add lit test

Author: alexbaden

test/TritonIntelGPU/blockptr_load.mlir

@@ -28,6 +28,37 @@ module attributes {"ttg.num-warps" = 8 : i32, "ttg.threads-per-warp" = 16 : i32}
 
 // -----
 
+// CHECK-DAG: llvm.func spir_funccc @_Z38intel_sub_group_f16_f16_matrix_mad_k16Dv8_sDv8_iDv8_f(vector<8xi16>, vector<8xi32>, vector<8xf32>) -> vector<8xf32> attributes {convergent, memory_effects = #llvm.memory_effects<other = none, argMem = none, inaccessibleMem = none>, no_unwind, will_return}
+// CHECK-DAG: llvm.func spir_funccc @_Z41intel_sub_group_2d_block_read_16b_8r16x2cPU3AS1viiiDv2_iPt(!llvm.ptr<1> {llvm.nonnull, llvm.readonly}, i32, i32, i32, vector<2xi32>, !llvm.ptr {llvm.nonnull, llvm.writeonly}) attributes {no_unwind, will_return}
+// CHECK-DAG: llvm.func spir_funccc @_Z52intel_sub_group_2d_block_read_transform_16b_32r16x1cPU3AS1viiiDv2_iPj(!llvm.ptr<1> {llvm.nonnull, llvm.readonly}, i32, i32, i32, vector<2xi32>, !llvm.ptr {llvm.nonnull, llvm.writeonly}) attributes {no_unwind, will_return}
+#blocked = #ttg.blocked<{sizePerThread = [1, 1], threadsPerWarp = [1, 16], warpsPerCTA = [2, 4], order = [1, 0]}>
+#dpas = #triton_intel_gpu.dpas<{repeatCount = 8, systolicDepth = 8, executionSize = 16, opsPerChan = 2, threadsPerWarp = 16, warpsPerCTA = [4, 2], repCluster = [1, 1], A = [8, 16], B = [16, 16], C = [8, 16]}>
+#dot0 = #ttg.dot_op<{opIdx = 0, parent = #dpas, kWidth=2}>
+#dot1 = #ttg.dot_op<{opIdx = 1, parent = #dpas, kWidth=2}>
+module attributes {"ttg.num-warps" = 8 : i32, "ttg.threads-per-warp" = 16 : i32} {
+  tt.func public @matmul_no_scf_with_add_kernel(%arg0: !tt.ptr<f16>, %arg1: !tt.ptr<f16>, %arg2: !tt.ptr<f32>, %arg3: i64, %arg4: i64, %arg5: i64, %arg6: i64, %arg8: i64) {
+    %C = arith.constant dense<0.000000e+00> : tensor<64x64xf32, #dpas>
+    %c0_i32 = arith.constant 0 : i32
+    %c1_i64 = arith.constant 1 : i64
+    %ptrA = tt.make_tensor_ptr %arg0, [%arg3, %arg5], [%arg6, %c1_i64], [%c0_i32, %c0_i32] {order = array<i32: 1, 0>} : <tensor<64x32xf16, #dot0>>
+    %ptrB = tt.make_tensor_ptr %arg1, [%arg5, %arg4], [%arg8, %c1_i64], [%c0_i32, %c0_i32] {order = array<i32: 1, 0>} : <tensor<32x64xf16, #dot1>>
+    // CHECK-COUNT-2: llvm.call spir_funccc @_Z41intel_sub_group_2d_block_read_16b_8r16x2cPU3AS1viiiDv2_iPt({{.*}}) {{.*}} : (!llvm.ptr<1>, i32, i32, i32, vector<2xi32>, !llvm.ptr) -> ()
+    // CHECK-COUNT-2: llvm.call spir_funccc @_Z52intel_sub_group_2d_block_read_transform_16b_32r16x1cPU3AS1viiiDv2_iPj({{.*}}) {{.*}} : (!llvm.ptr<1>, i32, i32, i32, vector<2xi32>, !llvm.ptr) -> ()
+    // CHECK-COUNT-8: llvm.call spir_funccc @_Z38intel_sub_group_f16_f16_matrix_mad_k16Dv8_sDv8_iDv8_f({{.*}}) {{.*}} : (vector<8xi16>, vector<8xi32>, vector<8xf32>) -> vector<8xf32>
+    %A = tt.load %ptrA {boundaryCheck = array<i32: 1>, padding = 1 : i32, triton_intel_gpu.block_io = "row_major"} : !tt.ptr<tensor<64x32xf16, #dot0>>
+    %B = tt.load %ptrB {boundaryCheck = array<i32: 0>, padding = 1 : i32, triton_intel_gpu.block_io = "row_major"} : !tt.ptr<tensor<32x64xf16, #dot1>>
+    %D = tt.dot %A, %B, %C, inputPrecision = tf32 : tensor<64x32xf16, #dot0> * tensor<32x64xf16, #dot1> -> tensor<64x64xf32, #dpas>
+    %ptrX = tt.make_tensor_ptr %arg2, [%arg3, %arg4], [%arg8, %c1_i64], [%c0_i32, %c0_i32] {order = array<i32: 1, 0>} : <tensor<64x64xf32, #dpas>>
+    // CHECK-COUNT-4: llvm.call spir_funccc @_Z41intel_sub_group_2d_block_read_32b_8r16x1cPU3AS1viiiDv2_iPj({{.*}}) {{.*}} : (!llvm.ptr<1>, i32, i32, i32, vector<2xi32>, !llvm.ptr) -> ()
+    %X = tt.load %ptrX {boundaryCheck = array<i32: 0, 1>, triton_intel_gpu.block_io = "row_major"} : !tt.ptr<tensor<64x64xf32, #dpas>>
+    // CHECK-COUNT-32: llvm.fadd {{.*}}, {{.*}}
+    %0 = arith.addf %D, %X : tensor<64x64xf32, #dpas>
+    tt.return
+  }
+}
+
+// -----
+
 #dpas = #triton_intel_gpu.dpas<{repeatCount = 8, systolicDepth = 8, executionSize = 16, opsPerChan = 1, threadsPerWarp = 16, warpsPerCTA = [8, 4], repCluster = [1, 1], A = [8, 8], B = [8, 16], C = [8, 16]}>
 #dot0 = #ttg.dot_op<{opIdx = 0, parent = #dpas, kWidth=1}>
 #dot1 = #ttg.dot_op<{opIdx = 1, parent = #dpas, kWidth=1}>