diff options
Diffstat (limited to 'llvm/test/Transforms/LoopVectorize/SystemZ')
9 files changed, 469 insertions, 0 deletions
diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/addressing.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/addressing.ll new file mode 100644 index 00000000000..1f7a6d29c57 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/addressing.ll @@ -0,0 +1,72 @@ +; RUN: opt -S -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize -dce \ +; RUN: -instcombine -force-vector-width=2 < %s | FileCheck %s +; +; Test that loop vectorizer does not generate vector addresses that must then +; always be extracted. + +; Check that the addresses for a scalarized memory access is not extracted +; from a vector register. +define i32 @foo(i32* nocapture %A) { +;CHECK-LABEL: @foo( +;CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] +;CHECK: %0 = shl nsw i64 %index, 2 +;CHECK: %1 = shl i64 %index, 2 +;CHECK: %2 = or i64 %1, 4 +;CHECK: %3 = getelementptr inbounds i32, i32* %A, i64 %0 +;CHECK: %4 = getelementptr inbounds i32, i32* %A, i64 %2 +;CHECK: store i32 4, i32* %3, align 4 +;CHECK: store i32 4, i32* %4, align 4 + +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %0 = shl nsw i64 %indvars.iv, 2 + %arrayidx = getelementptr inbounds i32, i32* %A, i64 %0 + store i32 4, i32* %arrayidx, align 4 + %indvars.iv.next = add i64 %indvars.iv, 1 + %lftr.wideiv = trunc i64 %indvars.iv.next to i32 + %exitcond = icmp eq i32 %lftr.wideiv, 10000 + br i1 %exitcond, label %for.end, label %for.body + +for.end: + ret i32 undef +} + + +; Check that a load of address is scalarized. +define i32 @foo1(i32* nocapture noalias %A, i32** nocapture %PtrPtr) { +;CHECK-LABEL: @foo1( +;CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ] +;CHECK: %0 = or i64 %index, 1 +;CHECK: %1 = getelementptr inbounds i32*, i32** %PtrPtr, i64 %index +;CHECK: %2 = getelementptr inbounds i32*, i32** %PtrPtr, i64 %0 +;CHECK: %3 = load i32*, i32** %1, align 8 +;CHECK: %4 = load i32*, i32** %2, align 8 +;CHECK: %5 = load i32, i32* %3, align 4 +;CHECK: %6 = load i32, i32* %4, align 4 +;CHECK: %7 = insertelement <2 x i32> undef, i32 %5, i32 0 +;CHECK: %8 = insertelement <2 x i32> %7, i32 %6, i32 1 +;CHECK: %9 = getelementptr inbounds i32, i32* %A, i64 %index +;CHECK: %10 = bitcast i32* %9 to <2 x i32>* +;CHECK: store <2 x i32> %8, <2 x i32>* %10, align 4 + +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] + %ptr = getelementptr inbounds i32*, i32** %PtrPtr, i64 %indvars.iv + %el = load i32*, i32** %ptr + %v = load i32, i32* %el + %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv + store i32 %v, i32* %arrayidx, align 4 + %indvars.iv.next = add i64 %indvars.iv, 1 + %lftr.wideiv = trunc i64 %indvars.iv.next to i32 + %exitcond = icmp eq i32 %lftr.wideiv, 10000 + br i1 %exitcond, label %for.end, label %for.body + +for.end: + ret i32 undef +} diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/branch-for-predicated-block.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/branch-for-predicated-block.ll new file mode 100644 index 00000000000..d2e59452033 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/branch-for-predicated-block.ll @@ -0,0 +1,38 @@ +; REQUIRES: asserts +; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize \ +; RUN: -force-vector-width=2 -debug-only=loop-vectorize \ +; RUN: -disable-output < %s 2>&1 | FileCheck %s + +; Check costs for branches inside a vectorized loop around predicated +; blocks. Each such branch will be guarded with an extractelement from the +; vector compare plus a test under mask instruction. This cost is modelled on +; the extractelement of i1. + +define void @fun(i32* %arr, i64 %trip.count) { +entry: + br label %for.body + +for.body: + %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ] + %arrayidx = getelementptr inbounds i32, i32* %arr, i64 %indvars.iv + %l = load i32, i32* %arrayidx, align 4 + %cmp55 = icmp sgt i32 %l, 0 + br i1 %cmp55, label %if.then, label %for.inc + +if.then: + %sub = sub nsw i32 0, %l + store i32 %sub, i32* %arrayidx, align 4 + br label %for.inc + +for.inc: + %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 + %exitcond = icmp eq i64 %indvars.iv.next, %trip.count + br i1 %exitcond, label %for.end.loopexit, label %for.body + +for.end.loopexit: + ret void + +; CHECK: LV: Found an estimated cost of 5 for VF 2 For instruction: br i1 %cmp55, label %if.then, label %for.inc +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: br label %for.inc +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: br i1 %exitcond, label %for.end.loopexit, label %for.body +} diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/lit.local.cfg b/llvm/test/Transforms/LoopVectorize/SystemZ/lit.local.cfg new file mode 100644 index 00000000000..2f3cf7d3f04 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/lit.local.cfg @@ -0,0 +1,2 @@ +if not 'SystemZ' in config.root.targets: + config.unsupported = True diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll new file mode 100644 index 00000000000..1925527eacf --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-0.ll @@ -0,0 +1,27 @@ +; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize \ +; RUN: -force-vector-width=2 -debug-only=loop-vectorize \ +; RUN: -disable-output < %s 2>&1 | FileCheck %s +; REQUIRES: asserts +; +; Check that a scalarized load does not get operands scalarization costs added. + +define void @fun(i64* %data, i64 %n, i64 %s, double* %Src) { +entry: + br label %for.body + +for.body: + %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] + %mul = mul nsw i64 %iv, %s + %gep = getelementptr inbounds double, double* %Src, i64 %mul + %bct = bitcast double* %gep to i64* + %ld = load i64, i64* %bct + %iv.next = add nuw nsw i64 %iv, 1 + %cmp110.us = icmp slt i64 %iv.next, %n + br i1 %cmp110.us, label %for.body, label %for.end + +for.end: + ret void + +; CHECK: LV: Found an estimated cost of 2 for VF 2 For instruction: %mul = mul nsw i64 %iv, %s +; CHECK: LV: Found an estimated cost of 2 for VF 2 For instruction: %ld = load i64, i64* %bct +} diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-1.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-1.ll new file mode 100644 index 00000000000..fbf8b114542 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/load-scalarization-cost-1.ll @@ -0,0 +1,28 @@ +; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize \ +; RUN: -force-vector-width=4 -debug-only=loop-vectorize \ +; RUN: -enable-interleaved-mem-accesses=false -disable-output < %s 2>&1 \ +; RUN: | FileCheck %s +; REQUIRES: asserts +; +; Check that a scalarized load does not get a zero cost in a vectorized +; loop. It can only be folded into the add operand in the scalar loop. + +define i32 @fun(i64* %data, i64 %n, i64 %s, i32* %Src) { +entry: + br label %for.body + +for.body: + %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] + %acc = phi i32 [ 0, %entry ], [ %acc_next, %for.body ] + %gep = getelementptr inbounds i32, i32* %Src, i64 %iv + %ld = load i32, i32* %gep + %acc_next = add i32 %acc, %ld + %iv.next = add nuw nsw i64 %iv, 2 + %cmp110.us = icmp slt i64 %iv.next, %n + br i1 %cmp110.us, label %for.body, label %for.end + +for.end: + ret i32 %acc_next + +; CHECK: Found an estimated cost of 4 for VF 4 For instruction: %ld = load i32, i32* %gep +} diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/load-store-scalarization-cost.