1*038b7e6bSXiang1 Zhang // RUN: %clang_cc1 %s -flax-vector-conversions=none -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512f -target-feature +amx-bf16 \
2*038b7e6bSXiang1 Zhang // RUN: -target-feature +amx-complex \
3*038b7e6bSXiang1 Zhang // RUN: -emit-llvm -o - -Werror -pedantic | FileCheck %s --check-prefixes=CHECK
4*038b7e6bSXiang1 Zhang
5*038b7e6bSXiang1 Zhang #include <immintrin.h>
6*038b7e6bSXiang1 Zhang
7*038b7e6bSXiang1 Zhang char buf[1024];
8*038b7e6bSXiang1 Zhang #define STRIDE 32
9*038b7e6bSXiang1 Zhang
10*038b7e6bSXiang1 Zhang char buf2[1024];
11*038b7e6bSXiang1 Zhang
test_tile_cmmimfp16ps(__tile1024i a,__tile1024i b,__tile1024i c)12*038b7e6bSXiang1 Zhang void test_tile_cmmimfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) {
13*038b7e6bSXiang1 Zhang //CHECK-LABEL: @test_tile_cmmimfp16ps
14*038b7e6bSXiang1 Zhang //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})
15*038b7e6bSXiang1 Zhang //CHECK-DAG: call x86_amx @llvm.x86.tcmmimfp16ps.internal
16*038b7e6bSXiang1 Zhang //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})
17*038b7e6bSXiang1 Zhang __tile_cmmimfp16ps(&c, a, b);
18*038b7e6bSXiang1 Zhang }
19*038b7e6bSXiang1 Zhang
test_tile_cmmrlfp16ps(__tile1024i a,__tile1024i b,__tile1024i c)20*038b7e6bSXiang1 Zhang void test_tile_cmmrlfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) {
21*038b7e6bSXiang1 Zhang //CHECK-LABEL: @test_tile_cmmrlfp16ps
22*038b7e6bSXiang1 Zhang //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})
23*038b7e6bSXiang1 Zhang //CHECK-DAG: call x86_amx @llvm.x86.tcmmrlfp16ps.internal
24*038b7e6bSXiang1 Zhang //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})
25*038b7e6bSXiang1 Zhang __tile_cmmrlfp16ps(&c, a, b);
26*038b7e6bSXiang1 Zhang }
27