1; RUN: opt < %s -mtriple=x86_64-unknown-unknown -mattr=+amx-int8 -mattr=+avx512f -lower-amx-type -S | FileCheck %s 2 3@buf = dso_local global [1024 x i8] zeroinitializer, align 16 4@buf2 = dso_local global [1024 x i8] zeroinitializer, align 16 5 6; Function Attrs: nounwind uwtable 7define dso_local void @test_api(i32 %cond, i16 signext %row, i16 signext %col) local_unnamed_addr { 8 9; CHECK-LABEL: entry: 10; CHECK: %{{[0-9]+}} = alloca <256 x i32>, align 1024 11; CHECK-NEXT: %{{[0-9]+}} = bitcast <256 x i32>* %{{[0-9]+}} to i8* 12; CHECK-NEXT: %{{[0-9]+}} = alloca <256 x i32>, align 1024 13; CHECK-NEXT: %{{[0-9]+}} = bitcast <256 x i32>* %{{[0-9]+}} to i8* 14; CHECK-NEXT: %{{[0-9]+}} = alloca <256 x i32>, align 1024 15; CHECK-NEXT: %{{[0-9]+}} = bitcast <256 x i32>* %{{[0-9]+}} to i8* 16; CHECK-NEXT: %{{[0-9]+}} = alloca <256 x i32>, align 1024 17; CHECK-NEXT: %{{[0-9]+}} = bitcast <256 x i32>* %{{[0-9]+}} to i8* 18; CHECK-NEXT: %tobool.not = icmp eq i32 %cond, 0 19; CHECK-NEXT: br i1 %tobool.not, label %if.else, label %if.then 20; CHECK: if.then: 21; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 8, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32) 22; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 %row, i16 8, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 23; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32) 24; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 8, i16 %col, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 25; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32) 26; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 %row, i16 %col, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 27; CHECK-NEXT: br label %if.end 28; CHECK: if.else: 29; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 8, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf2, i64 0, i64 0), i64 32) 30; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 %row, i16 8, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 31; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf2, i64 0, i64 0), i64 32) 32; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 8, i16 %col, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 33; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf2, i64 0, i64 0), i64 32) 34; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 %row, i16 %col, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 35; CHECK-NEXT: br label %if.end 36; CHECK: if.end: 37; CHECK-NEXT: %{{[0-9]+}} = call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 8, i8* %{{[0-9]+}}, i64 64) 38; CHECK-NEXT: %{{[0-9]+}} = call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %col, i8* %{{[0-9]+}}, i64 64) 39; CHECK-NEXT: %{{[0-9]+}} = call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 %col, i8* %{{[0-9]+}}, i64 64) 40; CHECK-NEXT: %{{[0-9]+}} = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %row, i16 %col, i16 8, x86_amx %{{[0-9]+}}, x86_amx %{{[0-9]+}}, x86_amx %{{[0-9]+}}) 41; CHECK-NEXT: call void @llvm.x86.tilestored64.internal(i16 %row, i16 %col, i8* %{{[0-9]+}}, i64 64, x86_amx %{{[0-9]+}}) 42; CHECK-NEXT: %{{[0-9]+}} = call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 %col, i8* %{{[0-9]+}}, i64 64) 43; CHECK-NEXT: tail call void @llvm.x86.tilestored64.internal(i16 %row, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %{{[0-9]+}}) 44; CHECK-NEXT: ret void 45 46entry: 47 %tobool.not = icmp eq i32 %cond, 0 48 br i1 %tobool.not, label %if.else, label %if.then 49 50if.then: ; preds = %entry 51 %0 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 8, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32) 52 %1 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32) 53 %2 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32) 54 br label %if.end 55 56if.else: ; preds = %entry 57 %3 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 8, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf2, i64 0, i64 0), i64 32) 58 %4 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 8, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf2, i64 0, i64 0), i64 32) 59 %5 = tail call x86_amx @llvm.x86.tileloadd64.internal(i16 %row, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf2, i64 0, i64 0), i64 32) 60 br label %if.end 61 62if.end: ; preds = %if.else, %if.then 63 %a.sroa.1094.0.in = phi x86_amx [ %3, %if.else ], [ %0, %if.then ] 64 %b.sroa.1069.0.in = phi x86_amx [ %4, %if.else ], [ %1, %if.then ] 65 %c.sroa.1044.0.in = phi x86_amx [ %5, %if.else ], [ %2, %if.then ] 66 %6 = tail call x86_amx @llvm.x86.tdpbssd.internal(i16 %row, i16 %col, i16 8, x86_amx %c.sroa.1044.0.in, x86_amx %a.sroa.1094.0.in, x86_amx %b.sroa.1069.0.in) 67 tail call void @llvm.x86.tilestored64.internal(i16 %row, i16 %col, i8* getelementptr inbounds ([1024 x i8], [1024 x i8]* @buf, i64 0, i64 0), i64 32, x86_amx %6) 68 ret void 69} 70 71; Function Attrs: nounwind 72declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, i8*, i64) 73 74; Function Attrs: nounwind 75declare x86_amx @llvm.x86.tdpbssd.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx) 76 77; Function Attrs: nounwind 78declare void @llvm.x86.tilestored64.internal(i16, i16, i8*, i64, x86_amx) 79