maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

Enable vectorizer-maximize-bandwidth by default. 

Summary:
vectorizer-maximize-bandwidth is generally useful in terms of performance. I've tested the impact of changing this to default on speccpu benchmarks on sandybridge machines. The result shows non-negative impact:

spec/2006/fp/C++/444.namd                 26.84  -0.31%
spec/2006/fp/C++/447.dealII               46.19  +0.89%
spec/2006/fp/C++/450.soplex               42.92  -0.44%
spec/2006/fp/C++/453.povray               38.57  -2.25%
spec/2006/fp/C/433.milc                   24.54  -0.76%
spec/2006/fp/C/470.lbm                    41.08  +0.26%
spec/2006/fp/C/482.sphinx3                47.58  -0.99%
spec/2006/int/C++/471.omnetpp             22.06  +1.87%
spec/2006/int/C++/473.astar               22.65  -0.12%
spec/2006/int/C++/483.xalancbmk           33.69  +4.97%
spec/2006/int/C/400.perlbench             33.43  +1.70%
spec/2006/int/C/401.bzip2                 23.02  -0.19%
spec/2006/int/C/403.gcc                   32.57  -0.43%
spec/2006/int/C/429.mcf                   40.35  +0.27%
spec/2006/int/C/445.gobmk                 26.96  +0.06%
spec/2006/int/C/456.hmmer                  24.4  +0.19%
spec/2006/int/C/458.sjeng                 27.91  -0.08%
spec/2006/int/C/462.libquantum            57.47  -0.20%
spec/2006/int/C/464.h264ref               46.52  +1.35%

geometric mean                                   +0.29%

The regression on 453.povray seems real, but is due to secondary effects as all hot functions are bit-identical with and without the flag.

I started this patch to consult upstream opinions on this. It will be greatly appreciated if the community can help test the performance impact of this change on other architectures so that we can decided if this should be target-dependent.

Reviewers: hfinkel, mkuper, davidxl, chandlerc

Reviewed By: chandlerc

Subscribers: rengolin, sanjoy, javed.absar, bjope, dorit, magabari, RKSimon, llvm-commits, mzolotukhin

Differential Revision: https://reviews.llvm.org/D33341

llvm-svn: 305960
This commit is contained in:
Dehao Chen 2017-06-21 22:01:32 +00:00
parent 7b871611b9
commit 014db29b89
12 changed files with 77 additions and 68 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
View File

@ -122,7 +122,7 @@ static cl::opt<unsigned> TinyTripCountVectorThreshold(
"value."));
static cl::opt<bool> MaximizeBandwidth(
"vectorizer-maximize-bandwidth", cl::init(false), cl::Hidden,
"vectorizer-maximize-bandwidth", cl::init(true), cl::Hidden,
cl::desc("Maximize bandwidth when selecting vectorization factor which "
"will be determined by the smallest type in loop."));

32
llvm/test/Transforms/LoopVectorize/AArch64/loop-vectorization-factors.ll
View File

@ -88,9 +88,9 @@ for.body: ; preds = %entry, %for.body
}
; CHECK-LABEL: @add_c(
; CHECK: load <8 x i8>, <8 x i8>*
; CHECK: add <8 x i16>
; CHECK: store <8 x i16>
; CHECK: load <16 x i8>, <16 x i8>*
; CHECK: add <16 x i16>
; CHECK: store <16 x i16>
; Function Attrs: nounwind
define void @add_c(i8* noalias nocapture readonly %p, i16* noalias nocapture %q, i32 %len) #0 {
entry:
@ -116,9 +116,9 @@ for.body: ; preds = %entry, %for.body
}
; CHECK-LABEL: @add_d(
; CHECK: load <4 x i16>
; CHECK: add nsw <4 x i32>
; CHECK: store <4 x i32>
; CHECK: load <8 x i16>
; CHECK: add nsw <8 x i32>
; CHECK: store <8 x i32>
define void @add_d(i16* noalias nocapture readonly %p, i32* noalias nocapture %q, i32 %len) #0 {
entry:
%cmp7 = icmp sgt i32 %len, 0
@ -187,16 +187,16 @@ for.body: ; preds = %for.body, %for.body
}
; CHECK-LABEL: @add_f
; CHECK: load <8 x i16>
; CHECK: trunc <8 x i16>
; CHECK: shl <8 x i8>
; CHECK: add <8 x i8>
; CHECK: or <8 x i8>
; CHECK: mul <8 x i8>
; CHECK: and <8 x i8>
; CHECK: xor <8 x i8>
; CHECK: mul <8 x i8>
; CHECK: store <8 x i8>
; CHECK: load <16 x i16>
; CHECK: trunc <16 x i16>
; CHECK: shl <16 x i8>
; CHECK: add <16 x i8>
; CHECK: or <16 x i8>
; CHECK: mul <16 x i8>
; CHECK: and <16 x i8>
; CHECK: xor <16 x i8>
; CHECK: mul <16 x i8>
; CHECK: store <16 x i8>
define void @add_f(i16* noalias nocapture readonly %p, i8* noalias nocapture %q, i8 %arg1, i8 %arg2, i32 %len) #0 {
entry:
%cmp.32 = icmp sgt i32 %len, 0

16
llvm/test/Transforms/LoopVectorize/AArch64/reduction-small-size.ll
View File

@ -123,16 +123,16 @@ for.body:
; }
;
; CHECK: vector.body:
; CHECK: phi <8 x i16>
; CHECK: [[Ld1:%[a-zA-Z0-9.]+]] = load <8 x i8>
; CHECK: zext <8 x i8> [[Ld1]] to <8 x i16>
; CHECK: [[Ld2:%[a-zA-Z0-9.]+]] = load <8 x i8>
; CHECK: zext <8 x i8> [[Ld2]] to <8 x i16>
; CHECK: add <8 x i16>
; CHECK: add <8 x i16>
; CHECK: phi <16 x i16>
; CHECK: [[Ld1:%[a-zA-Z0-9.]+]] = load <16 x i8>
; CHECK: zext <16 x i8> [[Ld1]] to <16 x i16>
; CHECK: [[Ld2:%[a-zA-Z0-9.]+]] = load <16 x i8>
; CHECK: zext <16 x i8> [[Ld2]] to <16 x i16>
; CHECK: add <16 x i16>
; CHECK: add <16 x i16>
;
; CHECK: middle.block:
; CHECK: [[Rdx:%[a-zA-Z0-9.]+]] = call i16 @llvm.experimental.vector.reduce.add.i16.v8i16(<8 x i16>
; CHECK: [[Rdx:%[a-zA-Z0-9.]+]] = call i16 @llvm.experimental.vector.reduce.add.i16.v16i16(<16 x i16>
; CHECK: zext i16 [[Rdx]] to i32
;
define i16 @reduction_i16_2(i8* nocapture readonly %a, i8* nocapture readonly %b, i32 %n) {

6
llvm/test/Transforms/LoopVectorize/ARM/gcc-examples.ll
View File

@ -35,9 +35,9 @@ define void @example1() nounwind uwtable ssp {
}
;CHECK-LABEL: @example10b(
;CHECK: load <4 x i16>
;CHECK: sext <4 x i16>
;CHECK: store <4 x i32>
;CHECK: load <8 x i16>
;CHECK: sext <8 x i16>
;CHECK: store <8 x i32>
;CHECK: ret void
define void @example10b(i16* noalias nocapture %sa, i16* noalias nocapture %sb, i16* noalias nocapture %sc, i32* noalias nocapture %ia, i32* noalias nocapture %ib, i32* noalias nocapture %ic) nounwind uwtable ssp {
br label %1

4
llvm/test/Transforms/LoopVectorize/X86/fp64_to_uint32-cost-model.ll
View File

@ -9,7 +9,9 @@ target triple = "x86_64-apple-macosx"
; If we need to scalarize the fptoui and then use inserts to build up the
; vector again, then there is certainly no value in going 256-bit wide.
; CHECK-NOT: vpinsrd
; But as we default to maximize bandwidth, we should convert it to 256-bit
; anyway.
; CHECK: vpinsrd
define void @convert() {
entry:

15
llvm/test/Transforms/LoopVectorize/X86/gcc-examples.ll
View File

@ -44,17 +44,16 @@ define void @example1() nounwind uwtable ssp {
ret void
}
; Select VF=4 because sext <8 x i1> to <8 x i32> is expensive.
;CHECK-LABEL: @example10b(
;CHECK: load <4 x i16>
;CHECK: sext <4 x i16>
;CHECK: store <4 x i32>
;CHECK: load <8 x i16>
;CHECK: sext <8 x i16>
;CHECK: store <8 x i32>
;CHECK: ret void
;UNROLL-LABEL: @example10b(
;UNROLL: load <4 x i16>
;UNROLL: load <4 x i16>
;UNROLL: store <4 x i32>
;UNROLL: store <4 x i32>
;UNROLL: load <8 x i16>
;UNROLL: load <8 x i16>
;UNROLL: store <8 x i32>
;UNROLL: store <8 x i32>
;UNROLL: ret void
define void @example10b(i16* noalias nocapture %sa, i16* noalias nocapture %sb, i16* noalias nocapture %sc, i32* noalias nocapture %ia, i32* noalias nocapture %ib, i32* noalias nocapture %ic) nounwind uwtable ssp {
br label %1

