[llvm] 073e65a - [LoopVectorize] Make needsExtract notice scalarized instructions (#119720)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jan 2 06:31:39 PST 2025
Author: John Brawn
Date: 2025-01-02T14:31:36Z
New Revision: 073e65a8e5f92ca9c63c3fcd1c0ce2a36913f9a6
URL: https://github.com/llvm/llvm-project/commit/073e65a8e5f92ca9c63c3fcd1c0ce2a36913f9a6
DIFF: https://github.com/llvm/llvm-project/commit/073e65a8e5f92ca9c63c3fcd1c0ce2a36913f9a6.diff
LOG: [LoopVectorize] Make needsExtract notice scalarized instructions (#119720)
LoopVectorizationCostModel::needsExtract should recognise instructions
that have been widened by scalarizing as scalar instructions, and thus
not needing an extract when used by later scalarized instructions.
This fixes an incorrect cost calculation in computePredInstDiscount,
where we are adding a scalarization overhead cost when we shouldn't,
though I haven't come up with a test case where it makes a difference.
It will make a difference when the cost model switches to using the cost
kind TCK_CodeSize for optsize, as not doing this causes the test
LoopVectorize/X86/small-size.ll to get worse.
Added:
Modified:
llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
llvm/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll
llvm/test/Transforms/LoopVectorize/ARM/mve-interleaved-cost.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 52186882b4f20f..f2f8a85b7cc233 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -1731,7 +1731,8 @@ class LoopVectorizationCostModel {
bool needsExtract(Value *V, ElementCount VF) const {
Instruction *I = dyn_cast<Instruction>(V);
if (VF.isScalar() || !I || !TheLoop->contains(I) ||
- TheLoop->isLoopInvariant(I))
+ TheLoop->isLoopInvariant(I) ||
+ getWideningDecision(I, VF) == CM_Scalarize)
return false;
// Assume we can vectorize V (and hence we need extraction) if the
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll
index dec124b55cd4e0..a550f1ca14c8b0 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleaved_cost.ll
@@ -170,8 +170,8 @@ entry:
; VF_2-LABEL: Checking a loop in 'i64_factor_8'
; VF_2: Found an estimated cost of 8 for VF 2 For instruction: %tmp2 = load i64, ptr %tmp0, align 8
; VF_2-NEXT: Found an estimated cost of 8 for VF 2 For instruction: %tmp3 = load i64, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i64 %tmp2, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i64 %tmp3, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 8 for VF 2 For instruction: store i64 %tmp2, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 8 for VF 2 For instruction: store i64 %tmp3, ptr %tmp1, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i64.8, ptr %data, i64 %i, i32 2
diff --git a/llvm/test/Transforms/LoopVectorize/ARM/mve-interleaved-cost.ll b/llvm/test/Transforms/LoopVectorize/ARM/mve-interleaved-cost.ll
index 976c6a9a570af9..551b85b7d03577 100644
--- a/llvm/test/Transforms/LoopVectorize/ARM/mve-interleaved-cost.ll
+++ b/llvm/test/Transforms/LoopVectorize/ARM/mve-interleaved-cost.ll
@@ -17,8 +17,8 @@ entry:
; VF_2-LABEL: Checking a loop in 'i8_factor_2'
; VF_2: Found an estimated cost of 12 for VF 2 For instruction: %tmp2 = load i8, ptr %tmp0, align 1
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp3 = load i8, ptr %tmp1, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp2, ptr %tmp0, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp3, ptr %tmp1, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp2, ptr %tmp0, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp3, ptr %tmp1, align 1
; VF_4-LABEL: Checking a loop in 'i8_factor_2'
; VF_4: Found an estimated cost of 4 for VF 4 For instruction: %tmp2 = load i8, ptr %tmp0, align 1
; VF_4-NEXT: Found an estimated cost of 0 for VF 4 For instruction: %tmp3 = load i8, ptr %tmp1, align 1
@@ -58,8 +58,8 @@ entry:
; VF_2-LABEL: Checking a loop in 'i16_factor_2'
; VF_2: Found an estimated cost of 12 for VF 2 For instruction: %tmp2 = load i16, ptr %tmp0, align 2
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp3 = load i16, ptr %tmp1, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp2, ptr %tmp0, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp3, ptr %tmp1, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp2, ptr %tmp0, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp3, ptr %tmp1, align 2
; VF_4-LABEL: Checking a loop in 'i16_factor_2'
; VF_4: Found an estimated cost of 4 for VF 4 For instruction: %tmp2 = load i16, ptr %tmp0, align 2
; VF_4-NEXT: Found an estimated cost of 0 for VF 4 For instruction: %tmp3 = load i16, ptr %tmp1, align 2
@@ -99,8 +99,8 @@ entry:
; VF_2-LABEL: Checking a loop in 'i32_factor_2'
; VF_2: Found an estimated cost of 12 for VF 2 For instruction: %tmp2 = load i32, ptr %tmp0, align 4
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp3 = load i32, ptr %tmp1, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp2, ptr %tmp0, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp3, ptr %tmp1, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp2, ptr %tmp0, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp3, ptr %tmp1, align 4
; VF_4-LABEL: Checking a loop in 'i32_factor_2'
; VF_4: Found an estimated cost of 4 for VF 4 For instruction: %tmp2 = load i32, ptr %tmp0, align 4
; VF_4-NEXT: Found an estimated cost of 0 for VF 4 For instruction: %tmp3 = load i32, ptr %tmp1, align 4
@@ -140,23 +140,23 @@ entry:
; VF_2-LABEL: Checking a loop in 'i64_factor_2'
; VF_2: Found an estimated cost of 22 for VF 2 For instruction: %tmp2 = load i64, ptr %tmp0, align 8
; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: %tmp3 = load i64, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp2, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp3, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp2, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp3, ptr %tmp1, align 8
; VF_4-LABEL: Checking a loop in 'i64_factor_2'
; VF_4: Found an estimated cost of 44 for VF 4 For instruction: %tmp2 = load i64, ptr %tmp0, align 8
; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: %tmp3 = load i64, ptr %tmp1, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp2, ptr %tmp0, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp3, ptr %tmp1, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp2, ptr %tmp0, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp3, ptr %tmp1, align 8
; VF_8-LABEL: Checking a loop in 'i64_factor_2'
; VF_8: Found an estimated cost of 88 for VF 8 For instruction: %tmp2 = load i64, ptr %tmp0, align 8
; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: %tmp3 = load i64, ptr %tmp1, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp2, ptr %tmp0, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp3, ptr %tmp1, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp2, ptr %tmp0, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp3, ptr %tmp1, align 8
; VF_16-LABEL: Checking a loop in 'i64_factor_2'
; VF_16: Found an estimated cost of 176 for VF 16 For instruction: %tmp2 = load i64, ptr %tmp0, align 8
; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: %tmp3 = load i64, ptr %tmp1, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp2, ptr %tmp0, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp3, ptr %tmp1, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp2, ptr %tmp0, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp3, ptr %tmp1, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i64.2, ptr %data, i64 %i, i32 0
@@ -181,8 +181,8 @@ entry:
; VF_2-LABEL: Checking a loop in 'f16_factor_2'
; VF_2: Found an estimated cost of 6 for VF 2 For instruction: %tmp2 = load half, ptr %tmp0, align 2
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp3 = load half, ptr %tmp1, align 2
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store half %tmp2, ptr %tmp0, align 2
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store half %tmp3, ptr %tmp1, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store half %tmp2, ptr %tmp0, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store half %tmp3, ptr %tmp1, align 2
; VF_4-LABEL: Checking a loop in 'f16_factor_2'
; VF_4: Found an estimated cost of 18 for VF 4 For instruction: %tmp2 = load half, ptr %tmp0, align 2
; VF_4-NEXT: Found an estimated cost of 0 for VF 4 For instruction: %tmp3 = load half, ptr %tmp1, align 2
@@ -263,23 +263,23 @@ entry:
; VF_2-LABEL: Checking a loop in 'f64_factor_2'
; VF_2: Found an estimated cost of 6 for VF 2 For instruction: %tmp2 = load double, ptr %tmp0, align 8
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp3 = load double, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp2, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp3, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp2, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp3, ptr %tmp1, align 8
; VF_4-LABEL: Checking a loop in 'f64_factor_2'
; VF_4: Found an estimated cost of 12 for VF 4 For instruction: %tmp2 = load double, ptr %tmp0, align 8
; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: %tmp3 = load double, ptr %tmp1, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp2, ptr %tmp0, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp3, ptr %tmp1, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp2, ptr %tmp0, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp3, ptr %tmp1, align 8
; VF_8-LABEL: Checking a loop in 'f64_factor_2'
; VF_8: Found an estimated cost of 24 for VF 8 For instruction: %tmp2 = load double, ptr %tmp0, align 8
; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: %tmp3 = load double, ptr %tmp1, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp2, ptr %tmp0, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp3, ptr %tmp1, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp2, ptr %tmp0, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp3, ptr %tmp1, align 8
; VF_16-LABEL: Checking a loop in 'f64_factor_2'
; VF_16: Found an estimated cost of 48 for VF 16 For instruction: %tmp2 = load double, ptr %tmp0, align 8
; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: %tmp3 = load double, ptr %tmp1, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp2, ptr %tmp0, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp3, ptr %tmp1, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp2, ptr %tmp0, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp3, ptr %tmp1, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %f64.2, ptr %data, i64 %i, i32 0
@@ -309,30 +309,30 @@ entry:
; VF_2: Found an estimated cost of 12 for VF 2 For instruction: %tmp3 = load i8, ptr %tmp0, align 1
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp4 = load i8, ptr %tmp1, align 1
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp5 = load i8, ptr %tmp2, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp3, ptr %tmp0, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp4, ptr %tmp1, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp5, ptr %tmp2, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp3, ptr %tmp0, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp4, ptr %tmp1, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp5, ptr %tmp2, align 1
; VF_4-LABEL: Checking a loop in 'i8_factor_3'
; VF_4: Found an estimated cost of 24 for VF 4 For instruction: %tmp3 = load i8, ptr %tmp0, align 1
