Skip to content

[LoopVectorize][X86][AMDLibm] Add Missing AMD LibM trig vector intrinsics #101125

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.

Sign up for GitHub

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

Merged
merged 1 commit into from
Aug 12, 2024
Merged

[LoopVectorize][X86][AMDLibm] Add Missing AMD LibM trig vector intrinsics #101125

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions llvm/include/llvm/Analysis/VecFuncs.def
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1371,7 +1371,9 @@ TLI_DEFINE_VECFUNC("llvm.asin.f32", "amd_vrs16_asinf", FIXED(16), NOMASK, "_ZGV_

TLI_DEFINE_VECFUNC("acosf", "amd_vrs4_acosf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")
TLI_DEFINE_VECFUNC("acosf", "amd_vrs8_acosf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")
TLI_DEFINE_VECFUNC("acosf", "amd_vrs16_acosf", FIXED(16), NOMASK, "_ZGV_LLVM_N16v")
Copy link
Member Author

@farzonl farzonl Jul 30, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@rohitaggarwal007 git blame shows you added the amd vector lib support: #78560. Can you review this change?

@phoebewang you approved the mentioned pr so if you could also take a look i'd appreciate it.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@phoebewang gentle ping. Could you take a look at this PR? Also @RKSimon you commented on the last PR (#78560) could you take a look at this PR as well?


TLI_DEFINE_VECFUNC("llvm.acos.f32", "amd_vrs16_acosf", FIXED(16), NOMASK, "_ZGV_LLVM_N16v")
TLI_DEFINE_VECFUNC("llvm.acos.f32", "amd_vrs8_acosf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")
TLI_DEFINE_VECFUNC("llvm.acos.f32", "amd_vrs4_acosf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")

Expand All @@ -1389,17 +1391,21 @@ TLI_DEFINE_VECFUNC("llvm.atan.f32", "amd_vrs4_atanf", FIXED(4), NOMASK, "_ZGV_LL
TLI_DEFINE_VECFUNC("llvm.atan.f32", "amd_vrs8_atanf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")
TLI_DEFINE_VECFUNC("llvm.atan.f32", "amd_vrs16_atanf", FIXED(16), NOMASK, "_ZGV_LLVM_N16v")

TLI_DEFINE_VECFUNC("cosh", "amd_vrd2_cosh" , FIXED(2), NOMASK, "_ZGV_LLVM_N2v")
TLI_DEFINE_VECFUNC("coshf", "amd_vrs4_coshf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")
TLI_DEFINE_VECFUNC("coshf", "amd_vrs8_coshf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")

TLI_DEFINE_VECFUNC("llvm.cosh.f64", "amd_vrd2_cosh" , FIXED(2), NOMASK, "_ZGV_LLVM_N2v")
TLI_DEFINE_VECFUNC("llvm.cosh.f32", "amd_vrs4_coshf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")
TLI_DEFINE_VECFUNC("llvm.cosh.f32", "amd_vrs8_coshf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")

TLI_DEFINE_VECFUNC("tanhf", "amd_vrs4_tanhf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")
TLI_DEFINE_VECFUNC("tanhf", "amd_vrs8_tanhf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")
TLI_DEFINE_VECFUNC("tanhf", "amd_vrs16_tanhf", FIXED(16), NOMASK, "_ZGV_LLVM_N16v")

TLI_DEFINE_VECFUNC("llvm.tanh.f32", "amd_vrs4_tanhf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")
TLI_DEFINE_VECFUNC("llvm.tanh.f32", "amd_vrs8_tanhf", FIXED(8), NOMASK, "_ZGV_LLVM_N8v")
TLI_DEFINE_VECFUNC("llvm.tanh.f32", "amd_vrs16_tanhf", FIXED(16), NOMASK, "_ZGV_LLVM_N16v")

TLI_DEFINE_VECFUNC("cbrt", "amd_vrd2_cbrt", FIXED(2), NOMASK, "_ZGV_LLVM_N2v")
TLI_DEFINE_VECFUNC("cbrtf", "amd_vrs4_cbrtf", FIXED(4), NOMASK, "_ZGV_LLVM_N4v")
Expand Down
62 changes: 62 additions & 0 deletions llvm/test/Transforms/LoopVectorize/X86/amdlibm-calls.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -412,6 +412,10 @@ define void @acos_f32(ptr nocapture %varray) {
; CHECK: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_acosf(<4 x float> [[TMP4:%.*]])
; CHECK: ret void
;
; CHECK-AVX512-VF16-LABEL: @acos_f32(
; CHECK-AVX512-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_acosf(<16 x float> [[TMP4:%.*]])
; CHECK-AVX512-VF16: ret void
;
entry:
br label %for.body

