[AArch64][compiler-rt] Add memcpy, memset, memmove, memchr builtins. #77496

dtemirbulatov · 2024-01-09T16:44:07Z

Add naive implementation of memcpy, memset, memmove, memchr for SME targets.
Co-authored-by: David Sherwood david.sherwood@arm.com

github-actions · 2024-01-09T16:47:00Z

✅ With the latest revision this PR passed the C/C++ code formatter.

…lementation RT builtins. Add naive implementation of memcpy, memset, memmove, memchr for SME targets. Patch co-authored by David Sherwood <david.sherwood@arm.com>

sdesmalen-arm · 2024-01-10T14:32:47Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+// restrict keywords here. This also matches the definition of the libc memcpy
+// according to the man page.
+void *__arm_sc_memcpy(void *__restrict__ dest, const void *__restrict__ src,
+                      size_t n) __arm_streaming_compatible __arm_preserves_za {


The compiler building compiler-rt may not support __arm_streaming_compatible. You'll want to check for the SME attributes when configuring the project with cmake.

I'd also suggest removing __arm_preserves_za in anticipation of this ACLE PR landing.

hmm, not sure what extra required for __arm_streaming_compatible avaliblity check, We have already check that in compiler-rt/lib/builtins/CMakeLists.txt?

sdesmalen-arm · 2024-01-10T15:07:25Z

compiler-rt/test/builtins/Unit/sme-string-test.c

+#include <time.h>
+
+#define N 1024
+#define NREPS 1234


It looks like this test is testing more than just functionality (i.e. it seems to be testing performance). This can be seen from e.g. NREPS and get_time_diff. I think you'll want to make this purely a functional test and try to test out specific edge cases for each of the routines.

sdesmalen-arm · 2024-01-12T08:52:04Z

compiler-rt/unittests/lit.common.unit.configured.in

@@ -7,6 +7,7 @@ config.llvm_obj_root = "@LLVM_BINARY_DIR@"
 config.llvm_tools_dir = lit_config.substitute("@LLVM_TOOLS_DIR@")
 config.compiler_rt_src_root = "@COMPILER_RT_SOURCE_DIR@"
 config.compiler_rt_libdir = lit_config.substitute("@COMPILER_RT_RESOLVED_LIBRARY_OUTPUT_DIR@")
+config.sme = "@COMPILER_RT_HAS_ASM_SME@"


Please call this aarch64.sme, just to avoid any future name clashes with other kinds of 'sme'.

sdesmalen-arm · 2024-01-12T08:53:05Z

compiler-rt/test/builtins/Unit/sme-string-test.c

 int main() {
-  struct timespec tv[2];

  init();

  // Test correctness of memcpy
  for (int i = 0; i < 67; i++) {


This is a test that goes from 0 to 67. Is 67a special number that was chosen for a reason?

sdesmalen-arm · 2024-01-12T08:57:17Z

compiler-rt/test/builtins/Unit/sme-string-test.c

+  }
+}
+
+void reinit_dst(int n) {


I think it's worth making this a .cpp file so that you can have a class with a constructor to do the init and have a sum method, rather than having to call functions like reinit_dst. We can reduce the sizes of the arrays a bit more to reduce runtime of the test, which makes the impact of reinitialising the entire array negligible.

In general I'd like to see a more structured way to test these routines, where we can see that some of the edge-cases are being tested. At the moment it looks like you've stripped the timing functionality from this file.

sdesmalen-arm · 2024-01-12T09:01:47Z

compiler-rt/lib/builtins/CMakeLists.txt

@@ -561,7 +561,7 @@ set(aarch64_SOURCES
 )

 if(COMPILER_RT_HAS_ASM_SME AND (COMPILER_RT_HAS_AUXV OR COMPILER_RT_BAREMETAL_BUILD))


COMPILER_RT_HAS_ASM_SME tests if the compiler supports SME assembly, but not if the compiler supports the C/C++ level attributes. You'll need to add another check to the function which sets this variable and probably rename the variable to COMPILER_RT_HAS_AARCH64_SME to make it clear the compiler supports more than just the ASM. (and also to add "aarch64" to the name to avoid future name clashes with other meanings of 'sme')

sdesmalen-arm · 2024-01-16T10:44:19Z

compiler-rt/cmake/builtin-config-ix.cmake

@@ -41,6 +41,12 @@ asm(\".arch armv9-a+sme\");
 asm(\"smstart\");
 ")

