57 lines · plain
1// RUN: mlir-opt %s --split-input-file -convert-amdgpu-to-rocdl=chipset=gfx950 | FileCheck %s2// RUN: not mlir-opt %s --split-input-file -convert-amdgpu-to-rocdl=chipset=gfx942 2>&1 | FileCheck %s --check-prefix=CHECK-OLD 3 4// CHECK-LABEL: func @transpose_load_to_rocdl_4xf165func.func @transpose_load_to_rocdl_4xf16(%idx1 : index, %idx2 : index, %wgmem : memref<128x72xf16, 3>) -> vector<4xf16> {6 // CHECK: rocdl.ds.read.tr16.b647 // CHECK-OLD: error: 'amdgpu.transpose_load' op Non-gfx950 chipset not supported8 %0 = amdgpu.transpose_load %wgmem[%idx1, %idx2] : memref<128x72xf16, 3> -> vector<4xf16>9 return %0 : vector<4xf16>10}11 12// -----13 14// CHECK-LABEL: func @transpose_load_to_rocdl_8xi815func.func @transpose_load_to_rocdl_8xi8(%idx1 : index, %idx2 : index, %wgmem : memref<128x128xi8, 3>) -> vector<8xi8> {16 // CHECK: %[[RES:.*]] = rocdl.ds.read.tr8.b6417 // CHECK-SAME: -> vector<2xi32>18 // CHECK-NEXT: llvm.bitcast %[[RES]] : vector<2xi32> to vector<8xi8>19 // CHECK-OLD: error: 'amdgpu.transpose_load' op Non-gfx950 chipset not supported20 %0 = amdgpu.transpose_load %wgmem[%idx1, %idx2] : memref<128x128xi8, 3> -> vector<8xi8>21 return %0 : vector<8xi8>22}23 24// -----25 26// CHECK-LABEL: func @transpose_load_to_rocdl_i4_memrefxi827func.func @transpose_load_to_rocdl_i4_memrefxi8(%idx1 : index, %idx2 : index, %wgmem : memref<128x32xi8, 3>) -> vector<16xi4> {28 // CHECK: %[[RES:.*]] = rocdl.ds.read.tr4.b6429 // CHECK-SAME: -> vector<2xi32>30 // CHECK-NEXT: llvm.bitcast %[[RES]] : vector<2xi32> to vector<16xi4>31 // CHECK-OLD: error: 'amdgpu.transpose_load' op Non-gfx950 chipset not supported32 %0 = amdgpu.transpose_load %wgmem[%idx1, %idx2] : memref<128x32xi8, 3> -> vector<16xi4>33 return %0 : vector<16xi4>34}35 36// -----37 38// CHECK-LABEL: func @transpose_load_to_rocdl_i6_memrefxi839func.func @transpose_load_to_rocdl_i6_memrefxi8(%idx1 : index, %idx2 : index, %wgmem : memref<128x32xi8, 3>) -> vector<16xi6> {40 // CHECK: %[[RES:.*]] = rocdl.ds.read.tr6.b9641 // CHECK-SAME: -> vector<3xi32>42 // CHECK-NEXT: llvm.bitcast %[[RES]] : vector<3xi32> to vector<16xi6>43 // CHECK-OLD: error: 'amdgpu.transpose_load' op Non-gfx950 chipset not supported44 %0 = amdgpu.transpose_load %wgmem[%idx1, %idx2] : memref<128x32xi8, 3> -> vector<16xi6>45 return %0 : vector<16xi6>46}47 48// -----49 50// CHECK-LABEL: func @transpose_load_to_rocdl_i16_memrefxi851func.func @transpose_load_to_rocdl_i16_memrefxi8(%idx1 : index, %idx2 : index, %wgmem : memref<128x32xi8, 3>) -> vector<4xi16> {52 // CHECK: rocdl.ds.read.tr16.b6453 // CHECK-OLD: error: 'amdgpu.transpose_load' op Non-gfx950 chipset not supported54 %0 = amdgpu.transpose_load %wgmem[%idx1, %idx2] : memref<128x32xi8, 3> -> vector<4xi16>55 return %0 : vector<4xi16>56}57