SLPVectorizer: Bring back the insertelement patch (r205965) with fixes

When can't assume a vectorized tree is rooted in an instruction. The IRBuilder
could have constant folded it. When we rebuild the build_vector (the series of
InsertElement instructions) use the last original InsertElement instruction. The
vectorized tree root is guaranteed to be before it.

Also, we can't assume that the n-th InsertElement inserts the n-th element into
a vector.

This reverts r207746 which reverted the revert of the revert of r205018 or so.

Fixes the test case in PR19621.

llvm-svn: 207939

1 files changed, 60 insertions, 0 deletions

diff --git a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
index 7537ea3b053..db33927cb79 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
@@ -195,6 +195,30 @@ define <4 x float> @simple_select_partial_vector(<4 x float> %a, <4 x float> %b,
   ret <4 x float> %rb
 }
 
+; Make sure that vectorization happens even if insertelements operations
+; must be rescheduled. The case here is from compiling Julia.
+define <4 x float> @reschedule_extract(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @reschedule_extract(
+; CHECK: %1 = fadd <4 x float> %a, %b
+  %a0 = extractelement <4 x float> %a, i32 0
+  %b0 = extractelement <4 x float> %b, i32 0
+  %c0 = fadd float %a0, %b0
+  %v0 = insertelement <4 x float> undef, float %c0, i32 0
+  %a1 = extractelement <4 x float> %a, i32 1
+  %b1 = extractelement <4 x float> %b, i32 1
+  %c1 = fadd float %a1, %b1
+  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
+  %a2 = extractelement <4 x float> %a, i32 2
+  %b2 = extractelement <4 x float> %b, i32 2
+  %c2 = fadd float %a2, %b2
+  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
+  %a3 = extractelement <4 x float> %a, i32 3
+  %b3 = extractelement <4 x float> %b, i32 3
+  %c3 = fadd float %a3, %b3
+  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
+  ret <4 x float> %v3
+}
+
 ; Check that cost model for vectorization takes credit for
 ; instructions that are erased.
 define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
@@ -219,4 +243,40 @@ define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
   ret <4 x float> %v3
 }
 
+; Make sure we handle multiple trees that feed one build vector correctly.
+define <4 x double> @multi_tree(double %w, double %x, double %y, double %z) {
+entry:
+  %t0 = fadd double %w , 0.000000e+00
+  %t1 = fadd double %x , 1.000000e+00
+  %t2 = fadd double %y , 2.000000e+00
+  %t3 = fadd double %z , 3.000000e+00
+  %t4 = fmul double %t0, 1.000000e+00
+  %i1 = insertelement <4 x double> undef, double %t4, i32 3
+  %t5 = fmul double %t1, 1.000000e+00
+  %i2 = insertelement <4 x double> %i1, double %t5, i32 2
+  %t6 = fmul double %t2, 1.000000e+00
+  %i3 = insertelement <4 x double> %i2, double %t6, i32 1
+  %t7 = fmul double %t3, 1.000000e+00
+  %i4 = insertelement <4 x double> %i3, double %t7, i32 0
+  ret <4 x double> %i4
+}
+; CHECK-LABEL: @multi_tree
+; CHECK-DAG:  %[[V0:.+]] = insertelement <2 x double> undef, double %w, i32 0
+; CHECK-DAG:  %[[V1:.+]] = insertelement <2 x double> %[[V0]], double %x, i32 1
+; CHECK-DAG:  %[[V2:.+]] = fadd <2 x double> %[[V1]], <double 0.000000e+00, double 1.000000e+00>
+; CHECK-DAG:  %[[V3:.+]] = insertelement <2 x double> undef, double %y, i32 0
+; CHECK-DAG:  %[[V4:.+]] = insertelement <2 x double> %[[V3]], double %z, i32 1
+; CHECK-DAG:  %[[V5:.+]] = fadd <2 x double> %[[V4]], <double 2.000000e+00, double 3.000000e+00>
+; CHECK-DAG:  %[[V6:.+]] = fmul <2 x double> <double 1.000000e+00, double 1.000000e+00>, %[[V2]]
+; CHECK-DAG:  %[[V7:.+]] = extractelement <2 x double> %[[V6]], i32 0
+; CHECK-DAG:  %[[I1:.+]] = insertelement <4 x double> undef, double %[[V7]], i32 3
+; CHECK-DAG:  %[[V8:.+]] = extractelement <2 x double> %[[V6]], i32 1
+; CHECK-DAG:  %[[I2:.+]] = insertelement <4 x double> %[[I1]], double %[[V8]], i32 2
+; CHECK-DAG:  %[[V9:.+]] = fmul <2 x double> <double 1.000000e+00, double 1.000000e+00>, %[[V5]]
+; CHECK-DAG:  %[[V10:.+]] = extractelement <2 x double> %[[V9]], i32 0
+; CHECK-DAG:  %[[I3:.+]] = insertelement <4 x double> %i2, double %[[V10]], i32 1
+; CHECK-DAG:  %[[V11:.+]] = extractelement <2 x double> %[[V9]], i32 1
+; CHECK-DAG:  %[[I4:.+]] = insertelement <4 x double> %i3, double %[[V11]], i32 0
+; CHECK:  ret <4 x double> %[[I4]]
+
 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }