remove index ops

Author: charithaintc

mlir/test/Dialect/XeGPU/subgroup-distribution.mlir

@@ -226,52 +226,3 @@ gpu.func @prefetch_1d(%arg0: memref<256xf16>){
   gpu.return
 }
 }
-
-
-// -----
-// CHECK-LABEL: gpu.func @gemm_loop
-// CHECK: (%[[ARG0:[0-9a-zA-Z]+]]: memref<1024x1024xbf16>, %[[ARG1:[0-9a-zA-Z]+]]: memref<1024x1024xbf16>, %[[ARG2:[0-9a-zA-Z]+]]: memref<1024x1024xf32>) {
-// CHECK: %[[BLOCK_ID_Y:.*]] = gpu.block_id y
-// CHECK: %[[Y_COORD:.*]] = arith.muli %[[BLOCK_ID_Y]], %c16 : index
-// CHECK: %[[BLOCK_ID_X:.*]] = gpu.block_id x
-// CHECK: %[[X_COORD:.*]] = arith.muli %[[BLOCK_ID_X]], %c8 : index
-// CHECK: %[[T2:.*]] = xegpu.create_nd_tdesc %[[ARG2]][%[[X_COORD]], %[[Y_COORD]]] : memref<1024x1024xf32> -> !xegpu.tensor_desc<8x16xf32>
-// CHECK: %[[T3:.*]] = xegpu.load_nd %[[T2]] : !xegpu.tensor_desc<8x16xf32> -> vector<8xf32>
-// CHECK: %[[T4:.*]] = vector.shape_cast %[[T3]] : vector<8xf32> to vector<8x1xf32>
-// CHECK: %[[T5:.*]] = scf.for %[[K:.*]] = %{{.*}} to %{{.*}} step %{{.*}} iter_args(%[[ARG4:.*]] = %[[T4]]) -> (vector<8x1xf32>) {
-// CHECK: %[[T10:.*]] = xegpu.create_nd_tdesc %[[ARG1]][%[[K]], %[[Y_COORD]]] : memref<1024x1024xbf16> -> !xegpu.tensor_desc<16x16xbf16>
-// CHECK: %[[T11:.*]] = xegpu.load_nd %[[T10]] <{packed}> : !xegpu.tensor_desc<16x16xbf16> -> vector<16xbf16>
-// CHECK: %[[T12:.*]] = xegpu.create_nd_tdesc %[[ARG0]][%[[X_COORD]], %[[K]]] : memref<1024x1024xbf16> -> !xegpu.tensor_desc<8x16xbf16>
-// CHECK: %[[T13:.*]] = xegpu.load_nd %[[T12]] : !xegpu.tensor_desc<8x16xbf16> -> vector<8xbf16>
-// CHECK: %[[T14:.*]] = vector.shape_cast %[[ARG4]] : vector<8x1xf32> to vector<8xf32>
-// CHECK: %[[T15:.*]] = xegpu.dpas %[[T13]], %[[T11]], %[[T14]] : vector<8xbf16>, vector<16xbf16>, vector<8xf32> -> vector<8xf32>
-// CHECK: %[[T16:.*]] = vector.shape_cast %[[T15]] : vector<8xf32> to vector<8x1xf32>
-// CHECK: scf.yield %[[T16]] : vector<8x1xf32>
-// CHECK: }
-// CHECK: %[[T8:.*]] = xegpu.create_nd_tdesc %[[ARG2]]{{.*}} : memref<1024x1024xf32> -> !xegpu.tensor_desc<8x16xf32>
-// CHECK: %[[T9:.*]] = vector.shape_cast %[[T5]] : vector<8x1xf32> to vector<8xf32>
-// CHECK: xegpu.store_nd %[[T9]], %[[T8]] : vector<8xf32>, !xegpu.tensor_desc<8x16xf32>
-gpu.module @test {
-gpu.func @gemm_loop(%arg0: memref<1024x1024xbf16>, %arg1: memref<1024x1024xbf16>, %arg2: memref<1024x1024xf32>){
-  %c0 = arith.constant 0 : index
-  %c16 = arith.constant 16 : index
-  %c8 = arith.constant 8 : index
-  %c1024 = arith.constant 1024 : index
-  %0 = gpu.block_id x
-  %1 = gpu.block_id y
-  %2 = arith.muli %0, %c8 : index
-  %3 = arith.muli %1, %c16 : index
-  %4 = xegpu.create_nd_tdesc %arg2[%2, %3] : memref<1024x1024xf32> -> !xegpu.tensor_desc<8x16xf32>
-  %5 = xegpu.load_nd %4 : !xegpu.tensor_desc<8x16xf32> -> vector<8x16xf32>
-  %6 = scf.for %arg3 = %c0 to %c1024 step %c16 iter_args(%arg4 = %5) -> (vector<8x16xf32>) {
-    %7 = xegpu.create_nd_tdesc %arg0[%2, %arg3] : memref<1024x1024xbf16> -> !xegpu.tensor_desc<8x16xbf16>
-    %8 = xegpu.create_nd_tdesc %arg1[%arg3, %3] : memref<1024x1024xbf16> -> !xegpu.tensor_desc<16x16xbf16>
-    %9 = xegpu.load_nd %7 : !xegpu.tensor_desc<8x16xbf16> -> vector<8x16xbf16>
-    %10 = xegpu.load_nd %8 : !xegpu.tensor_desc<16x16xbf16> -> vector<16x16xbf16>
-    %11 = xegpu.dpas %9, %10, %arg4 : vector<8x16xbf16>, vector<16x16xbf16>, vector<8x16xf32> -> vector<8x16xf32>
-    scf.yield %11 : vector<8x16xf32>
-  }
-  xegpu.store_nd %6, %4 : vector<8x16xf32>, !xegpu.tensor_desc<8x16xf32>
-  gpu.return
-}
-}