[Mlir-commits] [mlir] [mlir][spir] Add floating point dot product (PR #73466)

[Ivan Butygin]

================
@@ -760,13 +791,14 @@ void mlir::populateVectorToSPIRVPatterns(SPIRVTypeConverter &typeConverter,
       VectorInsertOpConvert, VectorReductionPattern<GL_INT_MAX_MIN_OPS>,
       VectorReductionPattern<CL_INT_MAX_MIN_OPS>,
       VectorReductionFloatMinMax<CL_FLOAT_MAX_MIN_OPS>,
-      VectorReductionFloatMinMax<GL_FLOAT_MAX_MIN_OPS>, VectorShapeCast,
+      VectorReductionFloatMinMax<GL_FLOAT_MAX_MIN_OPS>,
+      VectorReductionToFPDotProd, VectorShapeCast,
----------------
Hardcode84 wrote:

I see there is already `VectorReductionPattern` patter in the list which can convert same op. Should this new pattern have higher benefit?

https://github.com/llvm/llvm-project/pull/73466