Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Fix scalar overload name constructed by ReplaceWithVeclib.cpp #111095

Open
wants to merge 6 commits into
base: main
Choose a base branch
from
+31 −23
Open

Fix scalar overload name constructed by ReplaceWithVeclib.cpp #111095

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
8227c91
Fix scalar overload name constructed by ReplaceWithVeclib.cpp
tex3d Oct 4, 2024
f0557de
Simplify logic a bit, handle scalar return overload type.
tex3d Oct 7, 2024
2c70285
Skip not_intrinsic; skip void return for OloadTys; update tests
tex3d Oct 7, 2024
5e92c3d
Rework logic to preserve "scalar" operand/return type, even if vector
tex3d Oct 9, 2024
9adc046
fix formatting
tex3d Oct 9, 2024
6a07812
Revert change to use isVectorIntrinsicWithScalarOpAtArg for return type
tex3d Oct 9, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 15 additions & 1 deletion llvm/lib/CodeGen/ReplaceWithVeclib.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,8 +108,22 @@ static bool replaceWithCallToVeclib(const TargetLibraryInfo &TLI,
// all vector operands match the previously found EC.
SmallVector<Type *, 8> ScalarArgTypes;
Intrinsic::ID IID = II->getIntrinsicID();

// OloadTys collects types used in scalar intrinsic overload name.
SmallVector<Type *, 3> OloadTys;
if (VTy && isVectorIntrinsicWithOverloadTypeAtArg(IID, -1))
OloadTys.push_back(VTy->getElementType());

for (auto Arg : enumerate(II->args())) {
auto *ArgTy = Arg.value()->getType();
// Gather type if it is used in the overload name.
if (isVectorIntrinsicWithOverloadTypeAtArg(IID, Arg.index())) {
tex3d marked this conversation as resolved.
Show resolved Hide resolved
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

are there any tests, testing this logic? I think the pow case is working thanks to the condition added to line 114

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The pow case is impacted by both this logic and return type logic, since OloadTys should include only the return type for pow, not each of the argument types. isVectorIntrinsicWithOverloadTypeAtArg() is how you determine the arguments that the intrinsic is overloaded on, when not void, and defaults to -1 (return type).

Other tests will be running through this logic (and turning out the same), but the pow one is the only case with two arguments which had existing tests meant to match a vector lib call right now. atan2 will be added soon, which also needs this. Other intrinsics could require this change in theory, if there's a matching vector lib call for them.

if (!isVectorIntrinsicWithScalarOpAtArg(IID, Arg.index()) && isa<VectorType>(ArgTy))
OloadTys.push_back(cast<VectorType>(ArgTy)->getElementType());
else
OloadTys.push_back(ArgTy);
}

if (isVectorIntrinsicWithScalarOpAtArg(IID, Arg.index())) {
ScalarArgTypes.push_back(ArgTy);
} else if (auto *VectorArgTy = dyn_cast<VectorType>(ArgTy)) {
Expand All @@ -129,7 +143,7 @@ static bool replaceWithCallToVeclib(const TargetLibraryInfo &TLI,
// using scalar argument types.
std::string ScalarName =
Intrinsic::isOverloaded(IID)
? Intrinsic::getName(IID, ScalarArgTypes, II->getModule())
farzonl marked this conversation as resolved.
Show resolved Hide resolved
? Intrinsic::getName(IID, OloadTys, II->getModule())
: Intrinsic::getName(IID).str();

// Try to find the mapping for the scalar version of this intrinsic and the
Expand Down
17 changes: 5 additions & 12 deletions llvm/test/CodeGen/AArch64/replace-with-veclib-armpl.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ declare <vscale x 2 x double> @llvm.cos.nxv2f64(<vscale x 2 x double>)
declare <vscale x 4 x float> @llvm.cos.nxv4f32(<vscale x 4 x float>)

;.
; CHECK: @llvm.compiler.used = appending global [60 x ptr] [ptr @armpl_vcosq_f64, ptr @armpl_vcosq_f32, ptr @armpl_svcos_f64_x, ptr @armpl_svcos_f32_x, ptr @armpl_vexpq_f64, ptr @armpl_vexpq_f32, ptr @armpl_svexp_f64_x, ptr @armpl_svexp_f32_x, ptr @armpl_vexp10q_f64, ptr @armpl_vexp10q_f32, ptr @armpl_svexp10_f64_x, ptr @armpl_svexp10_f32_x, ptr @armpl_vexp2q_f64, ptr @armpl_vexp2q_f32, ptr @armpl_svexp2_f64_x, ptr @armpl_svexp2_f32_x, ptr @armpl_vlogq_f64, ptr @armpl_vlogq_f32, ptr @armpl_svlog_f64_x, ptr @armpl_svlog_f32_x, ptr @armpl_vlog10q_f64, ptr @armpl_vlog10q_f32, ptr @armpl_svlog10_f64_x, ptr @armpl_svlog10_f32_x, ptr @armpl_vlog2q_f64, ptr @armpl_vlog2q_f32, ptr @armpl_svlog2_f64_x, ptr @armpl_svlog2_f32_x, ptr @armpl_vsinq_f64, ptr @armpl_vsinq_f32, ptr @armpl_svsin_f64_x, ptr @armpl_svsin_f32_x, ptr @armpl_vtanq_f64, ptr @armpl_vtanq_f32, ptr @armpl_svtan_f64_x, ptr @armpl_svtan_f32_x, ptr @armpl_vacosq_f64, ptr @armpl_vacosq_f32, ptr @armpl_svacos_f64_x, ptr @armpl_svacos_f32_x, ptr @armpl_vasinq_f64, ptr @armpl_vasinq_f32, ptr @armpl_svasin_f64_x, ptr @armpl_svasin_f32_x, ptr @armpl_vatanq_f64, ptr @armpl_vatanq_f32, ptr @armpl_svatan_f64_x, ptr @armpl_svatan_f32_x, ptr @armpl_vcoshq_f64, ptr @armpl_vcoshq_f32, ptr @armpl_svcosh_f64_x, ptr @armpl_svcosh_f32_x, ptr @armpl_vsinhq_f64, ptr @armpl_vsinhq_f32, ptr @armpl_svsinh_f64_x, ptr @armpl_svsinh_f32_x, ptr @armpl_vtanhq_f64, ptr @armpl_vtanhq_f32, ptr @armpl_svtanh_f64_x, ptr @armpl_svtanh_f32_x], section "llvm.metadata"
; CHECK: @llvm.compiler.used = appending global [64 x ptr] [ptr @armpl_vcosq_f64, ptr @armpl_vcosq_f32, ptr @armpl_svcos_f64_x, ptr @armpl_svcos_f32_x, ptr @armpl_vexpq_f64, ptr @armpl_vexpq_f32, ptr @armpl_svexp_f64_x, ptr @armpl_svexp_f32_x, ptr @armpl_vexp10q_f64, ptr @armpl_vexp10q_f32, ptr @armpl_svexp10_f64_x, ptr @armpl_svexp10_f32_x, ptr @armpl_vexp2q_f64, ptr @armpl_vexp2q_f32, ptr @armpl_svexp2_f64_x, ptr @armpl_svexp2_f32_x, ptr @armpl_vlogq_f64, ptr @armpl_vlogq_f32, ptr @armpl_svlog_f64_x, ptr @armpl_svlog_f32_x, ptr @armpl_vlog10q_f64, ptr @armpl_vlog10q_f32, ptr @armpl_svlog10_f64_x, ptr @armpl_svlog10_f32_x, ptr @armpl_vlog2q_f64, ptr @armpl_vlog2q_f32, ptr @armpl_svlog2_f64_x, ptr @armpl_svlog2_f32_x, ptr @armpl_vpowq_f64, ptr @armpl_vpowq_f32, ptr @armpl_svpow_f64_x, ptr @armpl_svpow_f32_x, ptr @armpl_vsinq_f64, ptr @armpl_vsinq_f32, ptr @armpl_svsin_f64_x, ptr @armpl_svsin_f32_x, ptr @armpl_vtanq_f64, ptr @armpl_vtanq_f32, ptr @armpl_svtan_f64_x, ptr @armpl_svtan_f32_x, ptr @armpl_vacosq_f64, ptr @armpl_vacosq_f32, ptr @armpl_svacos_f64_x, ptr @armpl_svacos_f32_x, ptr @armpl_vasinq_f64, ptr @armpl_vasinq_f32, ptr @armpl_svasin_f64_x, ptr @armpl_svasin_f32_x, ptr @armpl_vatanq_f64, ptr @armpl_vatanq_f32, ptr @armpl_svatan_f64_x, ptr @armpl_svatan_f32_x, ptr @armpl_vcoshq_f64, ptr @armpl_vcoshq_f32, ptr @armpl_svcosh_f64_x, ptr @armpl_svcosh_f32_x, ptr @armpl_vsinhq_f64, ptr @armpl_vsinhq_f32, ptr @armpl_svsinh_f64_x, ptr @armpl_svsinh_f32_x, ptr @armpl_vtanhq_f64, ptr @armpl_vtanhq_f32, ptr @armpl_svtanh_f64_x, ptr @armpl_svtanh_f32_x], section "llvm.metadata"

;.
define <2 x double> @llvm_cos_f64(<2 x double> %in) {
Expand Down Expand Up @@ -333,17 +333,10 @@ declare <4 x float> @llvm.pow.v4f32(<4 x float>, <4 x float>)
declare <vscale x 2 x double> @llvm.pow.nxv2f64(<vscale x 2 x double>, <vscale x 2 x double>)
declare <vscale x 4 x float> @llvm.pow.nxv4f32(<vscale x 4 x float>, <vscale x 4 x float>)

;
; There is a bug in the replace-with-veclib pass, and for intrinsics which take
; more than one arguments, but has just one overloaded type, it incorrectly
; reconstructs the scalar name, for pow specifically it is searching for:
; llvm.pow.f64.f64 and llvm.pow.f32.f32
;

define <2 x double> @llvm_pow_f64(<2 x double> %in, <2 x double> %power) {
; CHECK-LABEL: define <2 x double> @llvm_pow_f64
; CHECK-SAME: (<2 x double> [[IN:%.*]], <2 x double> [[POWER:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call fast <2 x double> @llvm.pow.v2f64(<2 x double> [[IN]], <2 x double> [[POWER]])
; CHECK-NEXT: [[TMP1:%.*]] = call fast <2 x double> @armpl_vpowq_f64(<2 x double> [[IN]], <2 x double> [[POWER]])
; CHECK-NEXT: ret <2 x double> [[TMP1]]
;
%1 = call fast <2 x double> @llvm.pow.v2f64(<2 x double> %in, <2 x double> %power)
Expand All @@ -353,7 +346,7 @@ define <2 x double> @llvm_pow_f64(<2 x double> %in, <2 x double> %power) {
define <4 x float> @llvm_pow_f32(<4 x float> %in, <4 x float> %power) {
; CHECK-LABEL: define <4 x float> @llvm_pow_f32
; CHECK-SAME: (<4 x float> [[IN:%.*]], <4 x float> [[POWER:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call fast <4 x float> @llvm.pow.v4f32(<4 x float> [[IN]], <4 x float> [[POWER]])
; CHECK-NEXT: [[TMP1:%.*]] = call fast <4 x float> @armpl_vpowq_f32(<4 x float> [[IN]], <4 x float> [[POWER]])
; CHECK-NEXT: ret <4 x float> [[TMP1]]
;
%1 = call fast <4 x float> @llvm.pow.v4f32(<4 x float> %in, <4 x float> %power)
Expand All @@ -363,7 +356,7 @@ define <4 x float> @llvm_pow_f32(<4 x float> %in, <4 x float> %power) {
define <vscale x 2 x double> @llvm_pow_vscale_f64(<vscale x 2 x double> %in, <vscale x 2 x double> %power) #0 {
; CHECK-LABEL: define <vscale x 2 x double> @llvm_pow_vscale_f64
; CHECK-SAME: (<vscale x 2 x double> [[IN:%.*]], <vscale x 2 x double> [[POWER:%.*]]) #[[ATTR1]] {
; CHECK-NEXT: [[TMP1:%.*]] = call fast <vscale x 2 x double> @llvm.pow.nxv2f64(<vscale x 2 x double> [[IN]], <vscale x 2 x double> [[POWER]])
; CHECK-NEXT: [[TMP1:%.*]] = call fast <vscale x 2 x double> @armpl_svpow_f64_x(<vscale x 2 x double> [[IN]], <vscale x 2 x double> [[POWER]], <vscale x 2 x i1> shufflevector (<vscale x 2 x i1> insertelement (<vscale x 2 x i1> poison, i1 true, i64 0), <vscale x 2 x i1> poison, <vscale x 2 x i32> zeroinitializer))
; CHECK-NEXT: ret <vscale x 2 x double> [[TMP1]]
;
%1 = call fast <vscale x 2 x double> @llvm.pow.nxv2f64(<vscale x 2 x double> %in, <vscale x 2 x double> %power)
Expand All @@ -373,7 +366,7 @@ define <vscale x 2 x double> @llvm_pow_vscale_f64(<vscale x 2 x double> %in, <vs
define <vscale x 4 x float> @llvm_pow_vscale_f32(<vscale x 4 x float> %in, <vscale x 4 x float> %power) #0 {
; CHECK-LABEL: define <vscale x 4 x float> @llvm_pow_vscale_f32
; CHECK-SAME: (<vscale x 4 x float> [[IN:%.*]], <vscale x 4 x float> [[POWER:%.*]]) #[[ATTR1]] {
; CHECK-NEXT: [[TMP1:%.*]] = call fast <vscale x 4 x float> @llvm.pow.nxv4f32(<vscale x 4 x float> [[IN]], <vscale x 4 x float> [[POWER]])
; CHECK-NEXT: [[TMP1:%.*]] = call fast <vscale x 4 x float> @armpl_svpow_f32_x(<vscale x 4 x float> [[IN]], <vscale x 4 x float> [[POWER]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer))
; CHECK-NEXT: ret <vscale x 4 x float> [[TMP1]]
;
%1 = call fast <vscale x 4 x float> @llvm.pow.nxv4f32(<vscale x 4 x float> %in, <vscale x 4 x float> %power)
Expand Down
Loading