[mlir][sparse][gpu] reduce boilerplate class declarations

[aartbik]

Summary

Macro is used to avoid repeating same pattern many times.
Also fixed the ordering of ops to be consistent.

mlir/lib/Conversion/GPUCommon/GPUToLLVMConversion.cpp

ConvertSetDefaultDeviceOpToGpuRuntimeCallPattern(
       : ConvertOpToGpuRuntimeCallPattern<gpu::SetDefaultDeviceOp>(
             typeConverter) {}
 
   LogicalResult
   matchAndRewrite(gpu::SetDefaultDeviceOp op, OpAdaptor adaptor,
                   ConversionPatternRewriter &rewriter) const override;
 };
 
-class ConvertCreateDnTensorOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::CreateDnTensorOp> {
-public:
-  ConvertCreateDnTensorOpToGpuRuntimeCallPattern(
-      LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::CreateDnTensorOp>(typeConverter) {
-  }
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::CreateDnTensorOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertDestroyDnTensorOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::DestroyDnTensorOp> {
-public:
-  ConvertDestroyDnTensorOpToGpuRuntimeCallPattern(
-      LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::DestroyDnTensorOp>(
-            typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::DestroyDnTensorOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertCreateCooOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::CreateCooOp> {
-public:
-  ConvertCreateCooOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::CreateCooOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::CreateCooOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertCreateCooAoSOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::CreateCooAoSOp> {
-public:
-  ConvertCreateCooAoSOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::CreateCooAoSOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::CreateCooAoSOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertCreateCsrOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::CreateCsrOp> {
-public:
-  ConvertCreateCsrOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::CreateCsrOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::CreateCsrOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertCreate2To4SpMatOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::Create2To4SpMatOp> {
-public:
-  ConvertCreate2To4SpMatOpToGpuRuntimeCallPattern(
-      LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::Create2To4SpMatOp>(
-            typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::Create2To4SpMatOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertDestroySpMatOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::DestroySpMatOp> {
-public:
-  ConvertDestroySpMatOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::DestroySpMatOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::DestroySpMatOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertSpMVBufferSizeOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::SpMVBufferSizeOp> {
-public:
-  ConvertSpMVBufferSizeOpToGpuRuntimeCallPattern(
-      LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::SpMVBufferSizeOp>(typeConverter) {
-  }
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::SpMVBufferSizeOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertSpMVOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::SpMVOp> {
-public:
-  ConvertSpMVOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::SpMVOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::SpMVOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertSpMMBufferSizeOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::SpMMBufferSizeOp> {
-public:
-  ConvertSpMMBufferSizeOpToGpuRuntimeCallPattern(
-      LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::SpMMBufferSizeOp>(typeConverter) {
-  }
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::SpMMBufferSizeOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertSDDMMBufferSizeOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::SDDMMBufferSizeOp> {
-public:
-  ConvertSDDMMBufferSizeOpToGpuRuntimeCallPattern(
-      LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::SDDMMBufferSizeOp>(
-            typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::SDDMMBufferSizeOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertSpMMOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::SpMMOp> {
-public:
-  ConvertSpMMOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::SpMMOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::SpMMOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-class ConvertSDDMMOpToGpuRuntimeCallPattern
-    : public ConvertOpToGpuRuntimeCallPattern<gpu::SDDMMOp> {
-public:
-  ConvertSDDMMOpToGpuRuntimeCallPattern(LLVMTypeConverter &typeConverter)
-      : ConvertOpToGpuRuntimeCallPattern<gpu::SDDMMOp>(typeConverter) {}
-
-private:
-  LogicalResult
-  matchAndRewrite(gpu::SDDMMOp op, OpAdaptor adaptor,
-                  ConversionPatternRewriter &rewriter) const override;
-};
-
-// TODO: Apply this pattern to all GPU ops.
+/// Generic rewriting rule for operation on sparse matrices.
+/// Currently supports CUDA (by means of cuSparse and cuSparseLt).
 #define DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(op_name)                \
   class Convert##op_name##ToGpuRuntimeCallPattern                              \
       : public ConvertOpToGpuRuntimeCallPattern<gpu::op_name> {                \
   public:                                                                      \
     Convert##op_name##ToGpuRuntimeCallPattern(                                 \
         LLVMTypeConverter &typeConverter)                                      \
         : ConvertOpToGpuRuntimeCallPattern<gpu::op_name>(typeConverter) {}     \
                                                                                \
   private:                                                                     \
     LogicalResult                                                              \
     matchAndRewrite(gpu::op_name op, OpAdaptor adaptor,                        \
                     ConversionPatternRewriter &rewriter) const override;       \
   };
 
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(CreateDnTensorOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(DestroyDnTensorOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(CreateCooOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(CreateCooAoSOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(CreateCsrOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(Create2To4SpMatOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(DestroySpMatOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpMVBufferSizeOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpMVOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpMMBufferSizeOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SDDMMBufferSizeOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpMMOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SDDMMOp)
 DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMCreateDescrOp)
 DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMDestroyDescrOp)
 DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMWorkEstimationOrComputeOp)
 DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMEstimateMemoryOp)
-DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMGetSizeOp)
 DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMCopyOp)
+DECLARE_CONVERT_OP_TO_GPU_RUNTIME_CALL_PATTERN(SpGEMMGetSizeOp)
 
 } // namespace
 
 void GpuToLLVMConversionPass::runOnOperation() {
   LowerToLLVMOptions options(&getContext());
   options.useOpaquePointers = useOpaquePointers;
   options.useBarePtrCallConv = hostBarePtrCallConv;
 
patterns.add<ConvertAllocOpToGpuRuntimeCallPattern,
                ConvertAsyncYieldToGpuRuntimeCallPattern,
                ConvertCreateDnTensorOpToGpuRuntimeCallPattern,
                ConvertDestroyDnTensorOpToGpuRuntimeCallPattern,
                ConvertCreateCooOpToGpuRuntimeCallPattern,
                ConvertCreateCooAoSOpToGpuRuntimeCallPattern,
                ConvertCreateCsrOpToGpuRuntimeCallPattern,
                ConvertCreate2To4SpMatOpToGpuRuntimeCallPattern,
                ConvertDestroySpMatOpToGpuRuntimeCallPattern,
+               ConvertSpMVBufferSizeOpToGpuRuntimeCallPattern,
+               ConvertSpMVOpToGpuRuntimeCallPattern,
+               ConvertSpMMBufferSizeOpToGpuRuntimeCallPattern,
+               ConvertSDDMMBufferSizeOpToGpuRuntimeCallPattern,
+               ConvertSpMMOpToGpuRuntimeCallPattern,
+               ConvertSDDMMOpToGpuRuntimeCallPattern,
                ConvertSpGEMMCreateDescrOpToGpuRuntimeCallPattern,
                ConvertSpGEMMDestroyDescrOpToGpuRuntimeCallPattern,
                ConvertSpGEMMWorkEstimationOrComputeOpToGpuRuntimeCallPattern,
                ConvertSpGEMMEstimateMemoryOpToGpuRuntimeCallPattern,
-               ConvertSpGEMMGetSizeOpToGpuRuntimeCallPattern,
                ConvertSpGEMMCopyOpToGpuRuntimeCallPattern,
-               ConvertSpMVBufferSizeOpToGpuRuntimeCallPattern,
-               ConvertSpMVOpToGpuRuntimeCallPattern,
-               ConvertSpMMBufferSizeOpToGpuRuntimeCallPattern,
-               ConvertSpMMOpToGpuRuntimeCallPattern,
-               ConvertSDDMMBufferSizeOpToGpuRuntimeCallPattern,
-               ConvertSDDMMOpToGpuRuntimeCallPattern>(converter);
+               ConvertSpGEMMGetSizeOpToGpuRuntimeCallPattern>(converter);
   patterns.add<ConvertLaunchFuncOpToGpuRuntimeCallPattern>(
       converter, gpuBinaryAnnotation, kernelBarePtrCallConv);
   patterns.add<EraseGpuModuleOpPattern>(&converter.getContext());
 }