replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[TargetLibraryInfo] add mappings from LLVM sin/cos intrinsics to SVML calls 

These weren't included in D19544 - probably just an oversight.
D40044 made it more likely that we'll have LLVM math intrinsics rather 
than libcalls, so this bug was more easily exposed.
As the tests/code show, we already have the complete mappings for pow/exp/log.

I don't have any experience with SVML, so I don't know if anything else is 
missing. It's also not clear to me that we should be doing this transform in 
IR rather than DAG/isel, but that's a separate issue.

Differential Revision: https://reviews.llvm.org/D47610

llvm-svn: 334211
This commit is contained in:
Sanjay Patel 2018-06-07 18:21:24 +00:00
parent c3bc63e54b
commit 4b1205b40f
2 changed files with 20 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
llvm/lib/Analysis/TargetLibraryInfo.cpp
View File

@ -1461,6 +1461,14 @@ void TargetLibraryInfoImpl::addVectorizableFunctionsFromVecLib(
{"sinf", "__svml_sinf8", 8},
{"sinf", "__svml_sinf16", 16},
{"llvm.sin.f64", "__svml_sin2", 2},
{"llvm.sin.f64", "__svml_sin4", 4},
{"llvm.sin.f64", "__svml_sin8", 8},
{"llvm.sin.f32", "__svml_sinf4", 4},
{"llvm.sin.f32", "__svml_sinf8", 8},
{"llvm.sin.f32", "__svml_sinf16", 16},
{"cos", "__svml_cos2", 2},
{"cos", "__svml_cos4", 4},
{"cos", "__svml_cos8", 8},
@ -1469,6 +1477,14 @@ void TargetLibraryInfoImpl::addVectorizableFunctionsFromVecLib(
{"cosf", "__svml_cosf8", 8},
{"cosf", "__svml_cosf16", 16},
{"llvm.cos.f64", "__svml_cos2", 2},
{"llvm.cos.f64", "__svml_cos4", 4},
{"llvm.cos.f64", "__svml_cos8", 8},
{"llvm.cos.f32", "__svml_cosf4", 4},
{"llvm.cos.f32", "__svml_cosf8", 8},
{"llvm.cos.f32", "__svml_cosf16", 16},
{"pow", "__svml_pow2", 2},
{"pow", "__svml_pow4", 4},
{"pow", "__svml_pow8", 8},

8
llvm/test/Transforms/LoopVectorize/X86/svml-calls.ll
View File

@ -77,7 +77,7 @@ for.end:
define void @sin_f64_intrinsic(double* nocapture %varray) {
; CHECK-LABEL: @sin_f64_intrinsic(
; CHECK: [[TMP5:%.*]] = call <4 x double> @llvm.sin.v4f64(<4 x double> [[TMP4:%.*]])
; CHECK: [[TMP5:%.*]] = call <4 x double> @__svml_sin4(<4 x double> [[TMP4:%.*]])
; CHECK: ret void
;
entry:
@ -100,7 +100,7 @@ for.end:
define void @sin_f32_intrinsic(float* nocapture %varray) {
; CHECK-LABEL: @sin_f32_intrinsic(
; CHECK: [[TMP5:%.*]] = call <4 x float> @llvm.sin.v4f32(<4 x float> [[TMP4:%.*]])
; CHECK: [[TMP5:%.*]] = call <4 x float> @__svml_sinf4(<4 x float> [[TMP4:%.*]])
; CHECK: ret void
;
entry:
@ -169,7 +169,7 @@ for.end:
define void @cos_f64_intrinsic(double* nocapture %varray) {
; CHECK-LABEL: @cos_f64_intrinsic(
; CHECK: [[TMP5:%.*]] = call <4 x double> @llvm.cos.v4f64(<4 x double> [[TMP4:%.*]])
; CHECK: [[TMP5:%.*]] = call <4 x double> @__svml_cos4(<4 x double> [[TMP4:%.*]])
; CHECK: ret void
;
entry:
@ -192,7 +192,7 @@ for.end:
define void @cos_f32_intrinsic(float* nocapture %varray) {
; CHECK-LABEL: @cos_f32_intrinsic(
; CHECK: [[TMP5:%.*]] = call <4 x float> @llvm.cos.v4f32(<4 x float> [[TMP4:%.*]])
; CHECK: [[TMP5:%.*]] = call <4 x float> @__svml_cosf4(<4 x float> [[TMP4:%.*]])
; CHECK: ret void
;
entry: