BBVectorize: Choose pair ordering to minimize shuffles
BBVectorize would, except for loads and stores, always fuse instructions
so that the first instruction (in the current source order) would always
represent the low part of the input vectors and the second instruction
would always represent the high part. This lead to too many shuffles
being produced because sometimes the opposite order produces fewer of them.
With this change, BBVectorize tracks the kind of pair connections that form
the DAG of candidate pairs, and uses that information to reorder the pairs to
avoid excess shuffles. Using this information, a future commit will be able
to add VTTI-based shuffle costs to the pair selection procedure. Importantly,
the number of remaining shuffles can now be estimated during pair selection.
There are some trivial instruction reorderings in the test cases, and one
simple additional test where we certainly want to do a reordering to
avoid an unnecessary shuffle.
llvm-svn: 167122
diff --git a/llvm/test/Transforms/BBVectorize/simple.ll b/llvm/test/Transforms/BBVectorize/simple.ll
index 7cd8133..d9a12ee 100644
--- a/llvm/test/Transforms/BBVectorize/simple.ll
+++ b/llvm/test/Transforms/BBVectorize/simple.ll
@@ -5,8 +5,8 @@
define double @test1(double %A1, double %A2, double %B1, double %B2) {
; CHECK: @test1
; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
-; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
%X1 = fsub double %A1, %B1
%X2 = fsub double %A2, %B2
@@ -29,8 +29,8 @@
define double @test2(double %A1, double %A2, double %B1, double %B2) {
; CHECK: @test2
; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
-; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
%X1 = fsub double %A1, %B1
%X2 = fsub double %A2, %B2
@@ -40,12 +40,13 @@
; CHECK: %Y1 = fmul <2 x double> %X1, %X1.v.i0.2
%Z1 = fadd double %Y2, %B1
%Z2 = fadd double %Y1, %B2
-; CHECK: %Z1.v.i0 = shufflevector <2 x double> %Y1, <2 x double> undef, <2 x i32> <i32 1, i32 0>
-; CHECK: %Z1 = fadd <2 x double> %Z1.v.i0, %X1.v.i1.2
+; CHECK: %Z1.v.i1.1 = insertelement <2 x double> undef, double %B2, i32 0
+; CHECK: %Z1.v.i1.2 = insertelement <2 x double> %Z1.v.i1.1, double %B1, i32 1
+; CHECK: %Z2 = fadd <2 x double> %Y1, %Z1.v.i1.2
%R = fmul double %Z1, %Z2
-; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0
-; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1
-; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2
+; CHECK: %Z2.v.r1 = extractelement <2 x double> %Z2, i32 0
+; CHECK: %Z2.v.r2 = extractelement <2 x double> %Z2, i32 1
+; CHECK: %R = fmul double %Z2.v.r2, %Z2.v.r1
ret double %R
; CHECK: ret double %R
}
@@ -54,8 +55,8 @@
define double @test3(double %A1, double %A2, double %B1, double %B2) {
; CHECK: @test3
; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
-; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
%X1 = fsub double %A1, %B1
%X2 = fsub double %A2, %B2
@@ -79,8 +80,8 @@
define double @test4(double %A1, double %A2, double %B1, double %B2) {
; CHECK: @test4
; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
-; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
%X1 = fsub double %A1, %B1
%X2 = fsub double %A2, %B2
@@ -148,3 +149,27 @@
; CHECK: ret <8 x i8> %R
}
+; Basic depth-3 chain (flipped order)
+define double @test7(double %A1, double %A2, double %B1, double %B2) {
+; CHECK: @test7
+; CHECK: %X1.v.i1.1 = insertelement <2 x double> undef, double %B1, i32 0
+; CHECK: %X1.v.i1.2 = insertelement <2 x double> %X1.v.i1.1, double %B2, i32 1
+; CHECK: %X1.v.i0.1 = insertelement <2 x double> undef, double %A1, i32 0
+; CHECK: %X1.v.i0.2 = insertelement <2 x double> %X1.v.i0.1, double %A2, i32 1
+ %X1 = fsub double %A1, %B1
+ %X2 = fsub double %A2, %B2
+; CHECK: %X1 = fsub <2 x double> %X1.v.i0.2, %X1.v.i1.2
+ %Y1 = fmul double %X1, %A1
+ %Y2 = fmul double %X2, %A2
+; CHECK: %Y1 = fmul <2 x double> %X1, %X1.v.i0.2
+ %Z2 = fadd double %Y2, %B2
+ %Z1 = fadd double %Y1, %B1
+; CHECK: %Z1 = fadd <2 x double> %Y1, %X1.v.i1.2
+ %R = fmul double %Z1, %Z2
+; CHECK: %Z1.v.r1 = extractelement <2 x double> %Z1, i32 0
+; CHECK: %Z1.v.r2 = extractelement <2 x double> %Z1, i32 1
+; CHECK: %R = fmul double %Z1.v.r1, %Z1.v.r2
+ ret double %R
+; CHECK: ret double %R
+}
+