New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[SIMD] code not recognized as all_true #50142
Comments
Sorry, copied from the wrong tab, here is a corrected test (Compiler Explorer: https://godbolt.org/z/ref5rf19z): #include <stdint.h>
#pragma clang diagnostic ignored "-Wmissing-prototypes"
/* Because of <#45959 >, not
* a WASM issue. */
#pragma clang diagnostic ignored "-Wsign-conversion"
typedef int8_t i8x16 __attribute__((__vector_size__(16)));
typedef int16_t i16x8 __attribute__((__vector_size__(16)));
typedef int32_t i32x4 __attribute__((__vector_size__(16)));
typedef int64_t i64x2 __attribute__((__vector_size__(16)));
int
i8x16_all_true(i8x16 a) {
int8_t r = 1;
#pragma omp simd reduction(&:r)
for (int i = 0 ; i < 16 ; i++) {
r &= !!a[i];
}
return r;
}
int
i8x16_all_true_intrin(i8x16 a) {
return __builtin_wasm_all_true_i8x16(a);
}
int
i16x8_all_true(i16x8 a) {
int16_t r = 1;
#pragma omp simd reduction(&:r)
for (int i = 0 ; i < 8 ; i++) {
r &= !!a[i];
}
return r;
}
int
i16x8_all_true_intrin(i16x8 a) {
return __builtin_wasm_all_true_i16x8(a);
}
int
i32x4_all_true(i32x4 a) {
int32_t r = 1;
#pragma omp simd reduction(&:r)
for (int i = 0 ; i < 4 ; i++) {
r &= !!a[i];
}
return r;
}
int
i32x4_all_true_intrin(i32x4 a) {
return __builtin_wasm_all_true_i32x4(a);
}
int
i64x2_all_true(i64x2 a) {
int64_t r = 1;
#pragma omp simd reduction(&:r)
for (int i = 0 ; i < 2 ; i++) {
r &= !!a[i];
}
return r;
}
int
i64x2_all_true_intrin(i64x2 a) {
return __builtin_wasm_all_true_i64x2(a);
} |
This one is going to be a bit trickier than llvm/llvm-bugzilla-archive#50796 because I don't believe there is an existing reduction intrinsic corresponding to this operation. |
I can confirm that this issue still exists in clang version 18.0.0 (https://github.com/llvm/llvm-project.git cf7d4f5) using the provided Compiler Explorer link: https://godbolt.org/z/ref5rf19z |
@llvm/issue-subscribers-openmp Author: None (llvmbot)
| | |
| --- | --- |
| Bugzilla Link | [50798](https://llvm.org/bz50798) |
| Version | trunk |
| OS | Windows NT |
| Reporter | LLVM Bugzilla Contributor |
| CC | @rotateright,@tlively |
Extended DescriptionLLVM doesn't recognize that it can optimize these functions to an *.all_true instruction.. This one might be a bit too specialized; it doesn't surprise me at all that LLVM doesn't generate an *.all_true instruction. Still, it would be nice since the scalar version is pretty bad. I've included an OpenMP SIMD annotation to try to help the compiler, but obviously it would be better if it weren't required. Here is the example, or on Compiler Explorer at https://godbolt.org/z/x3xshzYrf if you prefer: #include <stdint.h> #pragma clang diagnostic ignored "-Wmissing-prototypes" /* Because of <#45959 >, not
typedef int8_t i8x16 attribute((vector_size(16))); int int int int int int int int |
Extended Description
LLVM doesn't recognize that it can optimize these functions to an *.all_true instruction.. This one might be a bit too specialized; it doesn't surprise me at all that LLVM doesn't generate an *.all_true instruction. Still, it would be nice since the scalar version is pretty bad.
I've included an OpenMP SIMD annotation to try to help the compiler, but obviously it would be better if it weren't required.
Here is the example, or on Compiler Explorer at https://godbolt.org/z/x3xshzYrf if you prefer:
The text was updated successfully, but these errors were encountered: