llvm-for-llvmta/test/CodeGen/Hexagon/csr-func-usedef.ll

73 lines
3.0 KiB
LLVM
Raw Normal View History

2022-04-25 10:02:23 +02:00
; RUN: llc -march=hexagon < %s
; REQUIRES: asserts
target triple = "hexagon"
declare i8* @llvm.hexagon.circ.ldb(i8*, i8*, i32, i32) #1
declare i8* @llvm.hexagon.circ.stb(i8*, i32, i32, i32) #1
define zeroext i8 @circular_loop_test10(i8* %A, i8* %B, i32 %x, i32 %y, i32 %z, i32 %w) #0 {
entry:
%element_load0 = alloca i8, align 1
%element_load2 = alloca i8, align 1
%element_load3 = alloca i8, align 1
%element_load5 = alloca i8, align 1
%or = or i32 %x, 100663296
%or5 = or i32 %z, 100663296
%or7 = or i32 %w, 100663296
br label %for.body
for.body: ; preds = %for.body, %entry
%p0.082 = phi i8* [ %A, %entry ], [ undef, %for.body ]
%element_load.080 = phi i32 [ 0, %entry ], [ %add18, %for.body ]
%p1.079 = phi i8* [ %B, %entry ], [ %1, %for.body ]
%p2.078 = phi i8* [ undef, %entry ], [ %3, %for.body ]
%p3.077 = phi i8* [ undef, %entry ], [ %4, %for.body ]
%0 = call i8* @llvm.hexagon.circ.ldb(i8* %p0.082, i8* nonnull %element_load0, i32 %or, i32 2)
%1 = call i8* @llvm.hexagon.circ.ldb(i8* %p1.079, i8* nonnull null, i32 0, i32 1)
%2 = call i8* @llvm.hexagon.circ.ldb(i8* %p2.078, i8* nonnull %element_load2, i32 %or5, i32 3)
%3 = call i8* @llvm.hexagon.circ.ldb(i8* %2, i8* nonnull %element_load5, i32 %or5, i32 1)
%4 = call i8* @llvm.hexagon.circ.ldb(i8* %p3.077, i8* nonnull %element_load3, i32 %or7, i32 1)
%5 = load i8, i8* null, align 1
%conv = zext i8 %5 to i32
%6 = load i8, i8* %element_load2, align 1
%conv8 = zext i8 %6 to i32
%7 = load i8, i8* %element_load3, align 1
%conv9 = zext i8 %7 to i32
%8 = load i8, i8* undef, align 1
%conv11 = zext i8 %8 to i32
%9 = load i8, i8* %element_load5, align 1
%conv13 = zext i8 %9 to i32
%10 = load i8, i8* %element_load0, align 1
%conv15 = zext i8 %10 to i32
%conv17 = and i32 %element_load.080, 255
%add = add nuw nsw i32 %conv, %conv17
%add10 = add nuw nsw i32 %add, %conv8
%add12 = add nuw nsw i32 %add10, %conv9
%add14 = add nuw nsw i32 %add12, %conv11
%add16 = add nuw nsw i32 %add14, %conv13
%add18 = add nuw nsw i32 %add16, %conv15
%exitcond84 = icmp eq i32 undef, 200
br i1 %exitcond84, label %for.body23, label %for.body
for.body23: ; preds = %for.body23, %for.body
%11 = call i8* @llvm.hexagon.circ.stb(i8* undef, i32 undef, i32 %or, i32 3)
br i1 undef, label %for.body34, label %for.body23
for.body34: ; preds = %for.body34, %for.body23
%element_load.173 = phi i32 [ %add38, %for.body34 ], [ %add18, %for.body23 ]
%arrayidx35 = getelementptr inbounds i8, i8* %B, i32 0
%12 = load i8, i8* %arrayidx35, align 1
%conv36 = zext i8 %12 to i32
%conv37 = and i32 %element_load.173, 255
%add38 = add nuw nsw i32 %conv36, %conv37
br i1 undef, label %for.end42, label %for.body34
for.end42: ; preds = %for.body34
%conv39 = trunc i32 %add38 to i8
ret i8 %conv39
}
attributes #0 = { nounwind optsize }
attributes #1 = { argmemonly nounwind }