60
llvm/test/Transforms/LoopVectorize/X86/masked_load_store.ll
View File

@ -260,20 +260,28 @@ for.end: ; preds = %for.cond
; }
;}
;AVX-LABEL: @foo3
;AVX: icmp slt <4 x i32> %wide.load, <i32 100, i32 100,
;AVX: call <4 x double> @llvm.masked.load.v4f64.p0v4f64
;AVX: sitofp <4 x i32> %wide.load to <4 x double>
;AVX: fadd <4 x double>
;AVX: call void @llvm.masked.store.v4f64.p0v4f64
;AVX: ret void
;AVX1-LABEL: @foo3
;AVX1: icmp slt <4 x i32> %wide.load, <i32 100, i32 100,
;AVX1: call <4 x double> @llvm.masked.load.v4f64.p0v4f64
;AVX1: sitofp <4 x i32> %wide.load to <4 x double>
;AVX1: fadd <4 x double>
;AVX1: call void @llvm.masked.store.v4f64.p0v4f64
;AVX1: ret void
;AVX2-LABEL: @foo3
;AVX2: icmp slt <8 x i32> %wide.load, <i32 100, i32 100,
;AVX2: call <8 x double> @llvm.masked.load.v8f64.p0v8f64
;AVX2: sitofp <8 x i32> %wide.load to <8 x double>
;AVX2: fadd <8 x double>
;AVX2: call void @llvm.masked.store.v8f64.p0v8f64
;AVX2: ret void
;AVX512-LABEL: @foo3
;AVX512: icmp slt <8 x i32> %wide.load, <i32 100, i32 100,
;AVX512: call <8 x double> @llvm.masked.load.v8f64.p0v8f64
;AVX512: sitofp <8 x i32> %wide.load to <8 x double>
;AVX512: fadd <8 x double>
;AVX512: call void @llvm.masked.store.v8f64.p0v8f64
;AVX512: icmp slt <16 x i32> %wide.load, <i32 100, i32 100,
;AVX512: call <16 x double> @llvm.masked.load.v16f64.p0v16f64
;AVX512: sitofp <16 x i32> %wide.load to <16 x double>
;AVX512: fadd <16 x double>
;AVX512: call void @llvm.masked.store.v16f64.p0v16f64
;AVX512: ret void
@ -502,19 +510,19 @@ for.end: ; preds = %for.cond
; }
;}
;AVX2-LABEL: @foo6
;AVX2: icmp sgt <4 x i32> %reverse, zeroinitializer
;AVX2: shufflevector <4 x i1>{{.*}}<4 x i32> <i32 3, i32 2, i32 1, i32 0>
;AVX2: call <4 x double> @llvm.masked.load.v4f64.p0v4f64
;AVX2: fadd <4 x double>
;AVX2: call void @llvm.masked.store.v4f64.p0v4f64
;AVX2: icmp sgt <8 x i32> %reverse, zeroinitializer
;AVX2: shufflevector <8 x i1>{{.*}}<8 x i32> <i32 7, i32 6, i32 5, i32 4
;AVX2: call <8 x double> @llvm.masked.load.v8f64.p0v8f64
;AVX2: fadd <8 x double>
;AVX2: call void @llvm.masked.store.v8f64.p0v8f64
;AVX2: ret void
;AVX512-LABEL: @foo6
;AVX512: icmp sgt <8 x i32> %reverse, zeroinitializer
;AVX512: shufflevector <8 x i1>{{.*}}<8 x i32> <i32 7, i32 6, i32 5, i32 4
;AVX512: call <8 x double> @llvm.masked.load.v8f64.p0v8f64
;AVX512: fadd <8 x double>
;AVX512: call void @llvm.masked.store.v8f64.p0v8f64
;AVX512: icmp sgt <16 x i32> %reverse, zeroinitializer
;AVX512: shufflevector <16 x i1>{{.*}}<16 x i32> <i32 15, i32 14, i32 13, i32 12
;AVX512: call <16 x double> @llvm.masked.load.v16f64.p0v16f64
;AVX512: fadd <16 x double>
;AVX512: call void @llvm.masked.store.v16f64.p0v16f64
;AVX512: ret void
@ -582,8 +590,8 @@ for.end: ; preds = %for.cond
; }
;AVX512-LABEL: @foo7
;AVX512: call <8 x double*> @llvm.masked.load.v8p0f64.p0v8p0f64(<8 x double*>*
;AVX512: call void @llvm.masked.store.v8f64.p0v8f64
;AVX512: call <64 x double*> @llvm.masked.load.v64p0f64.p0v64p0f64(<64 x double*>*
;AVX512: call void @llvm.masked.store.v64f64.p0v64f64
;AVX512: ret void
define void @foo7(double* noalias %out, double** noalias %in, i8* noalias %trigger, i32 %size) #0 {
@ -654,8 +662,8 @@ for.end: ; preds = %for.cond
;}
;AVX512-LABEL: @foo8
;AVX512: call <8 x i32 ()*> @llvm.masked.load.v8p0f_i32f.p0v8p0f_i32f(<8 x i32 ()*>* %
;AVX512: call void @llvm.masked.store.v8f64.p0v8f64
;AVX512: call <64 x i32 ()*> @llvm.masked.load.v64p0f_i32f.p0v64p0f_i32f(<64 x i32 ()*>* %
;AVX512: call void @llvm.masked.store.v64f64.p0v64f64
;AVX512: ret void
define void @foo8(double* noalias %out, i32 ()** noalias %in, i8* noalias %trigger, i32 %size) #0 {

