[mlir][vector] Use vector.broadcast in place of vector.splat (#148028)

Part of deprecation of vector.splat

RFC:
https://discourse.llvm.org/t/rfc-mlir-vector-deprecate-then-remove-vector-splat/87143/4
More complete deprecation:
#147818

Author: newling

mlir/lib/Dialect/Arith/Transforms/EmulateUnsupportedFloats.cpp

@@ -123,7 +123,7 @@ void mlir::arith::populateEmulateUnsupportedFloatsLegality(
       vector::FMAOp, vector::OuterProductOp, vector::MatmulOp, vector::ScanOp>(
       [&](Operation *op) { return converter.isLegal(op); });
   target.addLegalOp<arith::BitcastOp, arith::ExtFOp, arith::TruncFOp,
-                    arith::ConstantOp, vector::SplatOp>();
+                    arith::ConstantOp, vector::SplatOp, vector::BroadcastOp>();
 }
 
 void EmulateUnsupportedFloatsPass::runOnOperation() {

mlir/lib/Dialect/NVGPU/TransformOps/NVGPUTransformOps.cpp

@@ -688,7 +688,7 @@ Value MmaSyncBuilder::buildMmaSyncMemRefLoadOperand(
 
   Type elementType = getElementTypeOrSelf(memref.getType());
   auto vt = VectorType::get(vectorShape, elementType);
-  Value res = b.create<vector::SplatOp>(loc, vt, loads[0]);
+  Value res = b.create<vector::BroadcastOp>(loc, vt, loads[0]);
   foreachIndividualVectorElement(
       res,
       /*applyFn=*/

mlir/lib/Dialect/XeGPU/Transforms/XeGPUUnroll.cpp

@@ -438,7 +438,8 @@ struct UnrollCreateDescOp : public UnrollPattern<xegpu::CreateDescOp> {
           // Compute the offset
           Value inc = rewriter.create<arith::ConstantIndexOp>(
               loc, i * blockedChunkSize);
-          Value incVec = rewriter.create<vector::SplatOp>(loc, indiceType, inc);
+          Value incVec =
+              rewriter.create<vector::BroadcastOp>(loc, indiceType, inc);
           Value offsetIndice =
               rewriter.create<arith::AddIOp>(loc, indice, incVec);
 

mlir/test/Dialect/NVGPU/transform-matmul-to-nvvm.mlir

@@ -20,14 +20,14 @@ func.func @matmul_16x8x4xf32_global(
 // CHECK:           %[[VAL_7:.*]] = affine.apply #[[$div4p8]]()[%[[TIDX]]]
 // CHECK:           %[[VAL_8:.*]] = affine.apply #[[$mod4]]()[%[[TIDX]]]
 // CHECK:           %[[VAL_9:.*]] = memref.load %[[VAL_0]][%[[VAL_7]], %[[VAL_8]]] : memref<16x4xf32>
-// CHECK:           %[[VAL_10:.*]] = vector.splat %[[VAL_6]] : vector<2x1xf32>
+// CHECK:           %[[VAL_10:.*]] = vector.broadcast %[[VAL_6]] : f32 to vector<2x1xf32>
 // CHECK:           %[[VAL_11:.*]] = vector.insert %[[VAL_6]], %[[VAL_10]] [0, 0] : f32 into vector<2x1xf32>
 // CHECK:           %[[LHS:.*]] = vector.insert %[[VAL_9]], %[[VAL_11]] [1, 0] : f32 into vector<2x1xf32>
 //
 // CHECK:           %[[VAL_13:.*]] = affine.apply #[[$mod4]]()[%[[TIDX]]]
 // CHECK:           %[[VAL_14:.*]] = affine.apply #[[$div4]]()[%[[TIDX]]]
 // CHECK:           %[[VAL_15:.*]] = memref.load %[[VAL_1]][%[[VAL_13]], %[[VAL_14]]] : memref<4x8xf32>
-// CHECK:           %[[VAL_16:.*]] = vector.splat %[[VAL_15]] : vector<1x1xf32>
+// CHECK:           %[[VAL_16:.*]] = vector.broadcast %[[VAL_15]] : f32 to vector<1x1xf32>
 // CHECK:           %[[RHS:.*]] = vector.insert %[[VAL_15]], %[[VAL_16]] [0, 0] : f32 into vector<1x1xf32>
 //
 // CHECK:           %[[VAL_18:.*]] = affine.apply #[[$div4]]()[%[[TIDX]]]
@@ -42,7 +42,7 @@ func.func @matmul_16x8x4xf32_global(
 // CHECK:           %[[VAL_27:.*]] = affine.apply #[[$div4p8]]()[%[[TIDX]]]
 // CHECK:           %[[VAL_28:.*]] = affine.apply #[[$map4]]()[%[[TIDX]]]
 // CHECK:           %[[VAL_29:.*]] = memref.load %[[VAL_2]][%[[VAL_27]], %[[VAL_28]]] : memref<16x8xf32>
-// CHECK:           %[[VAL_30:.*]] = vector.splat %[[VAL_20]] : vector<2x2xf32>
+// CHECK:           %[[VAL_30:.*]] = vector.broadcast %[[VAL_20]] : f32 to vector<2x2xf32>
 // CHECK:           %[[VAL_31:.*]] = vector.insert %[[VAL_20]], %[[VAL_30]] [0, 0] : f32 into vector<2x2xf32>
 // CHECK:           %[[VAL_32:.*]] = vector.insert %[[VAL_23]], %[[VAL_31]] [0, 1] : f32 into vector<2x2xf32>
 // CHECK:           %[[VAL_33:.*]] = vector.insert %[[VAL_26]], %[[VAL_32]] [1, 0] : f32 into vector<2x2xf32>