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp4 = load i8, ptr %tmp1, align 1
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp5 = load i8, ptr %tmp2, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp3, ptr %tmp0, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp4, ptr %tmp1, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp5, ptr %tmp2, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp3, ptr %tmp0, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp4, ptr %tmp1, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp5, ptr %tmp2, align 1
; VF_8-LABEL: Checking a loop in 'i8_factor_3'
; VF_8: Found an estimated cost of 48 for VF 8 For instruction: %tmp3 = load i8, ptr %tmp0, align 1
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp4 = load i8, ptr %tmp1, align 1
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp5 = load i8, ptr %tmp2, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp3, ptr %tmp0, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp4, ptr %tmp1, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp5, ptr %tmp2, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp3, ptr %tmp0, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp4, ptr %tmp1, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp5, ptr %tmp2, align 1
; VF_16-LABEL: Checking a loop in 'i8_factor_3'
; VF_16: Found an estimated cost of 96 for VF 16 For instruction: %tmp3 = load i8, ptr %tmp0, align 1
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp4 = load i8, ptr %tmp1, align 1
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp5 = load i8, ptr %tmp2, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp3, ptr %tmp0, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp4, ptr %tmp1, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp5, ptr %tmp2, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp3, ptr %tmp0, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp4, ptr %tmp1, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp5, ptr %tmp2, align 1
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i8.3, ptr %data, i64 %i, i32 0
@@ -361,30 +361,30 @@ entry:
; VF_2: Found an estimated cost of 12 for VF 2 For instruction: %tmp3 = load i16, ptr %tmp0, align 2
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp4 = load i16, ptr %tmp1, align 2
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp5 = load i16, ptr %tmp2, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp3, ptr %tmp0, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp4, ptr %tmp1, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp5, ptr %tmp2, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp3, ptr %tmp0, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp4, ptr %tmp1, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp5, ptr %tmp2, align 2
; VF_4-LABEL: Checking a loop in 'i16_factor_3'
; VF_4: Found an estimated cost of 24 for VF 4 For instruction: %tmp3 = load i16, ptr %tmp0, align 2
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp4 = load i16, ptr %tmp1, align 2
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp5 = load i16, ptr %tmp2, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp3, ptr %tmp0, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp4, ptr %tmp1, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp5, ptr %tmp2, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp3, ptr %tmp0, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp4, ptr %tmp1, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp5, ptr %tmp2, align 2
; VF_8-LABEL: Checking a loop in 'i16_factor_3'
; VF_8: Found an estimated cost of 48 for VF 8 For instruction: %tmp3 = load i16, ptr %tmp0, align 2
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp4 = load i16, ptr %tmp1, align 2
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp5 = load i16, ptr %tmp2, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp3, ptr %tmp0, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp4, ptr %tmp1, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp5, ptr %tmp2, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp3, ptr %tmp0, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp4, ptr %tmp1, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp5, ptr %tmp2, align 2
; VF_16-LABEL: Checking a loop in 'i16_factor_3'
; VF_16: Found an estimated cost of 96 for VF 16 For instruction: %tmp3 = load i16, ptr %tmp0, align 2
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp4 = load i16, ptr %tmp1, align 2
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp5 = load i16, ptr %tmp2, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp3, ptr %tmp0, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp4, ptr %tmp1, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp5, ptr %tmp2, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp3, ptr %tmp0, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp4, ptr %tmp1, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp5, ptr %tmp2, align 2
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i16.3, ptr %data, i64 %i, i32 0
@@ -413,9 +413,9 @@ entry:
; VF_2: Found an estimated cost of 12 for VF 2 For instruction: %tmp3 = load i32, ptr %tmp0, align 4
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp4 = load i32, ptr %tmp1, align 4
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp5 = load i32, ptr %tmp2, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp3, ptr %tmp0, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp4, ptr %tmp1, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp5, ptr %tmp2, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp3, ptr %tmp0, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp4, ptr %tmp1, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp5, ptr %tmp2, align 4