Expand All @@ -435,6 +439,10 @@ define void @acos_f32_intrinsic(ptr nocapture %varray) {
; CHECK: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_acosf(<4 x float> [[TMP4:%.*]])
; CHECK: ret void
;
; CHECK-AVX512-VF16-LABEL: @acos_f32_intrinsic(
; CHECK-AVX512-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_acosf(<16 x float> [[TMP4:%.*]])
; CHECK-AVX512-VF16: ret void
;
entry:
br label %for.body

Expand Down Expand Up @@ -669,6 +677,29 @@ for.end:
ret void
}

define void @cosh_f64(ptr nocapture %varray) {
; CHECK-AVX-VF2-LABEL: @cosh_f64(
; CHECK-AVX-VF2: [[TMP5:%.*]] = call <2 x double> @amd_vrd2_cosh(<2 x double> [[TMP4:%.*]])
; CHECK-AVX-VF2: ret void
;
entry:
br label %for.body

for.body:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
%tmp = trunc i64 %iv to i32
%conv = sitofp i32 %tmp to double
%call = tail call double @cosh(double %conv)
%arrayidx = getelementptr inbounds double, ptr %varray, i64 %iv
store double %call, ptr %arrayidx, align 4
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, 1000
br i1 %exitcond, label %for.end, label %for.body

for.end:
ret void
}

define void @cosh_f32(ptr nocapture %varray) {
; CHECK-LABEL: @cosh_f32(
; CHECK: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_coshf(<4 x float> [[TMP4:%.*]])
Expand All @@ -692,6 +723,29 @@ for.end:
ret void
}

define void @cosh_f64_intrinsic(ptr nocapture %varray) {
; CHECK-AVX-VF2-LABEL: @cosh_f64_intrinsic(
; CHECK-AVX-VF2: [[TMP5:%.*]] = call <2 x double> @amd_vrd2_cosh(<2 x double> [[TMP4:%.*]])
; CHECK-AVX-VF2: ret void
;
entry:
br label %for.body

for.body:
%iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]
%tmp = trunc i64 %iv to i32
%conv = sitofp i32 %tmp to double
%call = tail call double @llvm.cosh.f64(double %conv)
%arrayidx = getelementptr inbounds double, ptr %varray, i64 %iv
store double %call, ptr %arrayidx, align 4
%iv.next = add nuw nsw i64 %iv, 1
%exitcond = icmp eq i64 %iv.next, 1000
br i1 %exitcond, label %for.end, label %for.body

for.end:
ret void
}

define void @cosh_f32_intrinsic(ptr nocapture %varray) {
; CHECK-LABEL: @cosh_f32_intrinsic(
; CHECK: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_coshf(<4 x float> [[TMP4:%.*]])
Expand Down Expand Up @@ -720,6 +774,10 @@ define void @tanh_f32(ptr nocapture %varray) {
; CHECK: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_tanhf(<4 x float> [[TMP4:%.*]])
; CHECK: ret void
;
; CHECK-AVX512-VF16-LABEL: @tanh_f32(
; CHECK-AVX512-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_tanhf(<16 x float> [[TMP4:%.*]])
; CHECK-AVX512-VF16: ret void
;
entry:
br label %for.body

Expand All @@ -743,6 +801,10 @@ define void @tanh_f32_intrinsic(ptr nocapture %varray) {
; CHECK: [[TMP5:%.*]] = call <4 x float> @amd_vrs4_tanhf(<4 x float> [[TMP4:%.*]])
; CHECK: ret void
;
; CHECK-AVX512-VF16-LABEL: @tanh_f32_intrinsic(
; CHECK-AVX512-VF16: [[TMP5:%.*]] = call <16 x float> @amd_vrs16_tanhf(<16 x float> [[TMP4:%.*]])
; CHECK-AVX512-VF16: ret void
;
entry:
br label %for.body

Expand Down
Loading