[llvm-branch-commits] [llvm] 202d359 - [X86] Add the FSRM feature (Fast Short Rep Mov) to Zen3.
Author: Hiroshi Yamauchi
Date: 2021-01-14T10:47:33-08:00
New Revision: 202d359753d1f130a228c3ad52dfaabf384250d1
URL:
https://github.com/llvm/llvm-project/commit/202d359753d1f130a228c3ad52dfaabf384250d1
DIFF:
https://github.com/llvm/llvm-project/commit/202d359753d1f130a228c3ad52dfaabf384250d1.diff
LOG: [X86] Add the FSRM feature (Fast Short Rep Mov) to Zen3.
Note -x86-use-fsrm-for-memcpy is still disabled by default and there's no
default behavior change.
Differential Revision: https://reviews.llvm.org/D94436
Added:
Modified:
llvm/lib/Target/X86/X86.td
llvm/test/CodeGen/X86/memcpy-inline-fsrm.ll
Removed:
diff --git a/llvm/lib/Target/X86/X86.td b/llvm/lib/Target/X86/X86.td
index 9096d9d54452..c492d686c52e 100644
--- a/llvm/lib/Target/X86/X86.td
+++ b/llvm/lib/Target/X86/X86.td
@@ -1071,7 +1071,8 @@ def ProcessorFeatures {
list ZN2Tuning = ZNTuning;
list ZN2Features =
!listconcat(ZNFeatures, ZN2AdditionalFeatures);
- list ZN3AdditionalFeatures = [FeatureINVPCID,
+ list ZN3AdditionalFeatures = [FeatureFSRM,
+ FeatureINVPCID,
FeaturePKU,
FeatureVAES,
FeatureVPCLMULQDQ];
diff --git a/llvm/test/CodeGen/X86/memcpy-inline-fsrm.ll
b/llvm/test/CodeGen/X86/memcpy-inline-fsrm.ll
index 9480d74723fc..77e97626b1c6 100644
--- a/llvm/test/CodeGen/X86/memcpy-inline-fsrm.ll
+++ b/llvm/test/CodeGen/X86/memcpy-inline-fsrm.ll
@@ -4,6 +4,7 @@
; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mcpu=haswell <
%s | FileCheck %s --check-prefix=NOFSRM
; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy
-mcpu=icelake-client < %s | FileCheck %s --check-prefix=FSRM
; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy
-mcpu=icelake-server < %s | FileCheck %s --check-prefix=FSRM
+; RUN: llc -mtriple=x86_64-linux-gnu -x86-use-fsrm-for-memcpy -mcpu=znver3 <
%s | FileCheck %s --check-prefix=FSRM
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture, i64, i1)
nounwind
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] cf5415c - [PGO][PGSO] Let unroll hints take precedence over PGSO.
Author: Hiroshi Yamauchi
Date: 2021-01-07T10:10:31-08:00
New Revision: cf5415c727dda0ea4b27ee16347d170f118b037b
URL:
https://github.com/llvm/llvm-project/commit/cf5415c727dda0ea4b27ee16347d170f118b037b
DIFF:
https://github.com/llvm/llvm-project/commit/cf5415c727dda0ea4b27ee16347d170f118b037b.diff
LOG: [PGO][PGSO] Let unroll hints take precedence over PGSO.
Differential Revision: https://reviews.llvm.org/D94199
Added:
Modified:
llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
Removed:
diff --git a/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
b/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
index 4cce05d595a8..d09f1ee22a75 100644
--- a/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
+++ b/llvm/lib/Transforms/Scalar/LoopUnrollPass.cpp
@@ -218,8 +218,10 @@ TargetTransformInfo::UnrollingPreferences
llvm::gatherUnrollingPreferences(
// Apply size attributes
bool OptForSize = L->getHeader()->getParent()->hasOptSize() ||
-llvm::shouldOptimizeForSize(L->getHeader(), PSI, BFI,
-PGSOQueryType::IRPass);
+// Let unroll hints / pragmas take precedence over PGSO.
+(hasUnrollTransformation(L) != TM_ForcedByUser &&
+ llvm::shouldOptimizeForSize(L->getHeader(), PSI, BFI,
+ PGSOQueryType::IRPass));
if (OptForSize) {
UP.Threshold = UP.OptSizeThreshold;
UP.PartialThreshold = UP.PartialOptSizeThreshold;
diff --git a/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
b/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
index e219349fb093..884981bf8bf0 100644
--- a/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
+++ b/llvm/test/Transforms/LoopUnroll/unroll-opt-attribute.ll
@@ -158,6 +158,38 @@ for.end: ; preds
= %for.body
; NPGSO-NOT: phi
; NPGSO-NOT: icmp
+;/ TEST 6 //
+
+; This test tests that unroll hints take precedence over PGSO and that this
loop
+; gets unrolled even though it's cold.
+
+define i32 @Test6() !prof !14 {
+entry:
+ br label %for.body
+
+for.body: ; preds = %for.body, %entry
+ %i.05 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
+ %arrayidx = getelementptr inbounds [24 x i32], [24 x i32]* @tab, i32 0, i32
%i.05
+ store i32 %i.05, i32* %arrayidx, align 4
+ %inc = add nuw nsw i32 %i.05, 1
+ %exitcond = icmp eq i32 %inc, 24
+ br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !15
+
+for.end: ; preds = %for.body
+ ret i32 42
+}
+
+; PGSO-LABEL: @Test6
+; PGSO: store
+; PGSO: store
+; PGSO: store
+; PGSO: store
+; NPGSO-LABEL: @Test6
+; NPGSO: store
+; NPGSO: store
+; NPGSO: store
+; NPGSO: store
+
!llvm.module.flags = !{!0}
!0 = !{i32 1, !"ProfileSummary", !1}
!1 = !{!2, !3, !4, !5, !6, !7, !8, !9}
@@ -174,3 +206,5 @@ for.end: ; preds =
%for.body
!12 = !{i32 999000, i64 100, i32 1}
!13 = !{i32 99, i64 1, i32 2}
!14 = !{!"function_entry_count", i64 0}
+!15 = !{!15, !16}
+!16 = !{!"llvm.loop.unroll.count", i32 4}
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] f9c3954 - Fix for Bug 48055.
Author: Hiroshi Yamauchi
Date: 2020-12-04T11:05:01-08:00
New Revision: f9c3954a6ec5d1066a67aadab848f02b9b78b056
URL:
https://github.com/llvm/llvm-project/commit/f9c3954a6ec5d1066a67aadab848f02b9b78b056
DIFF:
https://github.com/llvm/llvm-project/commit/f9c3954a6ec5d1066a67aadab848f02b9b78b056.diff
LOG: Fix for Bug 48055.
Differential Revision: https://reviews.llvm.org/D92599
Added:
llvm/test/Transforms/GlobalOpt/evaluate-bitcast-4.ll
Modified:
llvm/lib/Transforms/Utils/Evaluator.cpp
Removed:
diff --git a/llvm/lib/Transforms/Utils/Evaluator.cpp
b/llvm/lib/Transforms/Utils/Evaluator.cpp
index 6fe29381c71a..a01dc7a4e609 100644
--- a/llvm/lib/Transforms/Utils/Evaluator.cpp
+++ b/llvm/lib/Transforms/Utils/Evaluator.cpp
@@ -183,11 +183,11 @@ evaluateBitcastFromPtr(Constant *Ptr, const DataLayout
&DL,
std::function Func) {
Constant *Val;
while (!(Val = Func(Ptr))) {
-// If Ty is a struct, we can convert the pointer to the struct
+// If Ty is a non-opaque struct, we can convert the pointer to the struct
// into a pointer to its first member.
// FIXME: This could be extended to support arrays as well.
Type *Ty = cast(Ptr->getType())->getElementType();
-if (!isa(Ty))
+if (!isa(Ty) || cast(Ty)->isOpaque())
break;
IntegerType *IdxTy = IntegerType::get(Ty->getContext(), 32);
diff --git a/llvm/test/Transforms/GlobalOpt/evaluate-bitcast-4.ll
b/llvm/test/Transforms/GlobalOpt/evaluate-bitcast-4.ll
new file mode 100644
index ..1e7c89894d2a
--- /dev/null
+++ b/llvm/test/Transforms/GlobalOpt/evaluate-bitcast-4.ll
@@ -0,0 +1,29 @@
+; PR48055. Check that this does not crash.
+; RUN: opt -globalopt %s -disable-output
+target datalayout =
"e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+%struct.g = type opaque
+%struct.a = type { i32 (...)** }
+
+@l = dso_local global i32 0, align 4
+@h = external dso_local global %struct.g, align 1
+@llvm.global_ctors = appending global [1 x { i32, void ()*, i8* }] [{ i32,
void ()*, i8* } { i32 65535, void ()* @_GLOBAL__sub_I_bug48055.cc, i8* null }]
+
+; Function Attrs: uwtable
+define internal void @__cxx_global_var_init() {
+entry:
+ %vtable = load i32 (%struct.a*)**, i32 (%struct.a*)*** bitcast (%struct.g*
@h to i32 (%struct.a*)***), align 1
+ %0 = load i32 (%struct.a*)*, i32 (%struct.a*)** %vtable, align 8
+ %call = call i32 %0(%struct.a* nonnull dereferenceable(8) bitcast
(%struct.g* @h to %struct.a*))
+ store i32 %call, i32* @l, align 4
+ ret void
+}
+
+; Function Attrs: uwtable
+define internal void @_GLOBAL__sub_I_bug48055.cc() {
+entry:
+ call void @__cxx_global_var_init()
+ ret void
+}
+
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
[llvm-branch-commits] [llvm] a5f5612 - [PGO] Adjust -vp-counters-per-site under dynamic linking.
Author: Hiroshi Yamauchi
Date: 2020-12-11T09:42:53-08:00
New Revision: a5f5612263ca650ae0aa433539278c2d5f567cf8
URL:
https://github.com/llvm/llvm-project/commit/a5f5612263ca650ae0aa433539278c2d5f567cf8
DIFF:
https://github.com/llvm/llvm-project/commit/a5f5612263ca650ae0aa433539278c2d5f567cf8.diff
LOG: [PGO] Adjust -vp-counters-per-site under dynamic linking.
Addressing clang bootstrap under the dynamic linking mode running out of static
allocation of value profile nodes, reported in D81682.
Differential Revision: https://reviews.llvm.org/D92669
Added:
Modified:
llvm/cmake/modules/HandleLLVMOptions.cmake
Removed:
diff --git a/llvm/cmake/modules/HandleLLVMOptions.cmake
b/llvm/cmake/modules/HandleLLVMOptions.cmake
index 78ed1c06ac9d..f313492ba978 100644
--- a/llvm/cmake/modules/HandleLLVMOptions.cmake
+++ b/llvm/cmake/modules/HandleLLVMOptions.cmake
@@ -917,6 +917,15 @@ if (LLVM_BUILD_INSTRUMENTED)
CMAKE_EXE_LINKER_FLAGS
CMAKE_SHARED_LINKER_FLAGS)
endif()
+# Set this to avoid running out of the value profile node section
+# under clang in dynamic linking mode.
+if (CMAKE_CXX_COMPILER_ID MATCHES "Clang" AND
+CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 11 AND
+LLVM_LINK_LLVM_DYLIB)
+ append("-Xclang -mllvm -Xclang -vp-counters-per-site=1.5"
+CMAKE_CXX_FLAGS
+CMAKE_C_FLAGS)
+endif()
elseif(uppercase_LLVM_BUILD_INSTRUMENTED STREQUAL "CSIR")
append("-fcs-profile-generate=\"${LLVM_CSPROFILE_DATA_DIR}\""
CMAKE_CXX_FLAGS
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
