@@ -112,21 +112,21 @@ gpu.func @prefetch_nd(%src: memref<24x32xf16>) {
112112 gpu.return
113113}
114114
115- // CHECK: gpu.func @prefetch_nd_2(%[[arg0:.*]]: memref<8x24x32x48x64xf16 >) {
116- gpu.func @prefetch_nd_2 (%src: memref <8 x 24 x 32 x 48 x 64 x f16 >) {
117- // CHECK: %[[R0:.*]] = xegpu.create_nd_tdesc %[[arg0]][0, 0, 0, 0, 0 ] : memref<8x24x32x48x64xf16 > -> !xegpu.tensor_desc<1x2x4x8x16xf16 >
118- %1 = xegpu.create_nd_tdesc %src [0 , 0 , 0 , 0 , 0 ] : memref <8 x 24 x 32 x 48 x 64 x f16 > -> !xegpu.tensor_desc <1 x 2 x 4 x 8 x 16 x f16 >
119- // CHECK: xegpu.prefetch_nd %[[R0]] <{l1_hint = #xegpu.cache_hint<cached>, l2_hint = #xegpu.cache_hint<uncached>}> : !xegpu.tensor_desc<1x2x4x8x16xf16 >
120- xegpu.prefetch_nd %1 <{l1_hint = #xegpu.cache_hint <cached >, l2_hint = #xegpu.cache_hint <uncached >}>: !xegpu.tensor_desc <1 x 2 x 4 x 8 x 16 x f16 >
115+ // CHECK: gpu.func @prefetch_nd_2(%[[arg0:.*]]: memref<48x64xf16 >) {
116+ gpu.func @prefetch_nd_2 (%src: memref <48 x 64 x f16 >) {
117+ // CHECK: %[[R0:.*]] = xegpu.create_nd_tdesc %[[arg0]][0, 0] : memref<48x64xf16 > -> !xegpu.tensor_desc<8x16xf16 >
118+ %1 = xegpu.create_nd_tdesc %src [0 , 0 ] : memref <48 x 64 x f16 > -> !xegpu.tensor_desc <8 x 16 x f16 >
119+ // CHECK: xegpu.prefetch_nd %[[R0]] <{l1_hint = #xegpu.cache_hint<cached>, l2_hint = #xegpu.cache_hint<uncached>}> : !xegpu.tensor_desc<8x16xf16 >
120+ xegpu.prefetch_nd %1 <{l1_hint = #xegpu.cache_hint <cached >, l2_hint = #xegpu.cache_hint <uncached >}>: !xegpu.tensor_desc <8 x 16 x f16 >
121121 gpu.return
122122}
123123
124- // CHECK: gpu.func @prefetch_nd_offset_1(%[[arg0:.*]]: memref<8x24x32x48x64xf16 >) {
125- gpu.func @prefetch_nd_offset_1 (%src: memref <8 x 24 x 32 x 48 x 64 x f16 >) {
126- // CHECK: %[[R0:.*]] = xegpu.create_nd_tdesc %[[arg0]][0, 0, 0, 0, 0 ] : memref<8x24x32x48x64xf16 > -> !xegpu.tensor_desc<1x2x4x8x16xf16 >
127- %1 = xegpu.create_nd_tdesc %src [0 , 0 , 0 , 0 , 0 ] : memref <8 x 24 x 32 x 48 x 64 x f16 > -> !xegpu.tensor_desc <1 x 2 x 4 x 8 x 16 x f16 >
128- // CHECK: xegpu.prefetch_nd %[[R0]][0, 0] <{l1_hint = #xegpu.cache_hint<cached>, l2_hint = #xegpu.cache_hint<uncached>}> : !xegpu.tensor_desc<1x2x4x8x16xf16 >
129- xegpu.prefetch_nd %1 [0 , 0 ] <{l1_hint = #xegpu.cache_hint <cached >, l2_hint = #xegpu.cache_hint <uncached >}>: !xegpu.tensor_desc <1 x 2 x 4 x 8 x 16 x f16 >
124+ // CHECK: gpu.func @prefetch_nd_offset_1(%[[arg0:.*]]: memref<48x64xf16 >) {
125+ gpu.func @prefetch_nd_offset_1 (%src: memref <48 x 64 x f16 >) {
126+ // CHECK: %[[R0:.*]] = xegpu.create_nd_tdesc %[[arg0]][0, 0] : memref<48x64xf16 > -> !xegpu.tensor_desc<8x16xf16 >
127+ %1 = xegpu.create_nd_tdesc %src [0 , 0 ] : memref <48 x 64 x f16 > -> !xegpu.tensor_desc <8 x 16 x f16 >
128+ // CHECK: xegpu.prefetch_nd %[[R0]][0, 0] <{l1_hint = #xegpu.cache_hint<cached>, l2_hint = #xegpu.cache_hint<uncached>}> : !xegpu.tensor_desc<8x16xf16 >
129+ xegpu.prefetch_nd %1 [0 , 0 ] <{l1_hint = #xegpu.cache_hint <cached >, l2_hint = #xegpu.cache_hint <uncached >}>: !xegpu.tensor_desc <8 x 16 x f16 >
130130 gpu.return
131131}
132132
0 commit comments