@@ -264,24 +264,6 @@ gpu.func @test_load_nd_simt_8(%src: memref<24x32xf32>) {
264264 gpu.return
265265}
266266
267- // CHECK: func @test_load_nd_vc_9(%[[arg0:.*]]: memref<24x32xf32>) {
268- gpu.func @test_load_nd_vc_9 (%src: memref <24 x32 xf32 >) {
269- // CHECK: %[[R0:.*]] = xegpu.create_nd_tdesc %arg0[0, 0] : memref<24x32xf32> -> !xegpu.tensor_desc<8x16xf32>
270- %1 = xegpu.create_nd_tdesc %src [0 , 0 ] : memref <24 x32 xf32 > -> !xegpu.tensor_desc <8 x16 xf32 >
271- // CHECK: %[[R1:.*]] = xegpu.load_nd %[[R0]] <{l1_hint = #xegpu.cache_hint<cached>, l2_hint = #xegpu.cache_hint<uncached>, transpose = array<i64: 1, 0>}> : !xegpu.tensor_desc<8x16xf32> -> vector<16x8xf32>
272- %2 = xegpu.load_nd %1 <{l1_hint = #xegpu.cache_hint <cached >, l2_hint = #xegpu.cache_hint <uncached >, transpose = array<i64 : 1 , 0 >}> : !xegpu.tensor_desc <8 x16 xf32 > -> vector <16 x8 xf32 >
273- gpu.return
274- }
275-
276- // CHECK: func @test_load_nd_simt_9(%[[arg0:.*]]: memref<24x32xf32>) {
277- gpu.func @test_load_nd_simt_9 (%src: memref <24 x32 xf32 >) {
278- // CHECK: %[[R0:.*]] = xegpu.create_nd_tdesc %arg0[0, 0] : memref<24x32xf32> -> !xegpu.tensor_desc<8x16xf32, #xegpu.sg_map<wi_layout = [1, 16], wi_data = [1, 1]>>
279- %1 = xegpu.create_nd_tdesc %src [0 , 0 ] : memref <24 x32 xf32 > -> !xegpu.tensor_desc <8 x16 xf32 , #xegpu.sg_map <wi_layout = [1 , 16 ], wi_data = [1 , 1 ]>>
280- // CHECK: %[[R1:.*]] = xegpu.load_nd %[[R0]] <{l1_hint = #xegpu.cache_hint<cached>, l2_hint = #xegpu.cache_hint<uncached>, transpose = array<i64: 1, 0>}> : !xegpu.tensor_desc<8x16xf32, #xegpu.sg_map<wi_layout = [1, 16], wi_data = [1, 1]>> -> vector<8x1xf32>
281- %2 = xegpu.load_nd %1 <{l1_hint = #xegpu.cache_hint <cached >, l2_hint = #xegpu.cache_hint <uncached >, transpose = array<i64 : 1 , 0 >}> : !xegpu.tensor_desc <8 x16 xf32 , #xegpu.sg_map <wi_layout = [1 , 16 ], wi_data = [1 , 1 ]>> -> vector <8 x1 xf32 >
282- gpu.return
283- }
284-
285267// CHECK: func @test_store_nd_vc(%[[arg0:.*]]: memref<24x32xf16>) {
286268gpu.func @test_store_nd_vc (%dst: memref <24 x32 xf16 >) {
287269 // CHECK: %[[C:.*]] = arith.constant dense<1.000000e+00> : vector<24x32xf16>
0 commit comments