[PATCH] D47673: [Coroutines] Less IR for noexcept await_resume
modocache created this revision.
modocache added reviewers: GorNishanov, EricWF.
Herald added a subscriber: cfe-commits.
In his review of https://reviews.llvm.org/D45860, @GorNishanov suggested
avoiding generating additional exception-handling IR in the case that
the resume function was marked as 'noexcept', and exceptions could not
occur. This implements that suggestion.
Test Plan: `check-clang`
Repository:
rC Clang
https://reviews.llvm.org/D47673
Files:
lib/CodeGen/CGCoroutine.cpp
test/CodeGenCoroutines/coro-await-resume-eh.cpp
Index: test/CodeGenCoroutines/coro-await-resume-eh.cpp
===
--- test/CodeGenCoroutines/coro-await-resume-eh.cpp
+++ test/CodeGenCoroutines/coro-await-resume-eh.cpp
@@ -18,18 +18,18 @@
void await_resume() { throw 42; }
};
-struct task {
+struct throwing_task {
struct promise_type {
-task get_return_object() { return task{}; }
+auto get_return_object() { return throwing_task{}; }
auto initial_suspend() { return throwing_awaitable{}; }
auto final_suspend() { return coro::suspend_never{}; }
void return_void() {}
void unhandled_exception() {}
};
};
// CHECK-LABEL: define void @_Z1fv()
-task f() {
+throwing_task f() {
// A variable RESUMETHREW is used to keep track of whether the body
// of 'await_resume' threw an exception. Exceptions thrown in
// 'await_resume' are unwound to RESUMELPAD.
@@ -50,7 +50,7 @@
// CHECK: [[RESUMELPAD]]:
// CHECK: br label %[[RESUMECATCH:.+]]
// CHECK: [[RESUMECATCH]]:
- // CHECK: invoke void @_ZN4task12promise_type19unhandled_exceptionEv
+ // CHECK: invoke void @_ZN13throwing_task12promise_type19unhandled_exceptionEv
// CHECK-NEXT: to label %[[RESUMEENDCATCH:.+]] unwind label
// CHECK: [[RESUMEENDCATCH]]:
// CHECK-NEXT: invoke void @__cxa_end_catch()
@@ -67,15 +67,42 @@
// CHECK-NEXT: br i1 %[[RESUMETHREWLOAD]], label %[[RESUMEDCONT:.+]], label %[[RESUMEDBODY:.+]]
// CHECK: [[RESUMEDBODY]]:
- // CHECK: invoke void @_ZN4task12promise_type11return_voidEv
+ // CHECK: invoke void @_ZN13throwing_task12promise_type11return_voidEv
// CHECK-NEXT: to label %[[REDUMEDBODYCONT:.+]] unwind label
// CHECK: [[REDUMEDBODYCONT]]:
// CHECK-NEXT: br label %[[COROFINAL:.+]]
// CHECK: [[RESUMEDCONT]]:
// CHECK-NEXT: br label %[[COROFINAL]]
// CHECK: [[COROFINAL]]:
- // CHECK-NEXT: invoke void @_ZN4task12promise_type13final_suspendEv
+ // CHECK-NEXT: invoke void @_ZN13throwing_task12promise_type13final_suspendEv
+ co_return;
+}
+
+struct noexcept_awaitable {
+ bool await_ready() { return true; }
+ void await_suspend(coro::coroutine_handle<>) {}
+ void await_resume() noexcept {}
+};
+
+struct noexcept_task {
+ struct promise_type {
+auto get_return_object() { return noexcept_task{}; }
+auto initial_suspend() { return noexcept_awaitable{}; }
+auto final_suspend() { return coro::suspend_never{}; }
+void return_void() {}
+void unhandled_exception() {}
+ };
+};
+
+// CHECK-LABEL: define void @_Z1gv()
+noexcept_task g() {
+ // If the await_resume function is marked as noexcept, none of the additional
+ // conditions that are present in f() above are added to the IR.
