[mlir][spirv] Add atan and atan2 pattern to MathToSPIRV Conversion pass #102633

meehatpa · 2024-08-09T15:56:47Z

Add missing math.atan to spirv.CL.atan and math.atan2 to spirv.CL.atan2 in MathToSPIRV.
Add math.atan to spirv.GL.atan too.

github-actions · 2024-08-09T15:57:06Z

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be
notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write
permissions for the repository. In which case you can instead tag reviewers by
name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review
by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate
is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

llvmbot · 2024-08-09T15:57:37Z

@llvm/pr-subscribers-mlir-spirv

@llvm/pr-subscribers-mlir

Author: None (meehatpa)

Changes

Add missing math.atan to spirv.CL.atan and math.atan2 to spirv.CL.atan2 in MathToSPIRV.

Full diff: https://github.com/llvm/llvm-project/pull/102633.diff

2 Files Affected:

(modified) mlir/lib/Conversion/MathToSPIRV/MathToSPIRV.cpp (+2)
(modified) mlir/test/Conversion/MathToSPIRV/math-to-opencl-spirv.mlir (+51-39)

diff --git a/mlir/lib/Conversion/MathToSPIRV/MathToSPIRV.cpp b/mlir/lib/Conversion/MathToSPIRV/MathToSPIRV.cpp
index 0b29c93e2d8909..1260b27a0a751f 100644
--- a/mlir/lib/Conversion/MathToSPIRV/MathToSPIRV.cpp
+++ b/mlir/lib/Conversion/MathToSPIRV/MathToSPIRV.cpp
@@ -431,6 +431,8 @@ void populateMathToSPIRVPatterns(SPIRVTypeConverter &typeConverter,
   patterns.add<Log1pOpPattern<spirv::CLLogOp>, ExpM1OpPattern<spirv::CLExpOp>,
                CheckedElementwiseOpPattern<math::AbsFOp, spirv::CLFAbsOp>,
                CheckedElementwiseOpPattern<math::AbsIOp, spirv::CLSAbsOp>,
+               CheckedElementwiseOpPattern<math::AtanOp, spirv::CLAtanOp>,
+               CheckedElementwiseOpPattern<math::Atan2Op, spirv::CLAtan2Op>,
                CheckedElementwiseOpPattern<math::CeilOp, spirv::CLCeilOp>,
                CheckedElementwiseOpPattern<math::CosOp, spirv::CLCosOp>,
                CheckedElementwiseOpPattern<math::ErfOp, spirv::CLErfOp>,
diff --git a/mlir/test/Conversion/MathToSPIRV/math-to-opencl-spirv.mlir b/mlir/test/Conversion/MathToSPIRV/math-to-opencl-spirv.mlir
index 03a5c2f2bc9b15..e9ca838354c0de 100644
--- a/mlir/test/Conversion/MathToSPIRV/math-to-opencl-spirv.mlir
+++ b/mlir/test/Conversion/MathToSPIRV/math-to-opencl-spirv.mlir
@@ -4,83 +4,91 @@ module attributes { spirv.target_env = #spirv.target_env<#spirv.vce<v1.0, [Kerne
 
 // CHECK-LABEL: @float32_unary_scalar
 func.func @float32_unary_scalar(%arg0: f32) {
+  // CHECK: spirv.CL.atan %{{.*}}: f32
+  %0 = math.atan %arg0 : f32
   // CHECK: spirv.CL.cos %{{.*}}: f32
-  %0 = math.cos %arg0 : f32
+  %1 = math.cos %arg0 : f32
   // CHECK: spirv.CL.exp %{{.*}}: f32
-  %1 = math.exp %arg0 : f32
+  %2 = math.exp %arg0 : f32
   // CHECK: %[[EXP:.+]] = spirv.CL.exp %arg0
   // CHECK: %[[ONE:.+]] = spirv.Constant 1.000000e+00 : f32
   // CHECK: spirv.FSub %[[EXP]], %[[ONE]]
-  %2 = math.expm1 %arg0 : f32
+  %3 = math.expm1 %arg0 : f32
   // CHECK: spirv.CL.log %{{.*}}: f32
-  %3 = math.log %arg0 : f32
+  %4 = math.log %arg0 : f32
   // CHECK: %[[ONE:.+]] = spirv.Constant 1.000000e+00 : f32
   // CHECK: %[[ADDONE:.+]] = spirv.FAdd %[[ONE]], %{{.+}}
   // CHECK: spirv.CL.log %[[ADDONE]]
-  %4 = math.log1p %arg0 : f32
+  %5 = math.log1p %arg0 : f32
   // CHECK: spirv.CL.rint %{{.*}}: f32
-  %5 = math.roundeven %arg0 : f32
+  %6 = math.roundeven %arg0 : f32
   // CHECK: spirv.CL.rsqrt %{{.*}}: f32
-  %6 = math.rsqrt %arg0 : f32
+  %7 = math.rsqrt %arg0 : f32
   // CHECK: spirv.CL.sqrt %{{.*}}: f32
-  %7 = math.sqrt %arg0 : f32
+  %8 = math.sqrt %arg0 : f32
   // CHECK: spirv.CL.tanh %{{.*}}: f32
-  %8 = math.tanh %arg0 : f32
+  %9 = math.tanh %arg0 : f32
   // CHECK: spirv.CL.sin %{{.*}}: f32
-  %9 = math.sin %arg0 : f32
+  %10 = math.sin %arg0 : f32
   // CHECK: spirv.CL.fabs %{{.*}}: f32
-  %10 = math.absf %arg0 : f32
+  %11 = math.absf %arg0 : f32
   // CHECK: spirv.CL.ceil %{{.*}}: f32
-  %11 = math.ceil %arg0 : f32
+  %12 = math.ceil %arg0 : f32
   // CHECK: spirv.CL.floor %{{.*}}: f32
-  %12 = math.floor %arg0 : f32
+  %13 = math.floor %arg0 : f32
   // CHECK: spirv.CL.erf %{{.*}}: f32
-  %13 = math.erf %arg0 : f32
+  %14 = math.erf %arg0 : f32
   // CHECK: spirv.CL.round %{{.*}}: f32
-  %14 = math.round %arg0 : f32
+  %15 = math.round %arg0 : f32
   return
 }
 
 // CHECK-LABEL: @float32_unary_vector
 func.func @float32_unary_vector(%arg0: vector<3xf32>) {
+  // CHECK: spirv.CL.atan %{{.*}}: vector<3xf32>
+  %0 = math.atan %arg0 : vector<3xf32>
   // CHECK: spirv.CL.cos %{{.*}}: vector<3xf32>
-  %0 = math.cos %arg0 : vector<3xf32>
+  %1 = math.cos %arg0 : vector<3xf32>
   // CHECK: spirv.CL.exp %{{.*}}: vector<3xf32>
-  %1 = math.exp %arg0 : vector<3xf32>
+  %2 = math.exp %arg0 : vector<3xf32>
   // CHECK: %[[EXP:.+]] = spirv.CL.exp %arg0
   // CHECK: %[[ONE:.+]] = spirv.Constant dense<1.000000e+00> : vector<3xf32>
   // CHECK: spirv.FSub %[[EXP]], %[[ONE]]
-  %2 = math.expm1 %arg0 : vector<3xf32>
+  %3 = math.expm1 %arg0 : vector<3xf32>
   // CHECK: spirv.CL.log %{{.*}}: vector<3xf32>
-  %3 = math.log %arg0 : vector<3xf32>
+  %4 = math.log %arg0 : vector<3xf32>
   // CHECK: %[[ONE:.+]] = spirv.Constant dense<1.000000e+00> : vector<3xf32>
   // CHECK: %[[ADDONE:.+]] = spirv.FAdd %[[ONE]], %{{.+}}
   // CHECK: spirv.CL.log %[[ADDONE]]
-  %4 = math.log1p %arg0 : vector<3xf32>
+  %5 = math.log1p %arg0 : vector<3xf32>
   // CHECK: spirv.CL.rint %{{.*}}: vector<3xf32>
-  %5 = math.roundeven %arg0 : vector<3xf32>
+  %6 = math.roundeven %arg0 : vector<3xf32>
   // CHECK: spirv.CL.rsqrt %{{.*}}: vector<3xf32>
-  %6 = math.rsqrt %arg0 : vector<3xf32>
+  %7 = math.rsqrt %arg0 : vector<3xf32>
   // CHECK: spirv.CL.sqrt %{{.*}}: vector<3xf32>
-  %7 = math.sqrt %arg0 : vector<3xf32>
+  %8 = math.sqrt %arg0 : vector<3xf32>
   // CHECK: spirv.CL.tanh %{{.*}}: vector<3xf32>
-  %8 = math.tanh %arg0 : vector<3xf32>
+  %9 = math.tanh %arg0 : vector<3xf32>
   // CHECK: spirv.CL.sin %{{.*}}: vector<3xf32>
-  %9 = math.sin %arg0 : vector<3xf32>
+  %10 = math.sin %arg0 : vector<3xf32>
   return
 }
 
 // CHECK-LABEL: @float32_binary_scalar
 func.func @float32_binary_scalar(%lhs: f32, %rhs: f32) {
+  // CHECK: spirv.CL.atan2 %{{.*}}: f32
+  %0 = math.atan2 %lhs, %rhs : f32
   // CHECK: spirv.CL.pow %{{.*}}: f32
-  %0 = math.powf %lhs, %rhs : f32
+  %1 = math.powf %lhs, %rhs : f32
   return
 }
 
