-; RUN: opt < %s -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine -licm -S | FileCheck %s
+; RUN: opt < %s -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.8.0"
-;CHECK: @reduction_sum
+;CHECK-LABEL: @reduction_sum(
;CHECK: phi <4 x i32>
;CHECK: load <4 x i32>
;CHECK: add <4 x i32>
ret i32 %sum.0.lcssa
}
-;CHECK: @reduction_prod
+;CHECK-LABEL: @reduction_prod(
;CHECK: phi <4 x i32>
;CHECK: load <4 x i32>
;CHECK: mul <4 x i32>
ret i32 %prod.0.lcssa
}
-;CHECK: @reduction_mix
+;CHECK-LABEL: @reduction_mix(
;CHECK: phi <4 x i32>
;CHECK: load <4 x i32>
;CHECK: mul nsw <4 x i32>
ret i32 %sum.0.lcssa
}
-;CHECK: @reduction_mul
+;CHECK-LABEL: @reduction_mul(
;CHECK: mul <4 x i32>
;CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
;CHECK: mul <4 x i32>
ret i32 %sum.0.lcssa
}
-;CHECK: @start_at_non_zero
+;CHECK-LABEL: @start_at_non_zero(
;CHECK: phi <4 x i32>
;CHECK: <i32 120, i32 0, i32 0, i32 0>
;CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
ret i32 %sum.0.lcssa
}
-;CHECK: @reduction_and
+;CHECK-LABEL: @reduction_and(
;CHECK: and <4 x i32>
;CHECK: <i32 -1, i32 -1, i32 -1, i32 -1>
;CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
ret i32 %result.0.lcssa
}
-;CHECK: @reduction_or
+;CHECK-LABEL: @reduction_or(
;CHECK: or <4 x i32>
;CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
;CHECK: or <4 x i32>
ret i32 %result.0.lcssa
}
-;CHECK: @reduction_xor
+;CHECK-LABEL: @reduction_xor(
;CHECK: xor <4 x i32>
;CHECK: shufflevector <4 x i32> %{{.*}}, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
;CHECK: xor <4 x i32>
ret i32 %result.0.lcssa
}
-;CHECK: @reduction_sub_rhs
+; In this code the subtracted variable is on the RHS and this is not an induction variable.
+;CHECK-LABEL: @reduction_sub_rhs(
;CHECK-NOT: phi <4 x i32>
;CHECK-NOT: sub nsw <4 x i32>
;CHECK: ret i32
define i32 @reduction_sub_rhs(i32 %n, i32* noalias nocapture %A) nounwind uwtable readonly {
entry:
- %cmp3 = icmp sgt i32 %n, 0
- br i1 %cmp3, label %for.body, label %for.end
+ %cmp4 = icmp sgt i32 %n, 0
+ br i1 %cmp4, label %for.body, label %for.end
for.body: ; preds = %entry, %for.body
%indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
- %x.04 = phi i32 [ %sub, %for.body ], [ 0, %entry ]
+ %x.05 = phi i32 [ %sub, %for.body ], [ 0, %entry ]
%arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
%0 = load i32* %arrayidx, align 4
- %sub = sub nsw i32 %x.04, %0
+ %sub = sub nsw i32 %0, %x.05
%indvars.iv.next = add i64 %indvars.iv, 1
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
%exitcond = icmp eq i32 %lftr.wideiv, %n
ret i32 %x.0.lcssa
}
-;CHECK: @reduction_sub_lhs
-;CHECK-NOT: phi <4 x i32>
-;CHECK-NOT: sub nsw <4 x i32>
+
+; In this test the reduction variable is on the LHS and we can vectorize it.
