83 lines · c
1// RUN: %clang_cc1 %s -flax-vector-conversions=none -ffreestanding -triple=x86_64-unknown-unknown -target-feature +avx512f -target-feature +amx-int8 \2// RUN: -target-feature +amx-bf16 -target-feature +amx-fp16 -emit-llvm -o - -Werror -pedantic | FileCheck %s --check-prefixes=CHECK3 4#include <immintrin.h>5 6char buf[1024];7#define STRIDE 328 9char buf2[1024];10 11// This is an example code and integration test.12void test_api(int cond, short row, short col) {13 //CHECK-LABEL: @test_api14 //CHECK-DAG: call x86_amx @llvm.x86.tileloadd64.internal15 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})16 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})17 //CHECK-DAG: call x86_amx @llvm.x86.tdpbssd.internal18 //CHECK-DAG: call void @llvm.x86.tilestored64.internal19 __tile1024i a = {row, 8};20 __tile1024i b = {8, col};21 __tile1024i c = {row, col};22 23 if (cond) {24 __tile_loadd(&a, buf, STRIDE);25 __tile_loadd(&b, buf, STRIDE);26 __tile_loadd(&c, buf, STRIDE);27 } else {28 __tile_loadd(&a, buf2, STRIDE);29 __tile_loadd(&b, buf2, STRIDE);30 __tile_loadd(&c, buf2, STRIDE);31 }32 __tile_dpbssd(&c, a, b);33 __tile_stored(buf, STRIDE, c);34}35 36void test_tile_dpbssd(__tile1024i a, __tile1024i b, __tile1024i c) {37 //CHECK-LABEL: @test_tile_dpbssd38 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})39 //CHECK-DAG: call x86_amx @llvm.x86.tdpbssd.internal40 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})41 __tile_dpbssd(&c, a, b);42}43 44void test_tile_dpbsud(__tile1024i a, __tile1024i b, __tile1024i c) {45 //CHECK-LABEL: @test_tile_dpbsud46 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})47 //CHECK-DAG: call x86_amx @llvm.x86.tdpbsud.internal48 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})49 __tile_dpbsud(&c, a, b);50}51 52void test_tile_dpbusd(__tile1024i a, __tile1024i b, __tile1024i c) {53 //CHECK-LABEL: @test_tile_dpbusd54 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})55 //CHECK-DAG: call x86_amx @llvm.x86.tdpbusd.internal56 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})57 __tile_dpbusd(&c, a, b);58}59 60void test_tile_dpbuud(__tile1024i a, __tile1024i b, __tile1024i c) {61 //CHECK-LABEL: @test_tile_dpbuud62 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})63 //CHECK-DAG: call x86_amx @llvm.x86.tdpbuud.internal64 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})65 __tile_dpbuud(&c, a, b);66}67 68void test_tile_dpbf16ps(__tile1024i a, __tile1024i b, __tile1024i c) {69 //CHECK-LABEL: @test_tile_dpbf16ps70 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})71 //CHECK-DAG: call x86_amx @llvm.x86.tdpbf16ps.internal72 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})73 __tile_dpbf16ps(&a, b, c);74}75 76void test_tile_dpfp16ps(__tile1024i a, __tile1024i b, __tile1024i c) {77 //CHECK-LABEL: @test_tile_dpfp16ps78 //CHECK-DAG: call x86_amx @llvm.x86.cast.vector.to.tile.v256i32(<256 x i32> {{%.*}})79 //CHECK-DAG: call x86_amx @llvm.x86.tdpfp16ps.internal80 //CHECK-DAG: call <256 x i32> @llvm.x86.cast.tile.to.vector.v256i32(x86_amx {{%.*}})81 __tile_dpfp16ps(&a, b, c);82}83