 // CHECK-LABEL: @float32_binary_vector
 func.func @float32_binary_vector(%lhs: vector<4xf32>, %rhs: vector<4xf32>) {
+  // CHECK: spirv.CL.atan2 %{{.*}}: vector<4xf32>
+  %0 = math.atan2 %lhs, %rhs : vector<4xf32>
   // CHECK: spirv.CL.pow %{{.*}}: vector<4xf32>
-  %0 = math.powf %lhs, %rhs : vector<4xf32>
+  %1 = math.powf %lhs, %rhs : vector<4xf32>
   return
 }
 
@@ -118,18 +126,20 @@ module attributes {
 
 // CHECK-LABEL: @vector_2d
 func.func @vector_2d(%arg0: vector<2x2xf32>) {
+  // CHECK-NEXT: math.atan {{.+}} : vector<2x2xf32>
+  %0 = math.atan %arg0 : vector<2x2xf32>
   // CHECK-NEXT: math.cos {{.+}} : vector<2x2xf32>
-  %0 = math.cos %arg0 : vector<2x2xf32>
+  %1 = math.cos %arg0 : vector<2x2xf32>
   // CHECK-NEXT: math.exp {{.+}} : vector<2x2xf32>
-  %1 = math.exp %arg0 : vector<2x2xf32>
+  %2 = math.exp %arg0 : vector<2x2xf32>
   // CHECK-NEXT: math.absf {{.+}} : vector<2x2xf32>
-  %2 = math.absf %arg0 : vector<2x2xf32>
+  %3 = math.absf %arg0 : vector<2x2xf32>
   // CHECK-NEXT: math.ceil {{.+}} : vector<2x2xf32>
-  %3 = math.ceil %arg0 : vector<2x2xf32>
+  %4 = math.ceil %arg0 : vector<2x2xf32>
   // CHECK-NEXT: math.floor {{.+}} : vector<2x2xf32>
-  %4 = math.floor %arg0 : vector<2x2xf32>
+  %5 = math.floor %arg0 : vector<2x2xf32>
   // CHECK-NEXT: math.powf {{.+}}, {{%.+}} : vector<2x2xf32>
-  %5 = math.powf %arg0, %arg0 : vector<2x2xf32>
+  %6 = math.powf %arg0, %arg0 : vector<2x2xf32>
   // CHECK-NEXT: return
   return
 }
@@ -138,18 +148,20 @@ func.func @vector_2d(%arg0: vector<2x2xf32>) {
 
 // CHECK-LABEL: @tensor_1d
 func.func @tensor_1d(%arg0: tensor<2xf32>) {
+  // CHECK-NEXT: math.atan {{.+}} : tensor<2xf32>
+  %0 = math.atan %arg0 : tensor<2xf32>
   // CHECK-NEXT: math.cos {{.+}} : tensor<2xf32>
-  %0 = math.cos %arg0 : tensor<2xf32>
+  %1 = math.cos %arg0 : tensor<2xf32>
   // CHECK-NEXT: math.exp {{.+}} : tensor<2xf32>
-  %1 = math.exp %arg0 : tensor<2xf32>
+  %2 = math.exp %arg0 : tensor<2xf32>
   // CHECK-NEXT: math.absf {{.+}} : tensor<2xf32>
-  %2 = math.absf %arg0 : tensor<2xf32>
+  %3 = math.absf %arg0 : tensor<2xf32>
   // CHECK-NEXT: math.ceil {{.+}} : tensor<2xf32>
-  %3 = math.ceil %arg0 : tensor<2xf32>
+  %4 = math.ceil %arg0 : tensor<2xf32>
   // CHECK-NEXT: math.floor {{.+}} : tensor<2xf32>
-  %4 = math.floor %arg0 : tensor<2xf32>
+  %5 = math.floor %arg0 : tensor<2xf32>
   // CHECK-NEXT: math.powf {{.+}}, {{%.+}} : tensor<2xf32>
-  %5 = math.powf %arg0, %arg0 : tensor<2xf32>
+  %6 = math.powf %arg0, %arg0 : tensor<2xf32>
   // CHECK-NEXT: return
   return
 }

kuhar

Could we add glsl support in one sweep?
https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html#:~:text=%3Cid%3E%0Ax-,Atan%0A%0AArc%20tangent,-.%20Result%20is%20an

See the new comment

meehatpa · 2024-08-09T17:24:17Z

Could we add glsl support in one sweep? https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html#:~:text=%3Cid%3E%0Ax-,Atan%0A%0AArc%20tangent,-.%20Result%20is%20an

I updated it for glsl only for atan as atan2 op is not available in spirv.
Thanks for the quick review!

kuhar · 2024-08-09T18:07:27Z

Could we add glsl support in one sweep? https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html#:~:text=%3Cid%3E%0Ax-,Atan%0A%0AArc%20tangent,-.%20Result%20is%20an

I updated it for glsl only for atan as atan2 op is not available in spirv. Thanks for the quick review!