2
llvm/test/Transforms/LoopVectorize/X86/no_fpmath.ll
View File

@ -2,7 +2,7 @@
; CHECK: remark: no_fpmath.c:6:11: loop not vectorized: cannot prove it is safe to reorder floating-point operations
; CHECK: remark: no_fpmath.c:6:14: loop not vectorized
; CHECK: remark: no_fpmath.c:17:14: vectorized loop (vectorization width: 2, interleaved count: 2)
; CHECK: remark: no_fpmath.c:17:14: vectorized loop (vectorization width: 4, interleaved count: 2)
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.10.0"

2
llvm/test/Transforms/LoopVectorize/X86/no_fpmath_with_hotness.ll
View File

@ -3,7 +3,7 @@
; CHECK: remark: no_fpmath.c:6:11: loop not vectorized: cannot prove it is safe to reorder floating-point operations (hotness: 300)
; CHECK: remark: no_fpmath.c:6:14: loop not vectorized
; CHECK: remark: no_fpmath.c:17:14: vectorized loop (vectorization width: 2, interleaved count: 2) (hotness: 300)
; CHECK: remark: no_fpmath.c:17:14: vectorized loop (vectorization width: 4, interleaved count: 2) (hotness: 300)
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-apple-macosx10.10.0"

2
llvm/test/Transforms/LoopVectorize/X86/reduction-crash.ll
View File

@ -7,7 +7,7 @@ target triple = "i386-apple-darwin"
define void @test1(float* nocapture %arg, i32 %arg1) nounwind {
; CHECK-LABEL: @test1(
; CHECK: preheader
; CHECK: insertelement <2 x double> zeroinitializer, double %tmp, i32 0
; CHECK: insertelement <4 x double> zeroinitializer, double %tmp, i32 0
; CHECK: vector.memcheck
bb:

2
llvm/test/Transforms/LoopVectorize/X86/vectorization-remarks-loopid-dbg.ll
View File

@ -6,7 +6,7 @@
; DEBUG-OUTPUT-NOT: .loc
; DEBUG-OUTPUT-NOT: {{.*}}.debug_info
; VECTORIZED: remark: vectorization-remarks.c:17:8: vectorized loop (vectorization width: 4, interleaved count: 1)
; VECTORIZED: remark: vectorization-remarks.c:17:8: vectorized loop (vectorization width: 16, interleaved count: 1)
; UNROLLED: remark: vectorization-remarks.c:17:8: interleaved loop (interleaved count: 4)
; NONE: remark: vectorization-remarks.c:17:8: loop not vectorized: vectorization and interleaving are explicitly disabled, or vectorize width and interleave count are both set to 1

2
llvm/test/Transforms/LoopVectorize/X86/vectorization-remarks.ll
View File

@ -6,7 +6,7 @@
; DEBUG-OUTPUT-NOT: .loc
; DEBUG-OUTPUT-NOT: {{.*}}.debug_info
; VECTORIZED: remark: vectorization-remarks.c:17:8: vectorized loop (vectorization width: 4, interleaved count: 1)
; VECTORIZED: remark: vectorization-remarks.c:17:8: vectorized loop (vectorization width: 16, interleaved count: 1)
; UNROLLED: remark: vectorization-remarks.c:17:8: interleaved loop (interleaved count: 4)
; NONE: remark: vectorization-remarks.c:17:8: loop not vectorized: vectorization and interleaving are explicitly disabled, or vectorize width and interleave count are both set to 1