+ // This means that no i1 are stored before or after calling await_resume:
+ // CHECK: init.ready:
+ // CHECK-NEXT: call void @_ZN18noexcept_awaitable12await_resumeEv
+ // CHECK-NOT: store i1 false, i1*
co_return;
}
Index: lib/CodeGen/CGCoroutine.cpp
===
--- lib/CodeGen/CGCoroutine.cpp
+++ lib/CodeGen/CGCoroutine.cpp
@@ -217,8 +217,19 @@
// Emit await_resume expression.
CGF.EmitBlock(ReadyBlock);
+
+ // Exception handling requires additional IR. If the 'await_resume' function
+ // is marked as 'noexcept', we avoid generating this additional IR.
+ bool ResumeCanThrow = true;
+ if (const auto *MCE = dyn_cast(S.getResumeExpr()))
+if (const auto *Proto =
+MCE->getMethodDecl()->getType()->getAs())
+ if (isNoexceptExceptionSpec(Proto->getExceptionSpecType()) &&
+ Proto->canThrow() == CT_Cannot)
+ResumeCanThrow = false;
+
CXXTryStmt *TryStmt = nullptr;
- if (Coro.ExceptionHandler && Kind == AwaitKind::Init) {
+ if (Coro.ExceptionHandler && Kind == AwaitKind::Init && ResumeCanThrow) {
Coro.ResumeEHVar =
CGF.CreateTempAlloca(Builder.getInt1Ty(), Prefix + Twine("resume.eh"));
Builder.CreateFlagStore(true, Coro.ResumeEHVar);
@@ -625,12 +636,20 @@
CurCoro.Data->CurrentAwaitKind = AwaitKind::Normal;
if (CurCoro.Data->ExceptionHandler) {
- BasicBlock *BodyBB = createBasicBlock("coro.resumed.body");
- BasicBlock *ContBB = createBasicBlock("coro.resumed.cont");
- Value *SkipBody =
- Builder.CreateFlagLoad(CurCoro.Data->ResumeEHVar, "coro.resumed.eh");
- Builder.CreateCondBr(SkipBody, ContBB, BodyBB);
- EmitBlock(BodyBB);
+ // If we gene
[PATCH] D46805: If some platforms do not support an attribute, we should exclude the platform
HLJ2009 added a comment.
In https://reviews.llvm.org/D46805#1117091, @sunfish wrote:
> In https://reviews.llvm.org/D46805#1115681, @rsmith wrote:
>
> > In https://reviews.llvm.org/D46805#1113358, @aaron.ballman wrote:
> >
> > > @rsmith -- do the object file formats listed look correct to you?
> >
> >
> > They look at least plausible. We should be able to test whether LLVM can
> > actually emit aliases on each of these targets easily enough...
> >
> > However, I get this error for any WAsm compilation I try:
> >
> > fatal error: error in backend: section size does not fit in a uint32_t
> >
> > ... so I have no idea if aliases are/will be supported there. Perhaps
> > @sunfish can tell us :)
>
>
> Yes, they are intended to be supported. It sounds like you found a bug,
> though I've not been able to reproduce it in simple tests.
Comment at: include/clang/Basic/Attr.td:322
+def TargetSupportsAlias : TargetSpec {
+ let ObjectFormats = ["COFF", "ELF", "Wasm"];
+}
aaron.ballman wrote:
> Did you verify that Wasm supports the alias attribute? If it is supported, it
> might be nice to add a test to `CodeGen/alias.c` to demonstrate it. Similar
> for COFF.
Yes, I used the following command line to test my test file.
clang -cc1 -triple wasm32-unknown-unknown -fsyntax-only -verify attr-alias-has.c
clang -cc1 -triple wasm64-unknown-unknown -fsyntax-only -verify attr-alias-has.c
The test result is ok.
ok, I will update it.
Comment at: test/Sema/attr-alias-has.c:5
+// RUN: %clang_cc1 -triple wasm32-unknown-unknown -fsyntax-only -verify %s
+// RUN: %clang_cc1 -triple wasm64-unknown-unknown -fsyntax-only -verify %s
+
aaron.ballman wrote:
> I'd like to see a test that the "attribute not supported on target"
> diagnostic is being generated. I'd recommend something along these lines:
> ```
> void g() {}
> void f() __attribute__((alias("g")));
> #if !__has_attribute(alias)
> // expected-error@-2{{expected diagnostic text}}
> #else
> // expected-no-diagnostics
> #endif
> ```
ok, I will update it.
