Add VectorToSPIRV patterns to GPUToSPIRVPass, and fix errors in e2e test

Author: angelz913

mlir/lib/Conversion/GPUToSPIRV/GPUToSPIRVPass.cpp

@@ -16,6 +16,7 @@
 #include "mlir/Conversion/ArithToSPIRV/ArithToSPIRV.h"
 #include "mlir/Conversion/FuncToSPIRV/FuncToSPIRV.h"
 #include "mlir/Conversion/GPUToSPIRV/GPUToSPIRV.h"
+#include "mlir/Conversion/VectorToSPIRV/VectorToSPIRV.h"
 #include "mlir/Conversion/MemRefToSPIRV/MemRefToSPIRV.h"
 #include "mlir/Conversion/SCFToSPIRV/SCFToSPIRV.h"
 #include "mlir/Dialect/Func/IR/FuncOps.h"
@@ -132,6 +133,7 @@ void GPUToSPIRVPass::runOnOperation() {
     mlir::arith::populateArithToSPIRVPatterns(typeConverter, patterns);
     populateMemRefToSPIRVPatterns(typeConverter, patterns);
     populateFuncToSPIRVPatterns(typeConverter, patterns);
+    populateVectorToSPIRVPatterns(typeConverter, patterns);
 
     if (failed(applyFullConversion(gpuModule, *target, std::move(patterns))))
       return signalPassFailure();

mlir/test/mlir-vulkan-runner/vector-interleave.mlir

@@ -9,33 +9,41 @@ module attributes {
     #spirv.vce<v1.0, [Shader], [SPV_KHR_storage_buffer_storage_class]>, #spirv.resource_limits<>>
 } {
   gpu.module @kernels {
-    gpu.func @kernel_vector_interleave(%arg0 : vector<2xi32>, %arg1 : vector<2xi32>, %arg2 : memref<4xi32>)
+    gpu.func @kernel_vector_interleave(%arg0 : memref<2xi32>, %arg1 : memref<2xi32>, %arg2 : memref<4xi32>)
       kernel attributes { spirv.entry_point_abi = #spirv.entry_point_abi<workgroup_size = [1, 1, 1]>} {
       %c0 = arith.constant 0 : index
-      %result = vector.interleave %arg0, %arg1 : vector<2xi32>
+      %vec0 = vector.load %arg0[%c0] : memref<2xi32>, vector<2xi32>
+      %vec1 = vector.load %arg1[%c0] : memref<2xi32>, vector<2xi32>
+      %result = vector.interleave %vec0, %vec1 : vector<2xi32> -> vector<4xi32>
       vector.store %result, %arg2[%c0] : memref<4xi32>, vector<4xi32>
       gpu.return
     }
   }
 
   func.func @main() {
     // Allocate 3 buffers.
-    %buf0 = arith.constant dense<[0, 1]> : vector<2xi32>
-    %buf1 = arith.constant dense<[2, 3]> : vector<2xi32>
+    %buf0 = memref.alloc() : memref<2xi32>
+    %buf1 = memref.alloc() : memref<2xi32>
     %buf2 = memref.alloc() : memref<4xi32>
 
     %idx0 = arith.constant 0 : index
     %idx1 = arith.constant 1 : index
     %idx4 = arith.constant 4 : index
 
+    // Initialize input buffer
+    %buf0_vals = arith.constant dense<[0, 1]> : vector<2xi32>
+    %buf1_vals = arith.constant dense<[2, 3]> : vector<2xi32>
+    vector.store %buf0_vals, %buf0[%idx0] : memref<2xi32>, vector<2xi32>
+    vector.store %buf1_vals, %buf1[%idx0] : memref<2xi32>, vector<2xi32>
+
     // Initialize output buffer.
     %value0 = arith.constant 0 : i32
     %buf3 = memref.cast %buf2 : memref<4xi32> to memref<?xi32>
     call @fillResource1DInt(%buf3, %value0) : (memref<?xi32>, i32) -> ()
 
     gpu.launch_func @kernels::@kernel_vector_interleave
         blocks in (%idx4, %idx1, %idx1) threads in (%idx1, %idx1, %idx1)
-        args(%buf0 : vector<2xi32>, %buf1 : vector<2xi32>, %buf2 : memref<4xi32>)
+        args(%buf0 : memref<2xi32>, %buf1 : memref<2xi32>, %buf2 : memref<4xi32>)
     %buf4 = memref.cast %buf3 : memref<?xi32> to memref<*xi32>
     call @printMemrefI32(%buf4) : (memref<*xi32>) -> ()
     return