[Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs #113637

tex3d · 2024-10-25T02:56:59Z

This change is part of this proposal: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294

Return true for atan2 from isTriviallyVectorizable
Add atan2 to VecFuncs.def for massv and accelerate libraries.
Add atan2 to hasOptimizedCodeGen
Add atan2 support in llvm/lib/Analysis/ValueTracking.cpp llvm::getIntrinsicForCallSite and update vectorization tests
Add atan2 name check to isLoweredToCall in llvm/include/llvm/Analysis/TargetTransformInfoImpl.h

Thanks to @jroelofs for the atan2 accelerate veclib and associated test additions, plus the hasOptimizedCodeGen addition.

Part of: Implement the atan2 HLSL Function #70096.

llvmbot · 2024-10-25T02:57:33Z

@llvm/pr-subscribers-llvm-analysis

@llvm/pr-subscribers-llvm-transforms

Author: Tex Riddell (tex3d)

Changes

This change is part of this proposal: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294

This returns true from isTriviallyVectorizable for llvm.atan2 intrinsic. It also adds llvm atan2 intrinsic equivalents to VecFuncs.def for massv.

Part of: Implement the atan2 HLSL Function #70096.

Full diff: https://github.com/llvm/llvm-project/pull/113637.diff

3 Files Affected:

(modified) llvm/include/llvm/Analysis/VecFuncs.def (+2)
(modified) llvm/lib/Analysis/VectorUtils.cpp (+1)
(modified) llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll (+46)

diff --git a/llvm/include/llvm/Analysis/VecFuncs.def b/llvm/include/llvm/Analysis/VecFuncs.def
index c4586894e3e490..6f0ba3f9e406c4 100644
--- a/llvm/include/llvm/Analysis/VecFuncs.def
+++ b/llvm/include/llvm/Analysis/VecFuncs.def
@@ -289,7 +289,9 @@ TLI_DEFINE_VECFUNC("acosf", "__acosf4", FIXED(4), "_ZGV_LLVM_N4v")
 TLI_DEFINE_VECFUNC("atan", "__atand2", FIXED(2), "_ZGV_LLVM_N2v")
 TLI_DEFINE_VECFUNC("atanf", "__atanf4", FIXED(4), "_ZGV_LLVM_N4v")
 TLI_DEFINE_VECFUNC("atan2", "__atan2d2", FIXED(2), "_ZGV_LLVM_N2vv")
+TLI_DEFINE_VECFUNC("llvm.atan2.f64", "__atan2d2", FIXED(2), "_ZGV_LLVM_N2vv")
 TLI_DEFINE_VECFUNC("atan2f", "__atan2f4", FIXED(4), "_ZGV_LLVM_N4vv")
+TLI_DEFINE_VECFUNC("llvm.atan2.f32", "__atan2f4", FIXED(4), "_ZGV_LLVM_N4vv")
 
 // Hyperbolic Functions
 TLI_DEFINE_VECFUNC("sinh", "__sinhd2", FIXED(2), "_ZGV_LLVM_N2v")
diff --git a/llvm/lib/Analysis/VectorUtils.cpp b/llvm/lib/Analysis/VectorUtils.cpp
index 37c443011719b6..f3ffddeb40d4b4 100644
--- a/llvm/lib/Analysis/VectorUtils.cpp
+++ b/llvm/lib/Analysis/VectorUtils.cpp
@@ -69,6 +69,7 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::asin:
   case Intrinsic::acos:
   case Intrinsic::atan:
+  case Intrinsic::atan2:
   case Intrinsic::sin:
   case Intrinsic::cos:
   case Intrinsic::tan:
diff --git a/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll b/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll
index ba6e85e2852952..badc67866d2bad 100644
--- a/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll
+++ b/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll
@@ -1244,6 +1244,52 @@ for.end:
   ret void
 }
 
+define void @atan2_f64_intrinsic(ptr nocapture %varray) {
+; CHECK-LABEL: @atan2_f64_intrinsic(
+; CHECK: __atan2d2{{.*}}<2 x double>
+; CHECK: 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.atan2.f64(double %conv, 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 @atan2_f32_intrinsic(ptr nocapture %varray) {
+; CHECK-LABEL: @atan2_f32_intrinsic(
+; CHECK: __atan2f4{{.*}}<4 x float>
+; CHECK: 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 float
+  %call = tail call float @llvm.atan2.f32(float %conv, float %conv)
+  %arrayidx = getelementptr inbounds float, ptr %varray, i64 %iv
+  store float %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 @sinh_f64(ptr nocapture %varray) {
 ; CHECK-LABEL: @sinh_f64(
 ; CHECK: __sinhd2{{.*}}<2 x double>

llvmbot · 2024-10-25T02:57:33Z

@llvm/pr-subscribers-backend-powerpc

Author: Tex Riddell (tex3d)

Changes

This change is part of this proposal: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294

This returns true from isTriviallyVectorizable for llvm.atan2 intrinsic. It also adds llvm atan2 intrinsic equivalents to VecFuncs.def for massv.