Can we use this one: https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html#:~:text=Atan2%0A%0AArc%20tangent.%20Result%20is%20an%20angle ?

meehatpa · 2024-08-09T18:23:10Z

Could we add glsl support in one sweep? https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html#:~:text=%3Cid%3E%0Ax-,Atan%0A%0AArc%20tangent,-.%20Result%20is%20an

I updated it for glsl only for atan as atan2 op is not available in spirv. Thanks for the quick review!

Can we use this one: https://registry.khronos.org/SPIR-V/specs/unified1/GLSL.std.450.html#:~:text=Atan2%0A%0AArc%20tangent.%20Result%20is%20an%20angle ?

The issue is it is not available in spirv mlir dialect https://mlir.llvm.org/docs/Dialects/SPIR-V/. We would need to add that op in the dialect first.

Add missing math.atan to spirv.CL.atan and math.atan2 to spirv.CL.atan2 in MathToSPIRV.

meehatpa · 2024-08-12T14:27:44Z

@kuhar would you please merge?

github-actions · 2024-08-12T14:59:11Z

@meehatpa Congratulations on having your first Pull Request (PR) merged into the LLVM Project!

Your changes will be combined with recent changes from other authors, then tested
by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR.

Please check whether problems have been caused by your change specifically, as
the builds can include changes from many authors. It is not uncommon for your
change to be included in a build that fails due to someone else's changes, or
infrastructure issues.

How to do this, and the rest of the post-merge process, is covered in detail here.

If your change does cause a problem, it may be reverted, or you can revert it yourself.
This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again.

If you don't get any reports, no action is required from you. Your changes are working as expected, well done!

kuhar · 2024-08-12T14:59:11Z

@kuhar would you please merge?

Done

meehatpa requested review from antiagainst and kuhar as code owners August 9, 2024 15:56

llvmbot added mlir:spirv mlir labels Aug 9, 2024

kuhar previously approved these changes Aug 9, 2024

View reviewed changes

kuhar reviewed Aug 9, 2024

View reviewed changes

kuhar self-requested a review August 9, 2024 17:01

meehatpa force-pushed the atan_atan2_spirv branch from bcbf003 to 9f97687 Compare August 9, 2024 17:22

[mlir][spirv] Add atan and atan2 pattern to MathToSPIRV Conversion pass

9f97687

Add missing math.atan to spirv.CL.atan and math.atan2 to spirv.CL.atan2 in MathToSPIRV.

kuhar approved these changes Aug 10, 2024

View reviewed changes

Merge branch 'main' into atan_atan2_spirv

2c1f738

kuhar merged commit 49777d7 into llvm:main Aug 12, 2024
6 of 7 checks passed

meehatpa deleted the atan_atan2_spirv branch August 12, 2024 15:10

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[mlir][spirv] Add atan and atan2 pattern to MathToSPIRV Conversion pass #102633

[mlir][spirv] Add atan and atan2 pattern to MathToSPIRV Conversion pass #102633

Uh oh!

meehatpa commented Aug 9, 2024 •

edited

Loading

Uh oh!

github-actions bot commented Aug 9, 2024

Uh oh!

llvmbot commented Aug 9, 2024 •

edited

Loading

Uh oh!

kuhar left a comment

Uh oh!

meehatpa commented Aug 9, 2024

Uh oh!

kuhar commented Aug 9, 2024

Uh oh!

meehatpa commented Aug 9, 2024 •

edited

Loading

Uh oh!

meehatpa commented Aug 12, 2024

Uh oh!

Uh oh!

github-actions bot commented Aug 12, 2024

Uh oh!

kuhar commented Aug 12, 2024

Uh oh!

Uh oh!

[mlir][spirv] Add atan and atan2 pattern to MathToSPIRV Conversion pass #102633

[mlir][spirv] Add atan and atan2 pattern to MathToSPIRV Conversion pass #102633

Uh oh!

Conversation

meehatpa commented Aug 9, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Aug 9, 2024

Uh oh!

llvmbot commented Aug 9, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kuhar left a comment

Choose a reason for hiding this comment

Uh oh!

meehatpa commented Aug 9, 2024

Uh oh!

kuhar commented Aug 9, 2024

Uh oh!

meehatpa commented Aug 9, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

meehatpa commented Aug 12, 2024

Uh oh!

Uh oh!

github-actions bot commented Aug 12, 2024

Uh oh!

kuhar commented Aug 12, 2024

Uh oh!

Uh oh!

meehatpa commented Aug 9, 2024 •

edited

Loading

llvmbot commented Aug 9, 2024 •

edited

Loading

meehatpa commented Aug 9, 2024 •

edited

Loading