; VF_4-LABEL: Checking a loop in 'i32_factor_3'
; VF_4: Found an estimated cost of 8 for VF 4 For instruction: %tmp3 = load i32, ptr %tmp0, align 4
; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: %tmp4 = load i32, ptr %tmp1, align 4
@@ -427,16 +427,16 @@ entry:
; VF_8: Found an estimated cost of 48 for VF 8 For instruction: %tmp3 = load i32, ptr %tmp0, align 4
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp4 = load i32, ptr %tmp1, align 4
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp5 = load i32, ptr %tmp2, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp3, ptr %tmp0, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp4, ptr %tmp1, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp5, ptr %tmp2, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp3, ptr %tmp0, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp4, ptr %tmp1, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp5, ptr %tmp2, align 4
; VF_16-LABEL: Checking a loop in 'i32_factor_3'
; VF_16: Found an estimated cost of 96 for VF 16 For instruction: %tmp3 = load i32, ptr %tmp0, align 4
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp4 = load i32, ptr %tmp1, align 4
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp5 = load i32, ptr %tmp2, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp3, ptr %tmp0, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp4, ptr %tmp1, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp5, ptr %tmp2, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp3, ptr %tmp0, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp4, ptr %tmp1, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp5, ptr %tmp2, align 4
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i32.3, ptr %data, i64 %i, i32 0
@@ -465,30 +465,30 @@ entry:
; VF_2: Found an estimated cost of 22 for VF 2 For instruction: %tmp3 = load i64, ptr %tmp0, align 8
; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: %tmp4 = load i64, ptr %tmp1, align 8
; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: %tmp5 = load i64, ptr %tmp2, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp3, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp4, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp5, ptr %tmp2, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp3, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp4, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp5, ptr %tmp2, align 8
; VF_4-LABEL: Checking a loop in 'i64_factor_3'
; VF_4: Found an estimated cost of 44 for VF 4 For instruction: %tmp3 = load i64, ptr %tmp0, align 8
; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: %tmp4 = load i64, ptr %tmp1, align 8
; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: %tmp5 = load i64, ptr %tmp2, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp3, ptr %tmp0, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp4, ptr %tmp1, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp5, ptr %tmp2, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp3, ptr %tmp0, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp4, ptr %tmp1, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp5, ptr %tmp2, align 8
; VF_8-LABEL: Checking a loop in 'i64_factor_3'
; VF_8: Found an estimated cost of 88 for VF 8 For instruction: %tmp3 = load i64, ptr %tmp0, align 8
; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: %tmp4 = load i64, ptr %tmp1, align 8
; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: %tmp5 = load i64, ptr %tmp2, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp3, ptr %tmp0, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp4, ptr %tmp1, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp5, ptr %tmp2, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp3, ptr %tmp0, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp4, ptr %tmp1, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp5, ptr %tmp2, align 8
; VF_16-LABEL: Checking a loop in 'i64_factor_3'
; VF_16: Found an estimated cost of 176 for VF 16 For instruction: %tmp3 = load i64, ptr %tmp0, align 8
; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: %tmp4 = load i64, ptr %tmp1, align 8
; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: %tmp5 = load i64, ptr %tmp2, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp3, ptr %tmp0, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp4, ptr %tmp1, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp5, ptr %tmp2, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp3, ptr %tmp0, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp4, ptr %tmp1, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp5, ptr %tmp2, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i64.3, ptr %data, i64 %i, i32 0
@@ -517,9 +517,9 @@ entry:
; VF_2: Found an estimated cost of 6 for VF 2 For instruction: %tmp3 = load half, ptr %tmp0, align 2
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp4 = load half, ptr %tmp1, align 2
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp5 = load half, ptr %tmp2, align 2
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store half %tmp3, ptr %tmp0, align 2
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store half %tmp4, ptr %tmp1, align 2
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store half %tmp5, ptr %tmp2, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store half %tmp3, ptr %tmp0, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store half %tmp4, ptr %tmp1, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store half %tmp5, ptr %tmp2, align 2