Part of: Implement the atan2 HLSL Function #70096.

Full diff: https://github.com/llvm/llvm-project/pull/113637.diff

3 Files Affected:

(modified) llvm/include/llvm/Analysis/VecFuncs.def (+2)
(modified) llvm/lib/Analysis/VectorUtils.cpp (+1)
(modified) llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll (+46)

diff --git a/llvm/include/llvm/Analysis/VecFuncs.def b/llvm/include/llvm/Analysis/VecFuncs.def
index c4586894e3e490..6f0ba3f9e406c4 100644
--- a/llvm/include/llvm/Analysis/VecFuncs.def
+++ b/llvm/include/llvm/Analysis/VecFuncs.def
@@ -289,7 +289,9 @@ TLI_DEFINE_VECFUNC("acosf", "__acosf4", FIXED(4), "_ZGV_LLVM_N4v")
 TLI_DEFINE_VECFUNC("atan", "__atand2", FIXED(2), "_ZGV_LLVM_N2v")
 TLI_DEFINE_VECFUNC("atanf", "__atanf4", FIXED(4), "_ZGV_LLVM_N4v")
 TLI_DEFINE_VECFUNC("atan2", "__atan2d2", FIXED(2), "_ZGV_LLVM_N2vv")
+TLI_DEFINE_VECFUNC("llvm.atan2.f64", "__atan2d2", FIXED(2), "_ZGV_LLVM_N2vv")
 TLI_DEFINE_VECFUNC("atan2f", "__atan2f4", FIXED(4), "_ZGV_LLVM_N4vv")
+TLI_DEFINE_VECFUNC("llvm.atan2.f32", "__atan2f4", FIXED(4), "_ZGV_LLVM_N4vv")
 
 // Hyperbolic Functions
 TLI_DEFINE_VECFUNC("sinh", "__sinhd2", FIXED(2), "_ZGV_LLVM_N2v")
diff --git a/llvm/lib/Analysis/VectorUtils.cpp b/llvm/lib/Analysis/VectorUtils.cpp
index 37c443011719b6..f3ffddeb40d4b4 100644
--- a/llvm/lib/Analysis/VectorUtils.cpp
+++ b/llvm/lib/Analysis/VectorUtils.cpp
@@ -69,6 +69,7 @@ bool llvm::isTriviallyVectorizable(Intrinsic::ID ID) {
   case Intrinsic::asin:
   case Intrinsic::acos:
   case Intrinsic::atan:
+  case Intrinsic::atan2:
   case Intrinsic::sin:
   case Intrinsic::cos:
   case Intrinsic::tan:
diff --git a/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll b/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll
index ba6e85e2852952..badc67866d2bad 100644
--- a/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll
+++ b/llvm/test/Transforms/LoopVectorize/PowerPC/massv-calls.ll
@@ -1244,6 +1244,52 @@ for.end:
   ret void
 }
 
+define void @atan2_f64_intrinsic(ptr nocapture %varray) {
+; CHECK-LABEL: @atan2_f64_intrinsic(
+; CHECK: __atan2d2{{.*}}<2 x double>
+; CHECK: 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.atan2.f64(double %conv, 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 @atan2_f32_intrinsic(ptr nocapture %varray) {
+; CHECK-LABEL: @atan2_f32_intrinsic(
+; CHECK: __atan2f4{{.*}}<4 x float>
+; CHECK: 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 float
+  %call = tail call float @llvm.atan2.f32(float %conv, float %conv)
+  %arrayidx = getelementptr inbounds float, ptr %varray, i64 %iv
+  store float %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 @sinh_f64(ptr nocapture %varray) {
 ; CHECK-LABEL: @sinh_f64(
 ; CHECK: __sinhd2{{.*}}<2 x double>

jroelofs

LGTM

farzonl · 2024-11-06T09:44:55Z

Add libfunc mapping to intrinsic in llvm/lib/Analysis/ValueTracking.cpp llvm::getIntrinsicForCallSite
Add atan2 name check in llvm/include/llvm/Analysis/TargetTransformInfoImpl.h isLoweredToCall
I noticed there were some examples that use a amdlibm vector form below. Should we do a follow on PR that adds atan2 amdlibm and updates llvm/test/Transforms/LoopVectorize/X86/amdlibm-calls.ll?

