34 lines · plain
1; REQUIRES: asserts2; RUN: opt -passes=loop-vectorize -mtriple riscv64 -mattr=+v -vectorizer-maximize-bandwidth -debug-only=loop-vectorize,vplan -disable-output -S < %s 2>&1 | FileCheck %s --check-prefixes=CHECK-REGS-VP3 4define i32 @dotp(ptr %a, ptr %b) {5; CHECK-REGS-VP: LV(REG): VF = vscale x 166; CHECK-REGS-VP-NEXT: LV(REG): Found max usage: 2 item7; CHECK-REGS-VP-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 6 registers8; CHECK-REGS-VP-NEXT: LV(REG): RegisterClass: RISCV::VRRC, 24 registers9; CHECK-REGS-VP-NEXT: LV(REG): Found invariant usage: 1 item10; CHECK-REGS-VP-NEXT: LV(REG): RegisterClass: RISCV::GPRRC, 1 registers11; CHECK-REGS-VP: LV: Selecting VF: vscale x 16.12entry:13 br label %for.body14 15for.body: ; preds = %for.body, %entry16 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ]17 %accum = phi i32 [ 0, %entry ], [ %add, %for.body ]18 %gep.a = getelementptr i8, ptr %a, i64 %iv19 %load.a = load i8, ptr %gep.a, align 120 %ext.a = zext i8 %load.a to i3221 %gep.b = getelementptr i8, ptr %b, i64 %iv22 %load.b = load i8, ptr %gep.b, align 123 %ext.b = zext i8 %load.b to i3224 %mul = mul i32 %ext.b, %ext.a25 %sub = sub i32 0, %mul26 %add = add i32 %accum, %sub27 %iv.next = add i64 %iv, 128 %exitcond.not = icmp eq i64 %iv.next, 102429 br i1 %exitcond.not, label %for.exit, label %for.body30 31for.exit: ; preds = %for.body32 ret i32 %add33}34