65 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=instcombine -S < %s | FileCheck %s3 4define void @foo(ptr %arrayidx16, ptr %arrayidx29, ptr %arrayidx35, i1 %c1, i1 %c2) {5; CHECK-LABEL: @foo(6; CHECK-NEXT: entry:7; CHECK-NEXT: br label [[FOR_COND9:%.*]]8; CHECK: for.cond9:9; CHECK-NEXT: br i1 [[C1:%.*]], label [[FOR_BODY14:%.*]], label [[EXIT:%.*]]10; CHECK: for.body14:11; CHECK-NEXT: [[T5:%.*]] = load <256 x i32>, ptr [[ARRAYIDX16:%.*]], align 6412; CHECK-NEXT: br label [[FOR_COND18:%.*]]13; CHECK: for.cond18:14; CHECK-NEXT: [[SUB_C_SROA_0_0:%.*]] = phi <256 x i32> [ [[T5]], [[FOR_BODY14]] ], [ [[T12:%.*]], [[FOR_BODY24:%.*]] ]15; CHECK-NEXT: br i1 [[C2:%.*]], label [[FOR_BODY24]], label [[FOR_COND_CLEANUP23:%.*]]16; CHECK: for.cond.cleanup23:17; CHECK-NEXT: store <256 x i32> [[SUB_C_SROA_0_0]], ptr [[ARRAYIDX16]], align 6418; CHECK-NEXT: br label [[FOR_COND9]]19; CHECK: for.body24:20; CHECK-NEXT: [[T6:%.*]] = load <256 x i32>, ptr [[ARRAYIDX29:%.*]], align 6421; CHECK-NEXT: [[T7:%.*]] = load <256 x i32>, ptr [[ARRAYIDX35:%.*]], align 6422; CHECK-NEXT: [[T8:%.*]] = bitcast <256 x i32> [[SUB_C_SROA_0_0]] to x86_amx23; CHECK-NEXT: [[T9:%.*]] = bitcast <256 x i32> [[T6]] to x86_amx24; CHECK-NEXT: [[T10:%.*]] = bitcast <256 x i32> [[T7]] to x86_amx25; CHECK-NEXT: [[T11:%.*]] = call x86_amx @llvm.x86.tdpbssd.internal(i16 1, i16 4, i16 4, x86_amx [[T8]], x86_amx [[T9]], x86_amx [[T10]])26; CHECK-NEXT: [[T12]] = bitcast x86_amx [[T11]] to <256 x i32>27; CHECK-NEXT: br label [[FOR_COND18]]28; CHECK: exit:29; CHECK-NEXT: ret void30;31entry:32 br label %for.cond933for.cond9: ; preds = %for.cond, %for.cond.cleanup2334 br i1 %c1, label %for.body14, label %exit35 36for.body14:37 %t5 = load <256 x i32>, ptr %arrayidx16, align 6438 br label %for.cond1839 40for.cond18: ; preds = %for.body24, %for.body1441 %sub_c.sroa.0.0 = phi <256 x i32> [ %t5, %for.body14 ], [ %t12, %for.body24 ]42 br i1 %c2, label %for.body24, label %for.cond.cleanup2343 44for.cond.cleanup23: ; preds = %for.cond1845 store <256 x i32> %sub_c.sroa.0.0, ptr %arrayidx16, align 6446 br label %for.cond947 48for.body24: ; preds = %for.cond1849 %t6 = load <256 x i32>, ptr %arrayidx29, align 6450 %t7 = load <256 x i32>, ptr %arrayidx35, align 6451 %t8 = bitcast <256 x i32> %sub_c.sroa.0.0 to x86_amx52 %t9 = bitcast <256 x i32> %t6 to x86_amx53 %t10 = bitcast <256 x i32> %t7 to x86_amx54 %t11 = call x86_amx @llvm.x86.tdpbssd.internal(i16 1, i16 4, i16 4, x86_amx %t8, x86_amx %t9, x86_amx %t10) #1255 %t12 = bitcast x86_amx %t11 to <256 x i32>56 br label %for.cond1857 58exit:59 ret void60}61 62declare x86_amx @llvm.x86.tileloadd64.internal(i16, i16, ptr, i64)63declare x86_amx @llvm.x86.tdpbssd.internal(i16, i16, i16, x86_amx, x86_amx, x86_amx)64declare void @llvm.x86.tilestored64.internal(i16, i16, ptr, i64, x86_amx)65