> why is it happening? is there any way to solve this?
>
> On Thu, Aug 17, 2017 at 10:09 PM, hameeza ahmed <hahmed2305 at gmail.com <mailto:hahmed2305 at gmail.com>> wrote:
> even if i make my code as follows: vectorized instructions not get emitted. What to do?
>
> int main(int argc, char** argv) {
> int a[1000], b[1000], c[1000]; int g=0;
> int aa=atoi(argv[1]), bb=atoi(argv[2]);
> for (int i=0; i<1000; i++) {
> a[i]=aa, b[i]=bb;
> c[i]=a[i] + b[i];
> g+=c[i];
> }
>
> printf("sum: %d\n", g);
>
> return 0;
> }
>
> On Thu, Aug 17, 2017 at 10:03 PM, Craig Topper <craig.topper at gmail.com <mailto:craig.topper at gmail.com>> wrote:
> Did you remove the printf completely? Meaning that nothing accesses 'c' after the loop? If so it got removed as dead code because it had no visible effect.
>
> ~Craig
>
> On Thu, Aug 17, 2017 at 10:01 AM, hameeza ahmed <hahmed2305 at gmail.com <mailto:hahmed2305 at gmail.com>> wrote:
> i removed printf from loop. Now getting no error. but the IR doesnot contain vectorized code. IR Output is as follows:
> ; ModuleID = 'sum-vec.ll'
> source_filename = "sum-vec.c"
> target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
> target triple = "x86_64-unknown-linux-gnu"
>
> ; Function Attrs: norecurse nounwind readnone uwtable
> define i32 @main(i32, i8** nocapture readnone) local_unnamed_addr #0 {
> ret i32 0
> }
>
> attributes #0 = { norecurse nounwind readnone uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="knl" "target-features"="+adx,+aes,+avx,+avx2,+avx512cd,+avx512er,+avx512f,+avx512pf,+bmi,+bmi2,+cx16,+f16c,+fma,+fsgsbase,+fxsr,+lzcnt,+mmx,+movbe,+pclmul,+popcnt,+prefetchwt1,+rdrnd,+rdseed,+rtm,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt" "unsafe-fp-math"="false" "use-soft-float"="false" }
>
> !llvm.ident = !{!0}
>
> !0 = !{!"clang version 4.0.0 (tags/RELEASE_400/final)"}
>
> what to do? please help.
>
>
> On Thu, Aug 17, 2017 at 9:57 PM, Nemanja Ivanovic <nemanja.i.ibm at gmail.com <mailto:nemanja.i.ibm at gmail.com>> wrote:
> Move the printf out of the loop and it should vectorize just fine.
>
> On Thu, Aug 17, 2017 at 6:52 PM, hameeza ahmed <hahmed2305 at gmail.com <mailto:hahmed2305 at gmail.com>> wrote:
> I want to vectorize the user given inputs. when opt does vectorization user supplied inputs (from a text file) will be added using AVX vector instructions.
>
> as you pointed; When i changed my code to following:
>
> int main(int argc, char** argv) {
> int a[1000], b[1000], c[1000];
> int aa=atoi(argv[1]), bb=atoi(argv[2]);
> for (int i=0; i<1000; i++) {
> a[i]=aa, b[i]=bb;
> c[i]=a[i] + b[i];
> printf("sum: %d\n", c[i]);
>
> }
>
> I am getting error remark: <unknown>:0:0: loop not vectorized: call instruction cannot be vectorized.
>
> I am running following commands:
> clang -S -emit-llvm sum-vec.c -march=knl -O3 -mllvm -disable-llvm-optzns -o sum-vec.ll
> opt -S -O3 -force-vector-width=64 sum-vec.ll -o sum-vec03.ll
>
> How to achieve this? Please help.
>
>
>
>
>
> On Thu, Aug 17, 2017 at 10:44 AM, Nemanja Ivanovic <nemanja.i.ibm at gmail.com <mailto:nemanja.i.ibm at gmail.com>> wrote:
> I'm not sure what you expect to have vectorized here. If you look at the emitted code, there's no loop. It's just an add and a multiply as you might expect when adding a loop-invariant sum 1000 times in a loop.
>
> On Wed, Aug 16, 2017 at 11:38 PM, hameeza ahmed via llvm-dev <llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>> wrote:
> Hello,
> I have written the following code. when i try to vectorize it through opt. i am not getting vectorized instructions.
>
> #include <stdio.h>
> #include<stdlib.h>
> int main(int argc, char** argv) {
> int sum=0; int a=atoi(argv[1]); int b=atoi(argv[2]);
> for (int i=0;i<1000;i++)
> {
> sum+=a+b;
> }
>
> printf("sum: %d\n", sum);
> return 0;
> }
> i use following commands:
> clang -S -emit-llvm sum-main.c -march=knl -O3 -mllvm -disable-llvm-optzns -o sum-main.ll
> opt -S -O3 -force-vector-width=64 sum-main.ll -o sum-main03.ll
>
> why is that so? where am i doing mistake? i am not getting vectorized operations rather getting scalar operations.
>
> Please help.
>
> Thank You
>
> Regards
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org <mailto:llvm-dev at lists.llvm.org>
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev <http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev>
>
>
>
>
>
>
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20170817/15a264d2/attachment.html>
More information about the llvm-dev
mailing list