Repository:
rC Clang
https://reviews.llvm.org/D46805
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D47201: [CUDA] Implement nv_weak attribute for functions
Hahnfeld added a comment. In https://reviews.llvm.org/D47201#1119254, @aaron.ballman wrote: > In https://reviews.llvm.org/D47201#1119249, @tra wrote: > > > IIUIC, nv_weak is a synonym for weak (why, oh why did they need > > it?) > > You may need to hunt down and change few other places that deal with the > > weak attribute. > > E.g.: > > https://github.com/llvm-project/llvm-project-20170507/blob/master/clang/lib/AST/Decl.cpp#L4267 > > > > https://github.com/llvm-project/llvm-project-20170507/blob/master/clang/lib/CodeGen/ItaniumCXXABI.cpp#L3045 > > > If it is truly a synonym for weak, then a better implementation would be to > make no semantic distinction between the two attributes -- just add new > spellings to weak. If you need to make minor distinctions between the > spellings, you can do it using accessors on the attribute. I first went with this approach but then thought it would be better to restrict the new attribute as much as possible. That's why I added a completely new one which is only applicable to functions, but not to variables and `CXXRecord`s. Let me know if you'd prefer `nv_weak` to be a full alias of `weak` and I'll revert to what @aaron.ballman suggested. Repository: rC Clang https://reviews.llvm.org/D47201 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D46602: [clang-tidy] Store checks profiling info as YAML files
lebedev.ri updated this revision to Diff 149610.
lebedev.ri marked 6 inline comments as done.
lebedev.ri added a comment.
Rebased.
Addressed review notes.
Repository:
rCTE Clang Tools Extra
https://reviews.llvm.org/D46602
Files:
clang-tidy/ClangTidy.cpp
clang-tidy/ClangTidy.h
clang-tidy/ClangTidyDiagnosticConsumer.cpp
clang-tidy/ClangTidyDiagnosticConsumer.h
clang-tidy/ClangTidyProfiling.cpp
clang-tidy/ClangTidyProfiling.h
clang-tidy/tool/ClangTidyMain.cpp
docs/ReleaseNotes.rst
docs/clang-tidy/index.rst
test/clang-tidy/clang-tidy-enable-check-profile-one-tu.cpp
test/clang-tidy/clang-tidy-enable-check-profile-two-tu.cpp
test/clang-tidy/clang-tidy-store-check-profile-one-tu.cpp
Index: test/clang-tidy/clang-tidy-store-check-profile-one-tu.cpp
===
--- /dev/null
+++ test/clang-tidy/clang-tidy-store-check-profile-one-tu.cpp
@@ -0,0 +1,37 @@
+// RUN: rm -rf %T/out
+// RUN: clang-tidy -enable-check-profile -checks='-*,readability-function-size' -store-check-profile=%T/out %s 2>&1 | FileCheck --match-full-lines -implicit-check-not='{{warning:|error:}}' -check-prefix=CHECK-CONSOLE %s
+// RUN: cat %T/out/*-clang-tidy-store-check-profile-one-tu.cpp.yaml | FileCheck --match-full-lines -implicit-check-not='{{warning:|error:}}' -check-prefix=CHECK-FILE %s
+// RUN: rm -rf %T/out
+// RUN: clang-tidy -enable-check-profile -checks='-*,readability-function-size' -store-check-profile=%T/out %s 2>&1 | FileCheck --match-full-lines -implicit-check-not='{{warning:|error:}}' -check-prefix=CHECK-CONSOLE %s
+// RUN: cat %T/out/*-clang-tidy-store-check-profile-one-tu.cpp.yaml | FileCheck --match-full-lines -implicit-check-not='{{warning:|error:}}' -check-prefix=CHECK-FILE %s
+
+// CHECK-CONSOLE-NOT: ===-===
+// CHECK-CONSOLE-NOT: {{.*}} --- Name ---
+// CHECK-CONSOLE-NOT: {{.*}} readability-function-size
+// CHECK-CONSOLE-NOT: {{.*}} Total
+// CHECK-CONSOLE-NOT: ===-===
+
+// CHECK-FILE: {
+// CHECK-FILE-NEXT:"file": "{{.*}}clang-tidy-store-check-profile-one-tu.cpp",
+// CHECK-FILE-NEXT:"timestamp": "{{[0-9]+}}-{{[0-9]+}}-{{[0-9]+}} {{[0-9]+}}:{{[0-9]+}}:{{[0-9]+}}.{{[0-9]+}}",
+// CHECK-FILE-NEXT:"profile": {
+// CHECK-FILE-NEXT: "time.clang-tidy.readability-function-size.wall": {{.*}}{{[0-9]}}.{{[0-9]+}}e{{[-+]}}{{[0-9]}}{{[0-9]}},
+// CHECK-FILE-NEXT: "time.clang-tidy.readability-function-size.user": {{.*}}{{[0-9]}}.{{[0-9]+}}e{{[-+]}}{{[0-9]}}{{[0-9]}},
+// CHECK-FILE-NEXT: "time.clang-tidy.readability-function-size.sys": {{.*}}{{[0-9]}}.{{[0-9]+}}e{{[-+]}}{{[0-9]}}{{[0-9]}}
+// CHECK-FILE-NEXT: }
+// CHECK-FILE-NEXT: }
+
+// CHECK-FILE-NOT: {
+// CHECK-FILE-NOT: "file": {{.*}}clang-tidy-store-check-profile-one-tu.cpp{{.*}},
+// CHECK-FILE-NOT: "timestamp": "{{[0-9]+}}-{{[0-9]+}}-{{[0-9]+}} {{[0-9]+}}:{{[0-9]+}}:{{[0-9]+}}.{{[0-9]+}}",
+// CHECK-FILE-NOT: "profile": {
+// CHECK-FILE-NOT: "time.clang-tidy.readability-function-size.wall": {{.*}}{{[0-9]}}.{{[0-9]+}}e{{[-+]}}{{[0-9]}}{{[0-9]}},
+// CHECK-FILE-NOT: "time.clang-tidy.readability-function-size.user": {{.*}}{{[0-9]}}.{{[0-9]+}}e{{[-+]}}{{[0-9]}}{{[0-9]}},
+// CHECK-FILE-NOT: "time.clang-tidy.readability-function-size.sys": {{.*}}{{[0-9]}}.{{[0-9]+}}e{{[-+]}}{{[0-9]}}{{[0-9]}}
+// CHECK-FILE-NOT: }
+// CHECK-FILE-NOT: }
+
+class A {
+ A() {}
+ ~A() {}
+};
Index: test/clang-tidy/clang-tidy-enable-check-profile-two-tu.cpp
===
--- test/clang-tidy/clang-tidy-enable-check-profile-two-tu.cpp
+++ test/clang-tidy/clang-tidy-enable-check-profile-two-tu.cpp
@@ -1,22 +1,31 @@
// RUN: clang-tidy -enable-check-profile -checks='-*,readability-function-size' %s %s 2>&1 | FileCheck --match-full-lines -implicit-check-not='{{warning:|error:}}' %s
// CHECK: ===-===
-// CHECK-NEXT: {{.*}} --- Name ---
+// CHECK-NEXT: clang-tidy checks profiling
+// CHECK-NEXT: ===-===
+// CHECK-NEXT: Total Execution Time: {{.*}} seconds ({{.*}} wall clock)
+
+// CHECK: {{.*}} --- Name ---
// CHECK-NEXT: {{.*}} readability-function-size
// CHECK-NEXT: {{.*}} Total
-// CHECK-NEXT: ===-===
// CHECK: ===-===
-// CHECK-NEXT: {{.*}} --- Name ---
+// CHECK-NEXT: clang-tidy checks profiling
+// CHECK-NEXT: ===-===
+// CHECK-NEXT: Total Execution Time: {{.*}} seconds ({{.*}} wall clock)
+
+// CHECK: {{.*}} --- Name ---
// CHECK-NEXT: {{.*}} readability-function-size
// CHECK-NEXT: {{.*}} Total
-
[PATCH] D46602: [clang-tidy] Store checks profiling info as YAML files
lebedev.ri added inline comments.