+;CHECK-LABEL: @reduction_sub_lhs(
+;CHECK: phi <4 x i32>
+;CHECK: sub nsw <4 x i32>
;CHECK: ret i32
define i32 @reduction_sub_lhs(i32 %n, i32* noalias nocapture %A) nounwind uwtable readonly {
entry:
%x.05 = phi i32 [ %sub, %for.body ], [ 0, %entry ]
%arrayidx = getelementptr inbounds i32* %A, i64 %indvars.iv
%0 = load i32* %arrayidx, align 4
- %sub = sub nsw i32 %0, %x.05
+ %sub = sub nsw i32 %x.05, %0
%indvars.iv.next = add i64 %indvars.iv, 1
%lftr.wideiv = trunc i64 %indvars.iv.next to i32
%exitcond = icmp eq i32 %lftr.wideiv, %n
%x.0.lcssa = phi i32 [ 0, %entry ], [ %sub, %for.body ]
ret i32 %x.0.lcssa
}
+
+; We can vectorize conditional reductions with multi-input phis.
+; CHECK: reduction_conditional
+; CHECK: fadd <4 x float>
+
+define float @reduction_conditional(float* %A, float* %B, float* %C, float %S) {
+entry:
+ br label %for.body
+
+for.body:
+ %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
+ %sum.033 = phi float [ %S, %entry ], [ %sum.1, %for.inc ]
+ %arrayidx = getelementptr inbounds float* %A, i64 %indvars.iv
+ %0 = load float* %arrayidx, align 4
+ %arrayidx2 = getelementptr inbounds float* %B, i64 %indvars.iv
+ %1 = load float* %arrayidx2, align 4
+ %cmp3 = fcmp ogt float %0, %1
+ br i1 %cmp3, label %if.then, label %for.inc
+
+if.then:
+ %cmp6 = fcmp ogt float %1, 1.000000e+00
+ br i1 %cmp6, label %if.then8, label %if.else
+
+if.then8:
+ %add = fadd fast float %sum.033, %0
+ br label %for.inc
+
+if.else:
+ %cmp14 = fcmp ogt float %0, 2.000000e+00
+ br i1 %cmp14, label %if.then16, label %for.inc
+
+if.then16:
+ %add19 = fadd fast float %sum.033, %1
+ br label %for.inc
+
+for.inc:
+ %sum.1 = phi float [ %add, %if.then8 ], [ %add19, %if.then16 ], [ %sum.033, %if.else ], [ %sum.033, %for.body ]
+ %indvars.iv.next = add i64 %indvars.iv, 1
+ %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+ %exitcond = icmp ne i32 %lftr.wideiv, 128
+ br i1 %exitcond, label %for.body, label %for.end
+
+for.end:
+ %sum.1.lcssa = phi float [ %sum.1, %for.inc ]
+ ret float %sum.1.lcssa
+}
+
+; We can't vectorize reductions with phi inputs from outside the reduction.
+; CHECK: noreduction_phi
+; CHECK-NOT: fadd <4 x float>
+define float @noreduction_phi(float* %A, float* %B, float* %C, float %S) {
+entry:
+ br label %for.body
+
+for.body:
+ %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
+ %sum.033 = phi float [ %S, %entry ], [ %sum.1, %for.inc ]
+ %arrayidx = getelementptr inbounds float* %A, i64 %indvars.iv
+ %0 = load float* %arrayidx, align 4
+ %arrayidx2 = getelementptr inbounds float* %B, i64 %indvars.iv
+ %1 = load float* %arrayidx2, align 4
+ %cmp3 = fcmp ogt float %0, %1
+ br i1 %cmp3, label %if.then, label %for.inc
+
+if.then:
+ %cmp6 = fcmp ogt float %1, 1.000000e+00
+ br i1 %cmp6, label %if.then8, label %if.else
+
+if.then8:
+ %add = fadd fast float %sum.033, %0
+ br label %for.inc
+
+if.else:
+ %cmp14 = fcmp ogt float %0, 2.000000e+00
+ br i1 %cmp14, label %if.then16, label %for.inc
+
+if.then16:
+ %add19 = fadd fast float %sum.033, %1
+ br label %for.inc
+
+for.inc:
+ %sum.1 = phi float [ %add, %if.then8 ], [ %add19, %if.then16 ], [ 0.000000e+00, %if.else ], [ %sum.033, %for.body ]
+ %indvars.iv.next = add i64 %indvars.iv, 1
+ %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+ %exitcond = icmp ne i32 %lftr.wideiv, 128
+ br i1 %exitcond, label %for.body, label %for.end
+
+for.end:
+ %sum.1.lcssa = phi float [ %sum.1, %for.inc ]
+ ret float %sum.1.lcssa
+}
+
+; We can't vectorize reductions that feed another header PHI.