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/load-store-scalarization-cost.ll new file mode 100644 index 00000000000..9fdf22ecd92 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/load-store-scalarization-cost.ll @@ -0,0 +1,33 @@ +; REQUIRES: asserts +; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize \ +; RUN: -force-vector-width=4 -debug-only=loop-vectorize \ +; RUN: -disable-output -enable-interleaved-mem-accesses=false < %s 2>&1 | \ +; RUN: FileCheck %s +; +; Check that a scalarized load/store does not get a cost for insterts/ +; extracts, since z13 supports element load/store. + +define void @fun(i32* %data, i64 %n) { +entry: + br label %for.body + +for.body: + %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] + %tmp0 = getelementptr inbounds i32, i32* %data, i64 %i + %tmp1 = load i32, i32* %tmp0, align 4 + %tmp2 = add i32 %tmp1, 1 + store i32 %tmp2, i32* %tmp0, align 4 + %i.next = add nuw nsw i64 %i, 2 + %cond = icmp slt i64 %i.next, %n + br i1 %cond, label %for.body, label %for.end + +for.end: + ret void + +; CHECK: LV: Scalarizing: %tmp1 = load i32, i32* %tmp0, align 4 +; CHECK: LV: Scalarizing: store i32 %tmp2, i32* %tmp0, align 4 + +; CHECK: LV: Found an estimated cost of 4 for VF 4 For instruction: %tmp1 = load i32, i32* %tmp0, align 4 +; CHECK: LV: Found an estimated cost of 4 for VF 4 For instruction: store i32 %tmp2, i32* %tmp0, align 4 +} + diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/mem-interleaving-costs-02.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/mem-interleaving-costs-02.ll new file mode 100644 index 00000000000..4c992cedd88 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/mem-interleaving-costs-02.ll @@ -0,0 +1,149 @@ +; REQUIRES: asserts +; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize \ +; RUN: -debug-only=loop-vectorize,vectorutils -max-interleave-group-factor=64\ +; RUN: -disable-output < %s 2>&1 | FileCheck %s +; +; Check that some cost estimations for interleave groups make sense. + +; This loop is loading four i16 values at indices [0, 1, 2, 3], with a stride +; of 4. At VF=4, memory interleaving means loading 4 * 4 * 16 bits = 2 vector +; registers. Each of the 4 vector values must then be constructed from the +; two vector registers using one vperm each, which gives a cost of 2 + 4 = 6. +; +; CHECK: LV: Checking a loop in "fun0" +; CHECK: LV: Found an estimated cost of 6 for VF 4 For instruction: %ld0 = load i16 +; CHECK: LV: Found an estimated cost of 0 for VF 4 For instruction: %ld1 = load i16 +; CHECK: LV: Found an estimated cost of 0 for VF 4 For instruction: %ld2 = load i16 +; CHECK: LV: Found an estimated cost of 0 for VF 4 For instruction: %ld3 = load i16 +define void @fun0(i16 *%ptr, i16 *%dst) { +entry: + br label %for.body + +for.body: + %ivptr = phi i16* [ %ptr.next, %for.body ], [ %ptr, %entry ] + %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ] + %inc = add i64 %iv, 4 + %ptr0 = getelementptr inbounds i16, i16* %ivptr, i64 0 + %ld0 = load i16, i16* %ptr0 + %ptr1 = getelementptr inbounds i16, i16* %ivptr, i64 1 + %ld1 = load i16, i16* %ptr1 + %ptr2 = getelementptr inbounds i16, i16* %ivptr, i64 2 + %ld2 = load i16, i16* %ptr2 + %ptr3 = getelementptr inbounds i16, i16* %ivptr, i64 3 + %ld3 = load i16, i16* %ptr3 + %a1 = add i16 %ld0, %ld1 + %a2 = add i16 %a1, %ld2 + %a3 = add i16 %a2, %ld3 + %dstptr = getelementptr inbounds i16, i16* %dst, i64 %iv + store i16 %a3, i16* %dstptr + %ptr.next = getelementptr inbounds i16, i16* %ivptr, i64 4 + %cmp = icmp eq i64 %inc, 100 + br i1 %cmp, label %for.end, label %for.body + +for.end: + ret void +} + +; This loop loads one i8 value in a stride of 3. At VF=16, this means loading +; 3 vector registers, and then constructing the vector value with two vperms, +; which gives a cost of 5. +; +; CHECK: LV: Checking a loop in "fun1" +; CHECK: LV: Found an estimated cost of 5 for VF 16 For instruction: %ld0 = load i8 +define void @fun1(i8 *%ptr, i8 *%dst) { +entry: + br label %for.body + +for.body: + %ivptr = phi i8* [ %ptr.next, %for.body ], [ %ptr, %entry ] + %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ] + %inc = add i64 %iv, 4 + %ptr0 = getelementptr inbounds i8, i8* %ivptr, i64 0 + %ld0 = load i8, i8* %ptr0 + %dstptr = getelementptr inbounds i8, i8* %dst, i64 %iv + store i8 %ld0, i8* %dstptr + %ptr.next = getelementptr inbounds i8, i8* %ivptr, i64 3 + %cmp = icmp eq i64 %inc, 100 + br i1 %cmp, label %for.end, label %for.body + +for.end: + ret void +} + +; This loop is loading 4 i8 values at indexes [0, 1, 2, 3], with a stride of +; 32. At VF=2, this means loading 2 vector registers, and using 4 vperms to +; produce the vector values, which gives a cost of 6. +; +; CHECK: LV: Checking a loop in "fun2" +; CHECK: LV: Found an estimated cost of 6 for VF 2 For instruction: %ld0 = load i8 +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %ld1 = load i8 +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %ld2 = load i8 +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %ld3 = load i8 +define void @fun2(i8 *%ptr, i8 *%dst) { +entry: + br label %for.body + +for.body: + %ivptr = phi i8* [ %ptr.next, %for.body ], [ %ptr, %entry ] + %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ] + %inc = add i64 %iv, 4 + %ptr0 = getelementptr inbounds i8, i8* %ivptr, i64 0 + %ld0 = load i8, i8* %ptr0 + %ptr1 = getelementptr inbounds i8, i8* %ivptr, i64 1 + %ld1 = load i8, i8* %ptr1 + %ptr2 = getelementptr inbounds i8, i8* %ivptr, i64 2 + %ld2 = load i8, i8* %ptr2 + %ptr3 = getelementptr inbounds i8, i8* %ivptr, i64 3 + %ld3 = load i8, i8* %ptr3 + %a1 = add i8 %ld0, %ld1 + %a2 = add i8 %a1, %ld2 + %a3 = add i8 %a2, %ld3 + %dstptr = getelementptr inbounds i8, i8* %dst, i64 %iv + store i8 %a3, i8* %dstptr + %ptr.next = getelementptr inbounds i8, i8* %ivptr, i64 32 + %cmp = icmp eq i64 %inc, 100 + br i1 %cmp, label %for.end, label %for.body + +for.end: + ret void +} + +; This loop is loading 4 i8 values at indexes [0, 1, 2, 3], with a stride of +; 30. At VF=2, this means loading 3 vector registers, and using 4 vperms to +; produce the vector values, which gives a cost of 7. This is the same loop +; as in fun2, except the stride makes the second iterations values overlap a +; vector register boundary. +; +; CHECK: LV: Checking a loop in "fun3" +; CHECK: LV: Found an estimated cost of 7 for VF 2 For instruction: %ld0 = load i8 +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %ld1 = load i8 +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %ld2 = load i8 +; CHECK: LV: Found an estimated cost of 0 for VF 2 For instruction: %ld3 = load i8 +define void @fun3(i8 *%ptr, i8 *%dst) { +entry: + br label %for.body + +for.body: + %ivptr = phi i8* [ %ptr.next, %for.body ], [ %ptr, %entry ] + %iv = phi i64 [ %inc, %for.body ], [ 0, %entry ] + %inc = add i64 %iv, 4 + %ptr0 = getelementptr inbounds i8, i8* %ivptr, i64 0 + %ld0 = load i8, i8* %ptr0 + %ptr1 = getelementptr inbounds i8, i8* %ivptr, i64 1 + %ld1 = load i8, i8* %ptr1 + %ptr2 = getelementptr inbounds i8, i8* %ivptr, i64 2 + %ld2 = load i8, i8* %ptr2 + %ptr3 = getelementptr inbounds i8, i8* %ivptr, i64 3 + %ld3 = load i8, i8* %ptr3 + %a1 = add i8 %ld0, %ld1 + %a2 = add i8 %a1, %ld2 + %a3 = add i8 %a2, %ld3 + %dstptr = getelementptr inbounds i8, i8* %dst, i64 %iv + store i8 %a3, i8* %dstptr + %ptr.next = getelementptr inbounds i8, i8* %ivptr, i64 30 + %cmp = icmp eq i64 %inc, 100 + br i1 %cmp, label %for.end, label %for.body + +for.end: + ret void +} diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/mem-interleaving-costs.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/mem-interleaving-costs.ll new file mode 100644 index 00000000000..4e04a3423ed --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/mem-interleaving-costs.ll @@ -0,0 +1,70 @@ +; REQUIRES: asserts +; RUN: opt -mtriple=s390x-unknown-linux -mcpu=z13 -loop-vectorize \ +; RUN: -force-vector-width=4 -debug-only=loop-vectorize,vectorutils \ +; RUN: -disable-output < %s 2>&1 | FileCheck %s +; +; Check that the loop vectorizer performs memory interleaving with accurate +; cost estimations. + + +; Simple case where just the load is interleaved, because the store group +; would have gaps. +define void @fun0(i32* %data, i64 %n) { +entry: + br label %for.body + +for.body: + %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] + %tmp0 = getelementptr inbounds i32, i32* %data, i64 %i + %tmp1 = load