rename CompileToPtx to CompileToTargetBinary

PiperOrigin-RevId: 937650110

Author: ermilovmaxim

third_party/xla/xla/backends/gpu/codegen/cubin_custom_kernel_compiler.cc

@@ -87,20 +87,21 @@ xla::Future<LlvmKernelSource> CubinCustomKernelCompiler::CompileMlirToLlvm(
       });
 }
 
-xla::Future<std::vector<uint8_t>> CubinCustomKernelCompiler::CompileToPtx(
+xla::Future<std::vector<uint8_t>>
+CubinCustomKernelCompiler::CompileToTargetBinary(
     LlvmKernelSource kernel_source) {
   if (!thread_pool_) {
-    return CompileToPtxImpl(std::move(kernel_source));
+    return CompileToCubinImpl(std::move(kernel_source));
   }
   return xla::MakeFutureOn(
       *thread_pool_->AsExecutor(),
       [this, kernel_source = std::move(kernel_source)]() mutable {
-        return CompileToPtxImpl(std::move(kernel_source));
+        return CompileToCubinImpl(std::move(kernel_source));
       });
 }
 
 absl::StatusOr<std::vector<uint8_t>>
-CubinCustomKernelCompiler::CompileToPtxImpl(LlvmKernelSource kernel_source) {
+CubinCustomKernelCompiler::CompileToCubinImpl(LlvmKernelSource kernel_source) {
   llvm::orc::ThreadSafeModule thread_safe_module =
       std::move(kernel_source).thread_safe_module();
   llvm::Module* llvm_module = thread_safe_module.getModuleUnlocked();
@@ -120,7 +121,7 @@ absl::StatusOr<std::unique_ptr<Thunk>> CubinCustomKernelCompiler::CompileImpl(
     const emitters::KernelArguments& kernel_arguments,
     const LaunchDimensions& launch_dimensions) {
   ASSIGN_OR_RETURN(std::vector<uint8_t> cubin,
-                   CompileToPtxImpl(std::move(kernel_source)));
+                   CompileToCubinImpl(std::move(kernel_source)));
 
   ASSIGN_OR_RETURN(
       CustomKernel custom_kernel,

third_party/xla/xla/backends/gpu/codegen/cubin_custom_kernel_compiler.h

@@ -80,7 +80,7 @@ class CubinCustomKernelCompiler final : public KernelCompiler {
       const std::string& entry_function_name, int unroll_factor,
       MlirKernelSource source, BorrowedMlirContext borrowed_context) override;
 
-  xla::Future<std::vector<uint8_t>> CompileToPtx(
+  xla::Future<std::vector<uint8_t>> CompileToTargetBinary(
       LlvmKernelSource kernel_source) override;
 
   xla::Future<TritonWrapperResult> CompileTritonToLlvm(
@@ -92,7 +92,7 @@ class CubinCustomKernelCompiler final : public KernelCompiler {
       bool is_xla_fusion) override;
 
  private:
-  absl::StatusOr<std::vector<uint8_t>> CompileToPtxImpl(
+  absl::StatusOr<std::vector<uint8_t>> CompileToCubinImpl(
       LlvmKernelSource kernel_source);
 
   absl::StatusOr<std::unique_ptr<Thunk>> CompileImpl(

third_party/xla/xla/backends/gpu/codegen/emitters/mlir_kernel_emitter.cc

@@ -393,7 +393,7 @@ AsyncThunkSequence MlirKernelFusion::Emit(
                                        ir_emitter_context.gpu_device_info());
 
               return ir_emitter_context.kernel_compiler()
-                  ->CompileToPtx(std::move(kernel_def).TakeSource())
+                  ->CompileToTargetBinary(std::move(kernel_def).TakeSource())
                   .Map([kernel_name = std::move(kernel_name),
                         launch_dims = std::move(launch_dims),
                         use_pdl](const std::vector<uint8_t>& cubin) {

third_party/xla/xla/backends/gpu/codegen/kernel_compiler.h

@@ -90,7 +90,7 @@ class KernelCompiler {
       TritonKernelSource triton_source, BorrowedMlirContext borrowed_context,
       bool is_xla_fusion) = 0;
 
-  virtual xla::Future<std::vector<uint8_t>> CompileToPtx(
+  virtual xla::Future<std::vector<uint8_t>> CompileToTargetBinary(
       LlvmKernelSource kernel_source) = 0;
 
   // Sets a callback to be called prior to llvm::Module compilation.

third_party/xla/xla/backends/gpu/codegen/triton/fusion.cc

@@ -236,7 +236,7 @@ xla::Future<TritonFusion::EmitResult> TritonFusion::Emit(
               local_module.getModuleUnlocked()));
 
           return kernel_compiler
-              ->CompileToPtx(LlvmKernelSource{std::move(local_module)})
+              ->CompileToTargetBinary(LlvmKernelSource{std::move(local_module)})
               .Map([kernel_name = sanitized_kernel_name,
                     launch_dims = std::move(launch_dimensions),
                     tma_metadata = triton_wrapper_result.tma_metadata,

third_party/xla/xla/service/gpu/compile_module_to_llvm_ir.cc

@@ -267,7 +267,8 @@ absl::StatusOr<CompileModuleResults> CompileModuleToLlvmIr(
 
   ASSIGN_OR_RETURN(
       results.constants_binary,
-      compiler->CompileToPtx(thunk_emitter.ConsumeConstantsModule()).Await());
+      compiler->CompileToTargetBinary(thunk_emitter.ConsumeConstantsModule())
+          .Await());
   ASSIGN_OR_RETURN(results.executable,
                    std::move(future_sequential_thunk).Await());
 

third_party/xla/xla/service/gpu/thunk_emitter.cc

@@ -1329,7 +1329,7 @@ AsyncThunkSequence ThunkEmitter::EmitTritonCustomCall(
               local_module.getModuleUnlocked()));
 
           return kernel_compiler
-              ->CompileToPtx(LlvmKernelSource{std::move(local_module)})
+              ->CompileToTargetBinary(LlvmKernelSource{std::move(local_module)})
               .Map([use_pdl = result.use_pdl, shmem_bytes = result.shmem_bytes,
                     launch_dimensions = std::move(launch_dimensions),
                     tma_metadata = result.tma_metadata,