; VF_4-LABEL: Checking a loop in 'f16_factor_3'
; VF_4: Found an estimated cost of 28 for VF 4 For instruction: %tmp3 = load half, ptr %tmp0, align 2
; VF_4-NEXT: Found an estimated cost of 0 for VF 4 For instruction: %tmp4 = load half, ptr %tmp1, align 2
@@ -621,30 +621,30 @@ entry:
; VF_2: Found an estimated cost of 6 for VF 2 For instruction: %tmp3 = load double, ptr %tmp0, align 8
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp4 = load double, ptr %tmp1, align 8
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp5 = load double, ptr %tmp2, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp3, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp4, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp5, ptr %tmp2, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp3, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp4, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp5, ptr %tmp2, align 8
; VF_4-LABEL: Checking a loop in 'f64_factor_3'
; VF_4: Found an estimated cost of 12 for VF 4 For instruction: %tmp3 = load double, ptr %tmp0, align 8
; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: %tmp4 = load double, ptr %tmp1, align 8
; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: %tmp5 = load double, ptr %tmp2, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp3, ptr %tmp0, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp4, ptr %tmp1, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp5, ptr %tmp2, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp3, ptr %tmp0, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp4, ptr %tmp1, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp5, ptr %tmp2, align 8
; VF_8-LABEL: Checking a loop in 'f64_factor_3'
; VF_8: Found an estimated cost of 24 for VF 8 For instruction: %tmp3 = load double, ptr %tmp0, align 8
; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: %tmp4 = load double, ptr %tmp1, align 8
; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: %tmp5 = load double, ptr %tmp2, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp3, ptr %tmp0, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp4, ptr %tmp1, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp5, ptr %tmp2, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp3, ptr %tmp0, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp4, ptr %tmp1, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp5, ptr %tmp2, align 8
; VF_16-LABEL: Checking a loop in 'f64_factor_3'
; VF_16: Found an estimated cost of 48 for VF 16 For instruction: %tmp3 = load double, ptr %tmp0, align 8
; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: %tmp4 = load double, ptr %tmp1, align 8
; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: %tmp5 = load double, ptr %tmp2, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp3, ptr %tmp0, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp4, ptr %tmp1, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp5, ptr %tmp2, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp3, ptr %tmp0, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp4, ptr %tmp1, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp5, ptr %tmp2, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %f64.3, ptr %data, i64 %i, i32 0
@@ -677,37 +677,37 @@ entry:
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp5 = load i8, ptr %tmp1, align 1
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp6 = load i8, ptr %tmp2, align 1
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp7 = load i8, ptr %tmp3, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp4, ptr %tmp0, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp5, ptr %tmp1, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp6, ptr %tmp2, align 1
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i8 %tmp7, ptr %tmp3, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp4, ptr %tmp0, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp5, ptr %tmp1, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp6, ptr %tmp2, align 1
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i8 %tmp7, ptr %tmp3, align 1
; VF_4-LABEL: Checking a loop in 'i8_factor_4'
; VF_4: Found an estimated cost of 24 for VF 4 For instruction: %tmp4 = load i8, ptr %tmp0, align 1
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp5 = load i8, ptr %tmp1, align 1
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp6 = load i8, ptr %tmp2, align 1
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp7 = load i8, ptr %tmp3, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp4, ptr %tmp0, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp5, ptr %tmp1, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp6, ptr %tmp2, align 1
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i8 %tmp7, ptr %tmp3, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp4, ptr %tmp0, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp5, ptr %tmp1, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp6, ptr %tmp2, align 1
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i8 %tmp7, ptr %tmp3, align 1
; VF_8-LABEL: Checking a loop in 'i8_factor_4'
; VF_8: Found an estimated cost of 48 for VF 8 For instruction: %tmp4 = load i8, ptr %tmp0, align 1
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp5 = load i8, ptr %tmp1, align 1
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp6 = load i8, ptr %tmp2, align 1
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp7 = load i8, ptr %tmp3, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp4, ptr %tmp0, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp5, ptr %tmp1, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp6, ptr %tmp2, align 1