+; CHECK: noredux_header_phi
+; CHECK-NOT: fadd <4 x float>
+
+define float @noredux_header_phi(float* %A, float* %B, float* %C, float %S) {
+entry:
+ br label %for.body
+
+for.body:
+ %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+ %sum2.09 = phi float [ 0.000000e+00, %entry ], [ %add1, %for.body ]
+ %sum.08 = phi float [ %S, %entry ], [ %add, %for.body ]
+ %arrayidx = getelementptr inbounds float* %B, i64 %indvars.iv
+ %0 = load float* %arrayidx, align 4
+ %add = fadd fast float %sum.08, %0
+ %add1 = fadd fast float %sum2.09, %add
+ %indvars.iv.next = add i64 %indvars.iv, 1
+ %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+ %exitcond = icmp ne i32 %lftr.wideiv, 128
+ br i1 %exitcond, label %for.body, label %for.end
+
+for.end:
+ %add1.lcssa = phi float [ %add1, %for.body ]
+ %add.lcssa = phi float [ %add, %for.body ]
+ %add2 = fadd fast float %add.lcssa, %add1.lcssa
+ ret float %add2
+}
+
+
+; When vectorizing a reduction whose loop header phi value is used outside the
+; loop special care must be taken. Otherwise, the reduced value feeding into the
+; outside user misses a few iterations (VF-1) of the loop.
+; PR16522
+
+; CHECK-LABEL: @phivalueredux(
+; CHECK-NOT: x i32>
+
+define i32 @phivalueredux(i32 %p) {
+entry:
+ br label %for.body
+
+for.body:
+ %t.03 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+ %p.addr.02 = phi i32 [ %p, %entry ], [ %xor, %for.body ]
+ %xor = xor i32 %p.addr.02, -1
+ %inc = add nsw i32 %t.03, 1
+ %exitcond = icmp eq i32 %inc, 16
+ br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+ ret i32 %p.addr.02
+}
+
+; Don't vectorize a reduction value that is not the last in a reduction cyle. We
+; would loose iterations (VF-1) on the operations after that use.
+; PR17498
+
+; CHECK-LABEL: not_last_operation
+; CHECK-NOT: x i32>
+define i32 @not_last_operation(i32 %p, i32 %val) {
+entry:
+ %tobool = icmp eq i32 %p, 0
+ br label %for.body
+
+for.body:
+ %inc613.1 = phi i32 [ 0, %entry ], [ %inc6.1, %for.body ]
+ %inc511.1 = phi i32 [ %val, %entry ], [ %inc5.1, %for.body ]
+ %0 = zext i1 %tobool to i32
+ %inc4.1 = xor i32 %0, 1
+ %inc511.1.inc4.1 = add nsw i32 %inc511.1, %inc4.1
+ %inc5.1 = add nsw i32 %inc511.1.inc4.1, 1
+ %inc6.1 = add nsw i32 %inc613.1, 1
+ %exitcond.1 = icmp eq i32 %inc6.1, 22
+ br i1 %exitcond.1, label %exit, label %for.body
+
+exit:
+ %inc.2 = add nsw i32 %inc511.1.inc4.1, 2
+ ret i32 %inc.2
+}
projects / oota-llvm.git / blobdiff