Comment at: clang-tidy/tool/ClangTidyMain.cpp:342-347
+if (!llvm::sys::fs::exists(AbsolutePath)) {
+ // If the destination prefix does not exist, don't try to use
real_path().
+ return AbsolutePath;
+}
+SmallString<256> dest;
+if (std::error_code EC = llvm::sys::fs::real_path(AbsolutePath, dest)) {
aaron.ballman wrote:
> This creates a TOCTOU bug; can you call `real_path()` without checking for
> existence and instead check the error code to decide whether to spit out an
> error or return `AbsolutePath`?
Hmm, i think this can just go away.
Repository:
rCTE Clang Tools Extra
https://reviews.llvm.org/D46602
___
cfe-commits mailing list
cfe-commits@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D47358: : Implement {un, }synchronized_pool_resource.
Quuxplusone updated this revision to Diff 149611.
Quuxplusone added a comment.
- Split up the unit tests.
- Refactor to shrink the memory layout of the resource object itself. Before
this patch `sizeof(unsynchronized_pool_resource)==48`. After this patch
`sizeof(unsynchronized_pool_resource) == 32`.
Repository:
rCXX libc++
https://reviews.llvm.org/D47358
Files:
include/experimental/memory_resource
src/experimental/memory_resource.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.ctor/ctor_does_not_allocate.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.ctor/sync_with_default_resource.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.ctor/unsync_with_default_resource.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/equality.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/sync_allocate.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/sync_allocate_overaligned_request.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/sync_allocate_reuse_blocks.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/unsync_allocate.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/unsync_allocate_overaligned_request.pass.cpp
test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/unsync_allocate_reuse_blocks.pass.cpp
test/std/experimental/memory/memory.resource.pool/pool_options.pass.cpp
test/support/count_new.hpp
Index: test/support/count_new.hpp
===
--- test/support/count_new.hpp
+++ test/support/count_new.hpp
@@ -211,6 +211,11 @@
return disable_checking || n != delete_called;
}
+bool checkDeleteCalledGreaterThan(int n) const
+{
+return disable_checking || delete_called > n;
+}
+
bool checkAlignedNewCalledEq(int n) const
{
return disable_checking || n == aligned_new_called;
Index: test/std/experimental/memory/memory.resource.pool/pool_options.pass.cpp
===
--- /dev/null
+++ test/std/experimental/memory/memory.resource.pool/pool_options.pass.cpp
@@ -0,0 +1,25 @@
+//===--===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===--===//
+
+// UNSUPPORTED: c++98, c++03, c++11, c++14
+// UNSUPPORTED: apple-clang-7
+
+//
+
+// struct pool_options
+
+#include
+#include
+
+int main()
+{
+const std::experimental::pmr::pool_options p;
+assert(p.max_blocks_per_chunk == 0);
+assert(p.largest_required_pool_block == 0);
+}
Index: test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/unsync_allocate_reuse_blocks.pass.cpp
===
--- /dev/null
+++ test/std/experimental/memory/memory.resource.pool/memory.resource.pool.mem/unsync_allocate_reuse_blocks.pass.cpp
@@ -0,0 +1,53 @@
+//===--===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===--===//
+
+// UNSUPPORTED: c++98, c++03
+
+//
+
+// class unsynchronized_pool_resource
+
+#include
+#include
+
+#include "count_new.hpp"
+
+static bool is_aligned_to(void *p, size_t alignment)
+{
+void *p2 = p;
+size_t space = 1;
+void *result = std::align(alignment, 1, p2, space);
+return (result == p);
+}
+
+int main()
+{
+globalMemCounter.reset();
+std::experimental::pmr::pool_options opts { 1, 256 };
+std::experimental::pmr::unsynchronized_pool_resource unsync1(opts, std::experimental::pmr::new_delete_resource());
+std::experimental::pmr::memory_resource & r1 = unsync1;
+
+void *ret = r1.allocate(8);
+assert(ret != nullptr);
+assert(is_aligned_to(ret, 8));
+assert(globalMemCounter.checkNewCalledGreaterThan(0));
+int new_called = globalMemCounter.new_called;
+
+// After deallocation, the pool for 8-byte blocks should have at least one vacancy.
+r1.deallocate(ret, 8);
+assert(globalMemCounter.new_called == new_called);
+assert(globalMemCounter.checkDeleteCalledEq(0));
+
+// This should return an existing block from the pool: no new allocations.
+ret = r1.allocate(8);
+assert(ret != nullptr);
+assert(is_aligned_to(ret, 8));
+
[PATCH] D45050: [clang-tidy] New checker for not null-terminated result caused by strlen or wcslen
Charusso added a comment. In https://reviews.llvm.org/D45050#1116446, @lebedev.ri wrote: > I would like to see more negative tests. > What does it do if the len/size is a constant? > Variable that wasn't just assigned with `strlen()` return? Thanks for the comment! What do you mean by negative test? https://reviews.llvm.org/D45050 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D45050: [clang-tidy] New checker for not null-terminated result caused by strlen or wcslen
lebedev.ri added a comment. In https://reviews.llvm.org/D45050#1119973, @Charusso wrote: > In https://reviews.llvm.org/D45050#1116446, @lebedev.ri wrote: > > > I would like to see more negative tests. > > What does it do if the len/size is a constant? > > Variable that wasn't just assigned with `strlen()` return? > > > Thanks for the comment! What do you mean by negative test? The tests where the check matches and issues a warning is a positive test. The tests where the check does not match and/or does not issue a warning is a negative test. https://reviews.llvm.org/D45050 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D45050: [clang-tidy] New checker for not null-terminated result caused by strlen or wcslen
xbolva00 added a comment. In https://reviews.llvm.org/D45050#1119973, @Charusso wrote: > In https://reviews.llvm.org/D45050#1116446, @lebedev.ri wrote: > > > I would like to see more negative tests. > > What does it do if the len/size is a constant? > > Variable that wasn't just assigned with `strlen()` return? > > > Thanks for the comment! What do you mean by negative test? To prove we have no false warnings. https://reviews.llvm.org/D45050 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D47121: [NEON] Support VLD1xN intrinsics in AArch32 mode (Clang part)
This revision was automatically updated to reflect the committed changes.
Closed by commit rL333819: [NEON] Support VLD1xN intrinsics in AArch32 mode
(Clang part) (authored by kosarev, committed by ).
Herald added a subscriber: llvm-commits.
Changed prior to commit:
https://reviews.llvm.org/D47121?vs=147718&id=149612#toc
Repository:
rL LLVM
https://reviews.llvm.org/D47121
Files:
llvm/trunk/include/llvm/IR/IntrinsicsARM.td
llvm/trunk/lib/Target/ARM/ARMBaseInstrInfo.cpp
llvm/trunk/lib/Target/ARM/ARMExpandPseudoInsts.cpp
llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
llvm/trunk/lib/Target/ARM/ARMInstrNEON.td
llvm/trunk/test/CodeGen/ARM/arm-vld1.ll
Index: llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
===
--- llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ llvm/trunk/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -1761,9 +1761,7 @@
case MVT::v4f32:
case MVT::v4i32: OpcodeIndex = 2; break;
case MVT::v2f64:
- case MVT::v2i64: OpcodeIndex = 3;
-assert(NumVecs == 1 && "v2i64 type only supported for VLD1");
-break;
+ case MVT::v2i64: OpcodeIndex = 3; break;
}
EVT ResTy;
@@ -3441,6 +3439,51 @@
return;
}
+case Intrinsic::arm_neon_vld1x2: {
+ static const uint16_t DOpcodes[] = { ARM::VLD1q8, ARM::VLD1q16,
+ ARM::VLD1q32, ARM::VLD1q64 };
+ static const uint16_t QOpcodes[] = { ARM::VLD1d8QPseudo,
+ ARM::VLD1d16QPseudo,
+ ARM::VLD1d32QPseudo,
+ ARM::VLD1d64QPseudo };
+ SelectVLD(N, false, 2, DOpcodes, QOpcodes, nullptr);
+ return;
+}
+
+case Intrinsic::arm_neon_vld1x3: {
+ static const uint16_t DOpcodes[] = { ARM::VLD1d8TPseudo,
+ ARM::VLD1d16TPseudo,
+ ARM::VLD1d32TPseudo,
+ ARM::VLD1d64TPseudo };
+ static const uint16_t QOpcodes0[] = { ARM::VLD1q8LowTPseudo_UPD,
+ARM::VLD1q16LowTPseudo_UPD,
+ARM::VLD1q32LowTPseudo_UPD,
+ARM::VLD1q64LowTPseudo_UPD };
+ static const uint16_t QOpcodes1[] = { ARM::VLD1q8HighTPseudo,
+ARM::VLD1q16HighTPseudo,
+ARM::VLD1q32HighTPseudo,
+ARM::VLD1q64HighTPseudo };
+ SelectVLD(N, false, 3, DOpcodes, QOpcodes0, QOpcodes1);
+ return;
+}
+
+case Intrinsic::arm_neon_vld1x4: {
+ static const uint16_t DOpcodes[] = { ARM::VLD1d8QPseudo,
+ ARM::VLD1d16QPseudo,
+ ARM::VLD1d32QPseudo,
+ ARM::VLD1d64QPseudo };
+ static const uint16_t QOpcodes0[] = { ARM::VLD1q8LowQPseudo_UPD,
+ARM::VLD1q16LowQPseudo_UPD,
+ARM::VLD1q32LowQPseudo_UPD,
+ARM::VLD1q64LowQPseudo_UPD };
+ static const uint16_t QOpcodes1[] = { ARM::VLD1q8HighQPseudo,
+ARM::VLD1q16HighQPseudo,
+ARM::VLD1q32HighQPseudo,
+ARM::VLD1q64HighQPseudo };
+ SelectVLD(N, false, 4, DOpcodes, QOpcodes0, QOpcodes1);
+ return;
+}
+
case Intrinsic::arm_neon_vld2: {
static const uint16_t DOpcodes[] = { ARM::VLD2d8, ARM::VLD2d16,
ARM::VLD2d32, ARM::VLD1q64 };
Index: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
===
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
@@ -12763,6 +12763,9 @@
case ISD::INTRINSIC_W_CHAIN:
switch (cast(N->getOperand(1))->getZExtValue()) {
case Intrinsic::arm_neon_vld1:
+case Intrinsic::arm_neon_vld1x2:
+case Intrinsic::arm_neon_vld1x3:
+case Intrinsic::arm_neon_vld1x4:
case Intrinsic::arm_neon_vld2:
case Intrinsic::arm_neon_vld3:
case Intrinsic::arm_neon_vld4:
@@ -14074,6 +14077,21 @@
Info.flags = MachineMemOperand::MOLoad;
return true;
}
+ case Intrinsic::arm_neon_vld1x2:
+ case Intrinsic::arm_neon_vld1x3:
+ case Intrinsic::arm_neon_vld1x4: {
+Info.opc = ISD::INTRINSIC_W_CHAIN;
+// Conservatively set memVT to the entire set of vectors loaded.
+auto &DL = I.getCalledFunction()->getParent()->getDataLayout();
+uint64_t NumElts = DL.getTypeSizeInBits(I.getType()) / 64;
+Info.memVT = EVT::getVectorVT(I.getType()->g
[PATCH] D47103: Implement strip.invariant.group
Prazek added a comment. friendly ping Repository: rL LLVM https://reviews.llvm.org/D47103 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
[PATCH] D47299: [CodeGenCXX] Emit strip.invariant.group with -fstrict-vtable-pointers
Prazek added a comment. @rjmccall @rsmith friendly ping Repository: rL LLVM https://reviews.llvm.org/D47299 ___ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