i32, i32* %tmp0, align 4 + %tmp2 = add i32 %tmp1, 1 + store i32 %tmp2, i32* %tmp0, align 4 + %i.next = add nuw nsw i64 %i, 2 + %cond = icmp slt i64 %i.next, %n + br i1 %cond, label %for.body, label %for.end + +for.end: + ret void + +; CHECK: LV: Creating an interleave group with: %tmp1 = load i32, i32* %tmp0, align 4 +; CHECK: LV: Found an estimated cost of 3 for VF 4 For instruction: %tmp1 = load i32, i32* %tmp0, align 4 +; (vl; vl; vperm) +} + +; Interleaving of both load and stores. +define void @fun1(i32* %data, i64 %n) { +entry: + br label %for.body + +for.body: + %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] + %tmp0 = getelementptr inbounds i32, i32* %data, i64 %i + %tmp1 = load i32, i32* %tmp0, align 4 + %i_1 = add i64 %i, 1 + %tmp2 = getelementptr inbounds i32, i32* %data, i64 %i_1 + %tmp3 = load i32, i32* %tmp2, align 4 + store i32 %tmp1, i32* %tmp2, align 4 + store i32 %tmp3, i32* %tmp0, align 4 + %i.next = add nuw nsw i64 %i, 2 + %cond = icmp slt i64 %i.next, %n + br i1 %cond, label %for.body, label %for.end + +for.end: + ret void + +; CHECK: LV: Creating an interleave group with: store i32 %tmp3, i32* %tmp0, align 4 +; CHECK: LV: Inserted: store i32 %tmp1, i32* %tmp2, align 4 +; CHECK: into the interleave group with store i32 %tmp3, i32* %tmp0, align 4 +; CHECK: LV: Creating an interleave group with: %tmp3 = load i32, i32* %tmp2, align 4 +; CHECK: LV: Inserted: %tmp1 = load i32, i32* %tmp0, align 4 +; CHECK: into the interleave group with %tmp3 = load i32, i32* %tmp2, align 4 + +; CHECK: LV: Found an estimated cost of 4 for VF 4 For instruction: %tmp1 = load i32, i32* %tmp0, align 4 +; CHECK: LV: Found an estimated cost of 0 for VF 4 For instruction: %tmp3 = load i32, i32* %tmp2, align 4 +; (vl; vl; vperm, vpkg) + +; CHECK: LV: Found an estimated cost of 0 for VF 4 For instruction: store i32 %tmp1, i32* %tmp2, align 4 +; CHECK: LV: Found an estimated cost of 4 for VF 4 For instruction: store i32 %tmp3, i32* %tmp0, align 4 +; (vmrlf; vmrhf; vst; vst) +} + diff --git a/llvm/test/Transforms/LoopVectorize/SystemZ/pr38110.ll b/llvm/test/Transforms/LoopVectorize/SystemZ/pr38110.ll new file mode 100644 index 00000000000..6c8fef9fc91 --- /dev/null +++ b/llvm/test/Transforms/LoopVectorize/SystemZ/pr38110.ll @@ -0,0 +1,50 @@ +; RUN: opt -passes='loop-vectorize' -mcpu=z13 -force-vector-width=2 -S < %s | FileCheck %s +; +; Forcing VF=2 to trigger vector code gen +; +; This is a test case to exercise more cases in truncateToMinimalBitWidths(). +; Test passes if vector code is generated w/o hitting llvm_unreachable(). +; +; Performing minimal check in the output to ensure the loop is actually +; vectorized. +; +; CHECK: vector.body + +target datalayout = "E-m:e-i1:8:16-i8:8:16-i64:64-f128:64-v128:64-a:8:16-n32:64" +target triple = "s390x-ibm-linux" + +define void @test(i32 zeroext %width, i8* nocapture %row, i16 zeroext %src, i16* nocapture readonly %dst) { +entry: + %cmp10 = icmp eq i32 %width, 0 + br i1 %cmp10, label %for.end, label %for.body.lr.ph + +for.body.lr.ph: ; preds = %entry + %conv1 = zext i16 %src to i32 + br label %for.body + +for.body: ; preds = %for.inc, %for.body.lr.ph + %i.012 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.inc ] + %sp.011 = phi i8* [ %row, %for.body.lr.ph ], [ %incdec.ptr, %for.inc ] + %0 = load i8, i8* %sp.011, align 1 + %conv = zext i8 %0 to i32 + %cmp2 = icmp eq i32 %conv, %conv1 + br i1 %cmp2, label %if.then, label %for.inc + +if.then: ; preds = %for.body + %1 = load i16, i16* %dst, align 2 + %conv4 = trunc i16 %1 to i8 + store i8 %conv4, i8* %sp.011, align 1 + br label %for.inc + +for.inc: ; preds = %for.body, %if.then + %inc = add nuw i32 %i.012, 1 + %incdec.ptr = getelementptr inbounds i8, i8* %sp.011, i64 1 + %exitcond = icmp eq i32 %inc, %width + br i1 %exitcond, label %for.end.loopexit, label %for.body + +for.end.loopexit: ; preds = %for.inc + br label %for.end + +for.end: ; preds = %for.end.loopexit, %entry + ret void +} |
