1 // RUN: %clang_cc1 %s -flax-vector-conversions=none -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512f \ 2 // RUN: -target-feature +amx-transpose -target-feature +amx-bf16 -target-feature +amx-fp16 -target-feature +amx-complex \ 3 // RUN: -emit-llvm -o - -Werror -pedantic | FileCheck %s --check-prefixes=CHECK 4 5 #include <immintrin.h> 6 7 char buf[2048]; 8 #define STRIDE 32 9 10 char buf2[2048]; 11 12 void test_tile_2rpntlvwz0(__tile1024i dst0, __tile1024i dst1) { 13 //CHECK-LABEL: @test_tile_2rpntlvwz0 14 //CHECK: call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0.internal 15 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 0 16 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 17 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 18 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 1 19 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 20 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 21 __tile_2rpntlvwz0(&dst0, &dst1, buf, STRIDE); 22 } 23 24 void test_tile_2rpntlvwz0t1(__tile1024i dst0, __tile1024i dst1) { 25 //CHECK-LABEL: @test_tile_2rpntlvwz0t1 26 //CHECK: call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz0t1.internal 27 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 0 28 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 29 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 30 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 1 31 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 32 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 33 __tile_2rpntlvwz0t1(&dst0, &dst1, buf, STRIDE); 34 } 35 36 void test_tile_2rpntlvwz1(__tile1024i dst0, __tile1024i dst1) { 37 //CHECK-LABEL: @test_tile_2rpntlvwz1 38 //CHECK: call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1.internal 39 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 0 40 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 41 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 42 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 1 43 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 44 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 45 __tile_2rpntlvwz1(&dst0, &dst1, buf, STRIDE); 46 } 47 48 void test_tile_2rpntlvwz1t1(__tile1024i dst0, __tile1024i dst1) { 49 //CHECK-LABEL: @test_tile_2rpntlvwz1t1 50 //CHECK: call { x86_amx, x86_amx } @llvm.x86.t2rpntlvwz1t1.internal 51 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 0 52 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 53 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 54 //CHECK-NEXT: {{%.*}} = extractvalue { x86_amx, x86_amx } {{%.*}}, 1 55 //CHECK-NEXT: {{%.*}} = call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 56 //CHECK-NEXT: store <256 x i32> {{%.*}}, ptr {{%.*}} 57 __tile_2rpntlvwz1t1(&dst0, &dst1, buf, STRIDE); 58 } 59 60 void test_tile_transposed(__tile1024i dst, __tile1024i src) { 61 //CHECK-LABEL: @test_tile_transposed 62 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 63 //CHECK-DAG: call x86_amx @llvm.x86.ttransposed.internal 64 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 65 __tile_transposed(&dst, src); 66 } 67 68 void test_tile_tdpbf16ps(__tile1024i a, __tile1024i b, __tile1024i c) { 69 //CHECK-LABEL: @test_tile_tdpbf16ps 70 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 71 //CHECK-DAG: call x86_amx @llvm.x86.ttdpbf16ps.internal 72 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 73 __tile_tdpbf16ps(&c, a, b); 74 } 75 76 void test_tile_tdpfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) { 77 //CHECK-LABEL: @test_tile_tdpfp16ps 78 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 79 //CHECK-DAG: call x86_amx @llvm.x86.ttdpfp16ps.internal 80 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 81 __tile_tdpfp16ps(&c, a, b); 82 } 83 84 void test_tile_tcmmimfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) { 85 //CHECK-LABEL: @test_tile_tcmmimfp16ps 86 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 87 //CHECK-DAG: call x86_amx @llvm.x86.ttcmmimfp16ps.internal 88 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 89 __tile_tcmmimfp16ps(&c, a, b); 90 } 91 92 void test_tile_tcmmrlfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) { 93 //CHECK-LABEL: @test_tile_tcmmrlfp16ps 94 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 95 //CHECK-DAG: call x86_amx @llvm.x86.ttcmmrlfp16ps.internal 96 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 97 __tile_tcmmrlfp16ps(&c, a, b); 98 } 99 100 void test_tile_conjtcmmimfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) { 101 //CHECK-LABEL: @test_tile_conjtcmmimfp16ps 102 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 103 //CHECK-DAG: call x86_amx @llvm.x86.tconjtcmmimfp16ps.internal 104 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 105 __tile_conjtcmmimfp16ps(&c, a, b); 106 } 107 108 void test_tile_conjtfp16(__tile1024i dst, __tile1024i src) { 109 //CHECK-LABEL: @test_tile_conjtfp16 110 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}}) 111 //CHECK-DAG: call x86_amx @llvm.x86.tconjtfp16.internal 112 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}}) 113 __tile_conjtfp16(&dst, src); 114 } 115