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i8 %tmp7, ptr %tmp3, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp4, ptr %tmp0, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp5, ptr %tmp1, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp6, ptr %tmp2, align 1
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i8 %tmp7, ptr %tmp3, align 1
; VF_16-LABEL: Checking a loop in 'i8_factor_4'
; VF_16: Found an estimated cost of 96 for VF 16 For instruction: %tmp4 = load i8, ptr %tmp0, align 1
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp5 = load i8, ptr %tmp1, align 1
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp6 = load i8, ptr %tmp2, align 1
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp7 = load i8, ptr %tmp3, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp4, ptr %tmp0, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp5, ptr %tmp1, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp6, ptr %tmp2, align 1
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i8 %tmp7, ptr %tmp3, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp4, ptr %tmp0, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp5, ptr %tmp1, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp6, ptr %tmp2, align 1
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i8 %tmp7, ptr %tmp3, align 1
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i8.4, ptr %data, i64 %i, i32 0
@@ -740,37 +740,37 @@ entry:
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp5 = load i16, ptr %tmp1, align 2
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp6 = load i16, ptr %tmp2, align 2
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp7 = load i16, ptr %tmp3, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp4, ptr %tmp0, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp5, ptr %tmp1, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp6, ptr %tmp2, align 2
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i16 %tmp7, ptr %tmp3, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp4, ptr %tmp0, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp5, ptr %tmp1, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp6, ptr %tmp2, align 2
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i16 %tmp7, ptr %tmp3, align 2
; VF_4-LABEL: Checking a loop in 'i16_factor_4'
; VF_4: Found an estimated cost of 24 for VF 4 For instruction: %tmp4 = load i16, ptr %tmp0, align 2
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp5 = load i16, ptr %tmp1, align 2
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp6 = load i16, ptr %tmp2, align 2
; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: %tmp7 = load i16, ptr %tmp3, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp4, ptr %tmp0, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp5, ptr %tmp1, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp6, ptr %tmp2, align 2
-; VF_4-NEXT: Found an estimated cost of 24 for VF 4 For instruction: store i16 %tmp7, ptr %tmp3, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp4, ptr %tmp0, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp5, ptr %tmp1, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp6, ptr %tmp2, align 2
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store i16 %tmp7, ptr %tmp3, align 2
; VF_8-LABEL: Checking a loop in 'i16_factor_4'
; VF_8: Found an estimated cost of 48 for VF 8 For instruction: %tmp4 = load i16, ptr %tmp0, align 2
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp5 = load i16, ptr %tmp1, align 2
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp6 = load i16, ptr %tmp2, align 2
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp7 = load i16, ptr %tmp3, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp4, ptr %tmp0, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp5, ptr %tmp1, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp6, ptr %tmp2, align 2
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i16 %tmp7, ptr %tmp3, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp4, ptr %tmp0, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp5, ptr %tmp1, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp6, ptr %tmp2, align 2
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i16 %tmp7, ptr %tmp3, align 2
; VF_16-LABEL: Checking a loop in 'i16_factor_4'
; VF_16: Found an estimated cost of 96 for VF 16 For instruction: %tmp4 = load i16, ptr %tmp0, align 2
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp5 = load i16, ptr %tmp1, align 2
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp6 = load i16, ptr %tmp2, align 2
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp7 = load i16, ptr %tmp3, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp4, ptr %tmp0, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp5, ptr %tmp1, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp6, ptr %tmp2, align 2
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i16 %tmp7, ptr %tmp3, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp4, ptr %tmp0, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp5, ptr %tmp1, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp6, ptr %tmp2, align 2
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i16 %tmp7, ptr %tmp3, align 2
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i16.4, ptr %data, i64 %i, i32 0
@@ -803,10 +803,10 @@ entry:
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp5 = load i32, ptr %tmp1, align 4
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp6 = load i32, ptr %tmp2, align 4
; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: %tmp7 = load i32, ptr %tmp3, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp4, ptr %tmp0, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp5, ptr %tmp1, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp6, ptr %tmp2, align 4
-; VF_2-NEXT: Found an estimated cost of 12 for VF 2 For instruction: store i32 %tmp7, ptr %tmp3, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp4, ptr %tmp0, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp5, ptr %tmp1, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp6, ptr %tmp2, align 4
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store i32 %tmp7, ptr %tmp3, align 4
; VF_4-LABEL: Checking a loop in 'i32_factor_4'
; VF_4: Found an estimated cost of 8 for VF 4 For instruction: %tmp4 = load i32, ptr %tmp0, align 4
; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: %tmp5 = load i32, ptr %tmp1, align 4
@@ -821,19 +821,19 @@ entry:
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp5 = load i32, ptr %tmp1, align 4
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp6 = load i32, ptr %tmp2, align 4
; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: %tmp7 = load i32, ptr %tmp3, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp4, ptr %tmp0, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp5, ptr %tmp1, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp6, ptr %tmp2, align 4
-; VF_8-NEXT: Found an estimated cost of 48 for VF 8 For instruction: store i32 %tmp7, ptr %tmp3, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp4, ptr %tmp0, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp5, ptr %tmp1, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp6, ptr %tmp2, align 4
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store i32 %tmp7, ptr %tmp3, align 4
; VF_16-LABEL: Checking a loop in 'i32_factor_4'
; VF_16: Found an estimated cost of 96 for VF 16 For instruction: %tmp4 = load i32, ptr %tmp0, align 4
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp5 = load i32, ptr %tmp1, align 4
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp6 = load i32, ptr %tmp2, align 4
; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: %tmp7 = load i32, ptr %tmp3, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp4, ptr %tmp0, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp5, ptr %tmp1, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp6, ptr %tmp2, align 4
-; VF_16-NEXT: Found an estimated cost of 96 for VF 16 For instruction: store i32 %tmp7, ptr %tmp3, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp4, ptr %tmp0, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp5, ptr %tmp1, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp6, ptr %tmp2, align 4
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store i32 %tmp7, ptr %tmp3, align 4
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i32.4, ptr %data, i64 %i, i32 0
@@ -866,37 +866,37 @@ entry:
; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: %tmp5 = load i64, ptr %tmp1, align 8
; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: %tmp6 = load i64, ptr %tmp2, align 8
; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: %tmp7 = load i64, ptr %tmp3, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp4, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp5, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp6, ptr %tmp2, align 8
-; VF_2-NEXT: Found an estimated cost of 22 for VF 2 For instruction: store i64 %tmp7, ptr %tmp3, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp4, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp5, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp6, ptr %tmp2, align 8
+; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store i64 %tmp7, ptr %tmp3, align 8
; VF_4-LABEL: Checking a loop in 'i64_factor_4'
; VF_4: Found an estimated cost of 44 for VF 4 For instruction: %tmp4 = load i64, ptr %tmp0, align 8
; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: %tmp5 = load i64, ptr %tmp1, align 8
; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: %tmp6 = load i64, ptr %tmp2, align 8
; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: %tmp7 = load i64, ptr %tmp3, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp4, ptr %tmp0, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp5, ptr %tmp1, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp6, ptr %tmp2, align 8
-; VF_4-NEXT: Found an estimated cost of 44 for VF 4 For instruction: store i64 %tmp7, ptr %tmp3, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp4, ptr %tmp0, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp5, ptr %tmp1, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp6, ptr %tmp2, align 8
+; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store i64 %tmp7, ptr %tmp3, align 8
; VF_8-LABEL: Checking a loop in 'i64_factor_4'
; VF_8: Found an estimated cost of 88 for VF 8 For instruction: %tmp4 = load i64, ptr %tmp0, align 8
; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: %tmp5 = load i64, ptr %tmp1, align 8
; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: %tmp6 = load i64, ptr %tmp2, align 8
; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: %tmp7 = load i64, ptr %tmp3, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp4, ptr %tmp0, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp5, ptr %tmp1, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp6, ptr %tmp2, align 8
-; VF_8-NEXT: Found an estimated cost of 88 for VF 8 For instruction: store i64 %tmp7, ptr %tmp3, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp4, ptr %tmp0, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp5, ptr %tmp1, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp6, ptr %tmp2, align 8
+; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store i64 %tmp7, ptr %tmp3, align 8
; VF_16-LABEL: Checking a loop in 'i64_factor_4'
; VF_16: Found an estimated cost of 176 for VF 16 For instruction: %tmp4 = load i64, ptr %tmp0, align 8
; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: %tmp5 = load i64, ptr %tmp1, align 8
; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: %tmp6 = load i64, ptr %tmp2, align 8
; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: %tmp7 = load i64, ptr %tmp3, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp4, ptr %tmp0, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp5, ptr %tmp1, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp6, ptr %tmp2, align 8
-; VF_16-NEXT: Found an estimated cost of 176 for VF 16 For instruction: store i64 %tmp7, ptr %tmp3, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp4, ptr %tmp0, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp5, ptr %tmp1, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp6, ptr %tmp2, align 8
+; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store i64 %tmp7, ptr %tmp3, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %i64.4, ptr %data, i64 %i, i32 0
@@ -1055,37 +1055,37 @@ entry:
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp5 = load double, ptr %tmp1, align 8
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp6 = load double, ptr %tmp2, align 8
; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: %tmp7 = load double, ptr %tmp3, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp4, ptr %tmp0, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp5, ptr %tmp1, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp6, ptr %tmp2, align 8
-; VF_2-NEXT: Found an estimated cost of 6 for VF 2 For instruction: store double %tmp7, ptr %tmp3, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp4, ptr %tmp0, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp5, ptr %tmp1, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp6, ptr %tmp2, align 8
+; VF_2-NEXT: Found an estimated cost of 4 for VF 2 For instruction: store double %tmp7, ptr %tmp3, align 8
; VF_4-LABEL: Checking a loop in 'f64_factor_4'
; VF_4: Found an estimated cost of 12 for VF 4 For instruction: %tmp4 = load double, ptr %tmp0, align 8
; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: %tmp5 = load double, ptr %tmp1, align 8
; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: %tmp6 = load double, ptr %tmp2, align 8
; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: %tmp7 = load double, ptr %tmp3, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp4, ptr %tmp0, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp5, ptr %tmp1, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp6, ptr %tmp2, align 8
-; VF_4-NEXT: Found an estimated cost of 12 for VF 4 For instruction: store double %tmp7, ptr %tmp3, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp4, ptr %tmp0, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp5, ptr %tmp1, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp6, ptr %tmp2, align 8
+; VF_4-NEXT: Found an estimated cost of 8 for VF 4 For instruction: store double %tmp7, ptr %tmp3, align 8
; VF_8-LABEL: Checking a loop in 'f64_factor_4'
; VF_8: Found an estimated cost of 24 for VF 8 For instruction: %tmp4 = load double, ptr %tmp0, align 8
; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: %tmp5 = load double, ptr %tmp1, align 8
; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: %tmp6 = load double, ptr %tmp2, align 8
; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: %tmp7 = load double, ptr %tmp3, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp4, ptr %tmp0, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp5, ptr %tmp1, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp6, ptr %tmp2, align 8
-; VF_8-NEXT: Found an estimated cost of 24 for VF 8 For instruction: store double %tmp7, ptr %tmp3, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp4, ptr %tmp0, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp5, ptr %tmp1, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp6, ptr %tmp2, align 8
+; VF_8-NEXT: Found an estimated cost of 16 for VF 8 For instruction: store double %tmp7, ptr %tmp3, align 8
; VF_16-LABEL: Checking a loop in 'f64_factor_4'
; VF_16: Found an estimated cost of 48 for VF 16 For instruction: %tmp4 = load double, ptr %tmp0, align 8
; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: %tmp5 = load double, ptr %tmp1, align 8
; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: %tmp6 = load double, ptr %tmp2, align 8
; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: %tmp7 = load double, ptr %tmp3, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp4, ptr %tmp0, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp5, ptr %tmp1, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp6, ptr %tmp2, align 8
-; VF_16-NEXT: Found an estimated cost of 48 for VF 16 For instruction: store double %tmp7, ptr %tmp3, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp4, ptr %tmp0, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp5, ptr %tmp1, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp6, ptr %tmp2, align 8
+; VF_16-NEXT: Found an estimated cost of 32 for VF 16 For instruction: store double %tmp7, ptr %tmp3, align 8
for.body:
%i = phi i64 [ 0, %entry ], [ %i.next, %for.body ]
%tmp0 = getelementptr inbounds %f64.4, ptr %data, i64 %i, i32 0
More information about the llvm-commits
mailing list