mirrors/ZLUDA
0
0
Fork 0
mirror of https://github.com/vosen/ZLUDA.git synced 2025-09-29 12:49:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

Disable wave mode IEEE

Browse source
This commit is contained in:
Violet 2025-09-12 18:51:33 +00:00
commit 5307412337
4 changed files with 4 additions and 3 deletions
Download patch file Download diff file Expand all files Collapse all files

1
ptx/src/pass/llvm/emit.rs
View file

@ -149,6 +149,7 @@ impl<'a, 'input> ModuleEmitContext<'a, 'input> {
llvm_ftz(method.flush_to_zero_f16f64),
);
}
self.emit_fn_attribute(fn_, "amdgpu-ieee", "false");
for (i, param) in method.input_arguments.iter().enumerate() {
let value = unsafe { LLVMGetParam(fn_, i as u32) };
let name = self.resolver.get_or_add(param.name);

2
ptx/src/test/ll/min.ll
View file

@ -36,5 +36,5 @@ define amdgpu_kernel void @min(ptr addrspace(4) byref(i64) %"36", ptr addrspace(
; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare i32 @llvm.smin.i32(i32, i32) #1
attributes #0 = { "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="preserve-sign" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
attributes #0 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="preserve-sign" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
attributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }

2
ptx/src/test/ll/min_f16.ll
View file

@ -39,5 +39,5 @@ define amdgpu_kernel void @min_f16(ptr addrspace(4) byref(i64) %"36", ptr addrsp
; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare half @llvm.minnum.f16(half, half) #1
attributes #0 = { "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="ieee" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
attributes #0 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="ieee" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
attributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }

2
ptx/src/test/ll/min_nan_f16.ll
View file

@ -41,5 +41,5 @@ define amdgpu_kernel void @min_nan_f16(ptr addrspace(4) byref(i64) %"36", ptr ad
; Function Attrs: nocallback nofree nosync nounwind speculatable willreturn memory(none)
declare half @llvm.minnum.f16(half, half) #1
attributes #0 = { "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="ieee" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
attributes #0 = { "amdgpu-ieee"="false" "amdgpu-unsafe-fp-atomics"="true" "denormal-fp-math"="ieee" "denormal-fp-math-f32"="preserve-sign" "no-trapping-math"="true" "uniform-work-group-size"="true" }
attributes #1 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }