[MLIR][NVVM] Print ptxas path in debug output for "serialize-to-binary"

[grypp]

No description provided.

[llvmbot]

@llvm/pr-subscribers-mlir-llvm

@llvm/pr-subscribers-mlir-gpu

Author: Guray Ozen (grypp)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/132373.diff

2 Files Affected:

• (modified) mlir/lib/Target/LLVM/NVVM/Target.cpp (+1)
• (modified) mlir/test/Integration/GPU/CUDA/command-line-arg.mlir (+1)

diff --git a/mlir/lib/Target/LLVM/NVVM/Target.cpp b/mlir/lib/Target/LLVM/NVVM/Target.cpp
index fa8c597da58b1..69a247b06bf16 100644
--- a/mlir/lib/Target/LLVM/NVVM/Target.cpp
+++ b/mlir/lib/Target/LLVM/NVVM/Target.cpp
@@ -455,6 +455,7 @@ NVPTXSerializer::compileToBinary(const std::string &ptxCode) {
   LLVM_DEBUG({
     llvm::dbgs() << "Tool invocation for module: "
                  << getOperation().getNameAttr() << "\n";
+    llvm::dbgs() << "ptxas executable:" << ptxasCompiler.value() << "\n";
     llvm::interleave(ptxasArgs, llvm::dbgs(), " ");
     llvm::dbgs() << "\n";
     if (createFatbin) {
diff --git a/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir b/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir
index 34dde6e03c80e..f65eda6a2a88d 100644
--- a/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir
+++ b/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir
@@ -16,6 +16,7 @@ func.func @host_function(%arg0 : f32, %arg1 : memref<?xf32>) {
     return
 }
 
+// CHECK: ptxas executable:
 // CHECK: ptxas -arch sm_80
 // CHECK-SAME: -v 
 // CHECK-SAME: --register-usage-level=8

[llvmbot]

@llvm/pr-subscribers-mlir

Author: Guray Ozen (grypp)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/132373.diff

2 Files Affected:

• (modified) mlir/lib/Target/LLVM/NVVM/Target.cpp (+1)
• (modified) mlir/test/Integration/GPU/CUDA/command-line-arg.mlir (+1)

diff --git a/mlir/lib/Target/LLVM/NVVM/Target.cpp b/mlir/lib/Target/LLVM/NVVM/Target.cpp
index fa8c597da58b1..69a247b06bf16 100644
--- a/mlir/lib/Target/LLVM/NVVM/Target.cpp
+++ b/mlir/lib/Target/LLVM/NVVM/Target.cpp
@@ -455,6 +455,7 @@ NVPTXSerializer::compileToBinary(const std::string &ptxCode) {
   LLVM_DEBUG({
     llvm::dbgs() << "Tool invocation for module: "
                  << getOperation().getNameAttr() << "\n";
+    llvm::dbgs() << "ptxas executable:" << ptxasCompiler.value() << "\n";
     llvm::interleave(ptxasArgs, llvm::dbgs(), " ");
     llvm::dbgs() << "\n";
     if (createFatbin) {
diff --git a/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir b/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir
index 34dde6e03c80e..f65eda6a2a88d 100644
--- a/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir
+++ b/mlir/test/Integration/GPU/CUDA/command-line-arg.mlir
@@ -16,6 +16,7 @@ func.func @host_function(%arg0 : f32, %arg1 : memref<?xf32>) {
     return
 }
 
+// CHECK: ptxas executable:
 // CHECK: ptxas -arch sm_80
 // CHECK-SAME: -v 
 // CHECK-SAME: --register-usage-level=8

[joker-eph]

I'm puzzled that we have this test in-tree, in general we never test debug output, and I don't understand how is this test passing in release mode?

[grypp]

I add a test for every code I write, so I probably added these test automatically :)

This is a CUDA integration test. Perhaps we don't run CUDA integration tests in release mode?

[joker-eph]

It's possible, can you remove it please?

[grypp]

I removed from this PR.
I have written other tests as well for debug runs. let me put a new PR that removes them all.