amd_vrs4_atan2f
amd_vrs8_atan2f
amd_vrd2_atan2
amd_vrd4_atan2

This change is part of this proposal: https://discourse.llvm.org/t/rfc-all-the-math-intrinsics/78294 This returns true from isTriviallyVectorizable for llvm.atan2 intrinsic. It also adds llvm atan2 intrinsic equivalents to VecFuncs.def for massv. Part of: Implement the atan2 HLSL Function llvm#70096.

…n tests

…sis/TargetTransformInfoImpl.h` TBD: Add testing for this change.

tex3d · 2024-11-07T01:32:55Z

@farzonl

Add libfunc mapping to intrinsic in llvm/lib/Analysis/ValueTracking.cpp llvm::getIntrinsicForCallSite

Added.

Add atan2 name check in llvm/include/llvm/Analysis/TargetTransformInfoImpl.h isLoweredToCall

Added, but it had no test impact, so I'm looking for feedback on how this can be tested.

I noticed there were some examples that use a amdlibm vector form below. Should we do a follow on PR that adds atan2 amdlibm and updates llvm/test/Transforms/LoopVectorize/X86/amdlibm-calls.ll?

It turns out these example files have all the code under #if 0, and there doesn't appear to be any declarations of these vector operations in the header. I'll leave this for a later update when the operations are added to amdlibm.

farzonl · 2024-11-07T01:43:45Z

See this pr: https://github.com/llvm/llvm-project/pull/95517/files

tests like this should show that the libcall name to intrinsic mapping in the check line

llvm/test/Transforms/SLPVectorizer/X86/call.ll

llvmbot added backend:PowerPC llvm:analysis llvm:transforms labels Oct 25, 2024

tex3d mentioned this pull request Oct 25, 2024

[llvm] Improve TLI for Darwin libsystem_m functions #109479

Open

tex3d force-pushed the atan2-vectorize branch from e6c6df3 to 72efe7b Compare October 28, 2024 23:06

jroelofs approved these changes Oct 29, 2024

View reviewed changes

tex3d changed the title ~~[Analysis] isTriviallyVectorizable add vectorization for atan2 intrinsic~~ [Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs Oct 29, 2024

farzonl approved these changes Nov 6, 2024

View reviewed changes

tex3d added 4 commits November 6, 2024 12:19

Add atan2 to accelerate vector library and hasOptimizedCodeGen

146292e

[Analysis] Add atan2 support in ValueTracking and update vectorizatio…

65d6652

…n tests

Add atan2 name check to isLoweredToCall in `llvm/include/llvm/Analy…

e3b92d1

…sis/TargetTransformInfoImpl.h` TBD: Add testing for this change.

tex3d force-pushed the atan2-vectorize branch from 72efe7b to e3b92d1 Compare November 7, 2024 01:28

tex3d requested a review from nikic as a code owner November 7, 2024 01:28

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs #113637

[Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs #113637

tex3d commented Oct 25, 2024 •

edited

Loading

llvmbot commented Oct 25, 2024 •

edited

Loading

llvmbot commented Oct 25, 2024

jroelofs left a comment

farzonl commented Nov 6, 2024 •

edited

Loading

tex3d commented Nov 7, 2024

farzonl commented Nov 7, 2024

[Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs #113637

Are you sure you want to change the base?

[Analysis] atan2: isTriviallyVectorizable; add to massv and accelerate veclibs #113637

Conversation

tex3d commented Oct 25, 2024 • edited Loading

llvmbot commented Oct 25, 2024 • edited Loading

llvmbot commented Oct 25, 2024

jroelofs left a comment

Choose a reason for hiding this comment

farzonl commented Nov 6, 2024 • edited Loading

tex3d commented Nov 7, 2024

farzonl commented Nov 7, 2024

tex3d commented Oct 25, 2024 •

edited

Loading

llvmbot commented Oct 25, 2024 •

edited

Loading

farzonl commented Nov 6, 2024 •

edited

Loading