| Issue |
79258
|
| Summary |
[Flang] TSVC s118: not vectorized because LICM doesn't work
|
| Labels |
flang:ir,
vectorization
|
| Assignees |
|
| Reporter |
yus3710-fj
|
Flang can't vectorize the loop in `s118` of [TSVC](https://www.netlib.org/benchmark/vectors) while Clang can vectorize the loop written in C.
```fortran
! Fortran version
module mod
integer ld, nloops
parameter (ld=1000,nloops=135)
real a(ld), b(ld), c(ld), d(ld), e(ld)
real aa(ld,ld), bb(ld,ld), cc(ld,ld)
interface
subroutine dummy(ld,n,a,b,c,d,e,aa,bb,cc,x)
integer ld, n
real a(ld), b(ld), c(ld), d(ld), e(ld)
real aa(ld,ld), bb(ld,ld), cc(ld,ld)
real, value :: x
end subroutine
end interface
end module
subroutine s118 (n)
use mod
integer n, i, j
call init(ld,n,a,b,c,d,e,aa,bb,cc,'s118 ')
do 10 i = 2,n
do 20 j = 1,i-1
a(i) = a(i) + bb(i,j) * a(i-j)
20 continue
10 continue
call dummy(ld,n,a,b,c,d,e,aa,bb,cc,1.)
end
```
```c
// C version
#define LEN 32000
#define LEN2 256
float a[LEN], b[LEN], c[LEN], d[LEN], e[LEN];
float aa[LEN2][LEN2], bb[LEN2][LEN2], cc[LEN2][LEN2];
int s118() {
init( "s118 ");
for (int i = 1; i < LEN2; i++) {
for (int j = 0; j <= i - 1; j++) {
a[i] += bb[j][i] * a[i-j-1];
}
}
dummy(a, b, c, d, e, aa, bb, cc, 0.);
return 0;
}
```
```console
$ flang-new -v -Ofast s118.f -S -Rpass=vector
flang-new version 18.0.0 (https://github.com/llvm/llvm-project.git 2759e47067ea286f6302adcfe93b653cfaf6f2eb)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /path/to/install/bin
Found candidate GCC installation: /usr/lib/gcc/x86_64-linux-gnu/12
Selected GCC installation: /usr/lib/gcc/x86_64-linux-gnu/12
Candidate multilib: .;@m64
Selected multilib: .;@m64
"/path/to/install/bin/flang-new" -fc1 -triple x86_64-unknown-linux-gnu -emit-obj -fcolor-diagnostics -mrelocation-model pic -pic-level 2 -pic-is-pie -ffast-math -target-cpu x86-64 -fstack-arrays -fversion-loops-for-stride -mframe-pointer=none -O3 -o /tmp/s118-5868cd.o -x f95-cpp-input s118.f
$ clang -Ofast s118.c -S -Rpass=vector
/path/to/s118.c:16:4: remark: vectorized loop (vectorization width: 4, interleaved count: 2) [-Rpass=loop-vectorize]
16 | for (int j = 0; j <= i - 1; j++) {
| ^
```
Hoisting the store outside the loop is necessary for vectorization, but it doesn't work because BasicAA says `a(i)` and `a(i-j)` may alias each other.
It's similar to #74262 but BasicAA won't do complicated analyses, so I suspect it's difficult to fix BasicAA.
```llvm
25: ; preds = %.lr.ph, %25
%indvars.iv = phi i64 [ 1, %.lr.ph ], [ %indvars.iv.next, %25 ], !dbg !30
%26 = phi float [ %.promoted, %.lr.ph ], [ %32, %25 ], !dbg !30
%27 = mul nuw nsw i64 %indvars.iv, 1000, !dbg !30
%gep13 = getelementptr float, ptr %invariant.gep, i64 %27, !dbg !30
%28 = load float, ptr %gep13, align 4, !dbg !30, !tbaa !31
%29 = sub nuw nsw i64 %indvars.iv21, %indvars.iv, !dbg !30
%gep12 = getelementptr float, ptr getelementptr ([1000 x float], ptr @_QMmodEa, i64 -1, i64 999), i64 %29, !dbg !30
%30 = load float, ptr %gep12, align 4, !dbg !30, !tbaa !27
%31 = fmul fast float %30, %28, !dbg !30
%32 = fadd fast float %31, %26, !dbg !30
store float %32, ptr %24, align 4, !dbg !30, !tbaa !27 ;; this can be hoisted
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !33
%exitcond.not = icmp eq i64 %indvars.iv.next, %indvars.iv21, !dbg !26
br i1 %exitcond.not, label %._crit_edge, label %25, !dbg !26
```
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