+builtin_check_c_compiler_source(COMPILER_RT_HAS_AARCH64_SME


Can you combine this with COMPILER_RT_HAS_ASM_SME and just have a single COMPILER_RT_HAS_AARCH64_SME ? That way we don't need to have two variables given that you're AND'ing the result later on anyway.

sdesmalen-arm · 2024-01-16T11:09:51Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+// WARNING: When building the scalar versions of these functions you need to
+// use the compiler flag "-mllvm -disable-loop-idiom-all" to prevent clang
+// from recognising a loop idiom and planting calls to memcpy!


There is a comment here that seems important, but doesn't seem to be addressed by your patch?

sdesmalen-arm · 2024-01-16T11:10:24Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+  // Test correctness of memcpy
+  void test_memcpy() {
+    for (int i = 0; i < 8; i++) {
+      int t[2];


Why do you use an array, rather than two separate variables?

sdesmalen-arm · 2024-01-16T11:57:51Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

@@ -0,0 +1,108 @@
+// REQUIRES: linux, aarch64-target-arch, aarch64-sme-available


Does it require linux?

sdesmalen-arm · 2024-01-16T14:08:18Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+void *__arm_sc_memchr(const void *, int, size_t);
+}
+
+class MemoryArea {


Just an idea (no need to copy this literally), but I think you'd get more value from using a class like this if you'd do something like this:

template <unsigned N> class Buffer { public: uint8_t ptr[N]; Buffer(int Stride = 0) { for (unsigned I=0; I < N; I+= Stride) { Src[I] = I; } } // to more quickly check that the result buffer matches a specific buffer. Buffer(std::initializer_list<...> S) { } bool assert_equal(const Buffer &Other) { // check here that ptr == Other.ptr } };

Then you can do things like:

{ // test 1 Memory<8> Src(1), Dst; __arm_sc_memcpy(Dst.ptr, Src.ptr, 8); Dst.assert_equal(Src); Dst.assert_equal({0, 1, 2, 3, 4, 5, 6, 7}); } { // test 2 Memory<8> SrcDst(1); __arm_sc_memcpy(SrcDst.ptr + 1, SrcDst.ptr, 7); SrcDst.assert_equal({0, 0, 1, 2, 3, 4, 5, 6}); }

You could also compare it to the result from a regular memcpy.

sdesmalen-arm · 2024-01-16T14:12:21Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+    for (int i = 0; i < 8; i++) {
+      int t[2];
+      if (!__arm_sc_memset(dst, src[i], i))
+        abort();
+      t[0] = sum_and_reset_dst(dst, N, i);
+      __arm_sc_memset(dst, src[i], i);
+      t[1] = sum_and_reset_dst(dst, N, i);
+      if (t[0] != t[1])
+        abort();
+    }


We discussed this offline, but please write some more specific tests that target the corner cases of the routines.

sdesmalen-arm

Thanks for rewriting the tests, this is looking much better!

sdesmalen-arm · 2024-01-19T14:16:00Z

compiler-rt/cmake/builtin-config-ix.cmake

 "
-asm(\".arch armv9-a+sme\");
-asm(\"smstart\");
+void foo(int a)  __arm_streaming_compatible {


nit:

Suggested change

void foo(int a) __arm_streaming_compatible {

void foo(void) __arm_streaming_compatible {

sdesmalen-arm · 2024-01-19T14:31:28Z

compiler-rt/lib/builtins/CMakeLists.txt

  message(STATUS "AArch64 SME ABI routines enabled")
+  if("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
+    set_source_files_properties(aarch64/sme-libc-routines.c PROPERTIES COMPILE_FLAGS "-mllvm -disable-loop-idiom-all")


It seems that you can use the option -fno-builtin to avoid builtins being used in the implementation of these functions. This flag is supported by both GNU and Clang, which is better than passing some internal LLVM flag.

It does mean you'll need to add: AND COMPILER_RT_HAS_FNO_BUILTIN_FLAG to the condition on line 563 though.

sdesmalen-arm · 2024-01-19T14:34:42Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+    if (Stride != 0) {
+      for (unsigned I = 0, Elem = 0; I < N; I++, Elem += Stride) {
+        ptr[I] = Elem;
+      }
+    }


nit:

Suggested change

if (Stride != 0) {

for (unsigned I = 0, Elem = 0; I < N; I++, Elem += Stride) {

ptr[I] = Elem;

}

}

if (Stride == 0)

return;

for (unsigned I = 0; I < N; ++I)

ptr[I] = I * Stride;

sdesmalen-arm · 2024-01-19T14:36:52Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+public:
+  uint8_t ptr[N];
+
+  Memory(int Stride = 0) {


nit:

Suggested change

Memory(int Stride = 0) {

Memory(unsigned Stride = 0) {

sdesmalen-arm · 2024-01-19T14:43:48Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+  }
+
+  void assert_equal(const Memory &Other) {
+    for (unsigned I = 0; I < N; I++) {


you can use memcmp here?

sdesmalen-arm · 2024-01-19T14:44:16Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+    }
+  }
+
+  void assert_equal(const Memory &Other) {


Do you also need to verify that this.N == Other.N ?

sdesmalen-arm · 2024-01-19T14:52:30Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+      if (!Elem)
+        abort();
+      Src.assert_elemt_equal_at(Elem - Src.ptr, *Elem);
+      assert(__arm_sc_memchr(Src.ptr, E, 8) == memchr(Src.ptr, E, 8));


I would this assert is sufficient, i.e.

for (unsigned I=0; I<8; ++I) assert(__arm_sc_memchr(Src.ptr, Src.ptr[I]) == memchr(Src.ptr, Src.ptr[I]));

sdesmalen-arm · 2024-01-19T14:54:46Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+  // Testing memset with different pointer offset.
+  {
+    for (unsigned I = 0; I < 16; I++) {
+      Memory<16> Array(2);
+      if (!__arm_sc_memset(Array.ptr + I, I, 16 - I))
+        abort();
+
+      uint8_t OrigElem = 0;
+      for (unsigned J = 0; J < 16; J++) {
+        if (I == 0) {
+          Array.assert_elemt_equal_at(J, 0);
+        } else if (J < I) {
+          Array.assert_elemt_equal_at(J, OrigElem);
+        } else {
+          Array.assert_elemt_equal_at(J, (uint8_t)I);
+        }
+        OrigElem += 2;
+      }
+    }
+  }
+


Is this testing anything that the above two tests aren't?

sdesmalen-arm

LGTM

MaskRay · 2024-01-22T08:32:39Z

Patch co-authored by David Sherwood david.sherwood@arm.com

Co-authored-by: needs to be at the line beginning https://docs.github.com/en/pull-requests/committing-changes-to-your-project/creating-and-editing-commits/creating-a-commit-with-multiple-authors

MaskRay · 2024-01-22T08:33:32Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+      return;
+    for (unsigned I = 0, Elem = 0; I < N; I++) {
+      ptr[I] = I * Stride;
+    }


drop braces around simple single statement

lower_case variable naming

MaskRay · 2024-01-22T08:34:17Z

compiler-rt/test/builtins/Unit/sme-string-test.cpp

+    assert(N == S.size());
+    auto It = S.begin();
+    for (unsigned I = 0; I < N; I++) {
+      assert(ptr[I] == *It++);


drop braces around simple single statement

compiler-rt/lib/builtins uses lower_case variable naming.

MaskRay · 2024-01-22T08:35:11Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+  // TODO: Improve performance by copying larger chunks in reverse, or by
+  // using SVE.
+  while (n > 0) {
+    n--;


pre-increment https://llvm.org/docs/CodingStandards.html#prefer-preincrement

MaskRay · 2024-01-22T08:35:32Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+void *__arm_sc_memset(void *dest, int c, size_t n) __arm_streaming_compatible {
+  unsigned char *destp = (unsigned char *)dest;
+  unsigned char c8 = (unsigned char)c;
+


delete blank line after variable declaration

MaskRay · 2024-01-22T08:35:41Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+  unsigned char c8 = (unsigned char)c;
+
+  for (size_t i = 0; i < n; i++) {
+    destp[i] = c8;


drop braces, ++i

MaskRay · 2024-01-22T08:35:49Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+                            size_t n) __arm_streaming_compatible {
+  const unsigned char *srcp = (const unsigned char *)src;
+  unsigned char c8 = (unsigned char)c;
+


delete blank line

MaskRay · 2024-01-22T08:35:55Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+  for (size_t i = 0; i < n; i++) {
+    if (srcp[i] == c8)
+      return &srcp[i];
+  }


drop braces

MaskRay · 2024-01-22T18:40:20Z

Add naive implementation of memcpy, memset, memmove, memchr for SME targets.

This can use more description why these functions need to go into compiler-rt. I assume that a future Clang will emit these calls (https://arm-software.github.io/acle/main/acle.html#streaming-compatible-versions-of-standard-routines).

…lementation RT builtins.

The title has been wrapped by GitHub. It will be more readable by updating the title not to be wrapped.

MaskRay · 2024-01-22T18:41:16Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

+  unsigned char *destp = (unsigned char *)dest;
+  const unsigned char *srcp = (const unsigned char *)src;
+  // If src and dest are identical there is nothing to do!
+  if ((destp == srcp) || (n == 0))


I wonder whether the two conditions (destp == srcp) || (n == 0) are useful.

A lot of libc implementations don't check destp == srcp. Some don't check n==0

yes, I think we can safely remove those now. Thanks for finding this.

or length parameter to move is zero.

dyung · 2024-01-23T00:35:41Z

@dtemirbulatov your change seems to cause some issues when trying to run the lit tests on some bots:

For example: https://lab.llvm.org/buildbot/#/builders/247/builds/13502

llvm-lit: /home/buildbot/buildbot-root/llvm-clang-x86_64-gcc-ubuntu/llvm-project/llvm/utils/lit/lit/TestingConfig.py:152: fatal: unable to parse config file '/home/buildbot/buildbot-root/llvm-clang-x86_64-gcc-ubuntu/build/runtimes/runtimes-bins/compiler-rt/test/lit.common.configured', traceback: Traceback (most recent call last):
  File "/home/buildbot/buildbot-root/llvm-clang-x86_64-gcc-ubuntu/llvm-project/llvm/utils/lit/lit/TestingConfig.py", line 140, in load_from_path
    exec(compile(data, path, "exec"), cfg_globals, None)
  File "/home/buildbot/buildbot-root/llvm-clang-x86_64-gcc-ubuntu/build/runtimes/runtimes-bins/compiler-rt/test/lit.common.configured", line 66, in <module>
    set_default("aarch64_sme", )
TypeError: set_default() missing 1 required positional argument: 'value'

Can you take a look?

nico · 2024-01-23T01:27:42Z

Do you need to add a pythonize_bool() for COMPILER_RT_HAS_AARCH64_SME to compiler-rt/test/CMakeLists.txt?

Also, you do set_default("aarch64_sme", @COMPILER_RT_HAS_AARCH64_SME@) in lit.common.configured.in but config.aarch64_sme = "@COMPILER_RT_HAS_AARCH64_SME@" (as a string) in lit.common.unit.configured.in. Presumably the latter shouldn't have quotes either?

dyung · 2024-01-23T02:22:45Z

@dtemirbulatov sorry, I reverted your change as it is preventing many bots from running tests.

nico · 2024-01-23T02:32:37Z

When this relands, it'd be cool if you could put a change like 9a03d94 in the reland as well.

…77496) Add naive implementation of memcpy, memset, memmove, memchr for SME targets. Co-authored-by: David Sherwood <david.sherwood@arm.com>

SLTozer · 2024-01-23T14:53:39Z

As mentioned above this seems to be breaking on the buildbots, e.g. https://lab.llvm.org/buildbot/#/builders/275/builds/3646

dtemirbulatov · 2024-01-23T15:39:10Z

Commited solution for the last failure with 5f47687

SLTozer · 2024-01-23T15:52:22Z

The above commit hasn't fully fixed the issue - the error has now changed from the original:

llvm-lit: /home/buildbot/buildbot-root/debug-iterators/llvm-project/llvm/utils/lit/lit/TestingConfig.py:152: fatal: unable to parse config file '/home/buildbot/buildbot-root/debug-iterators/build/runtimes/runtimes-bins/compiler-rt/unittests/lit.common.unit.configured', traceback: Traceback (most recent call last):
  File "/home/buildbot/buildbot-root/debug-iterators/llvm-project/llvm/utils/lit/lit/TestingConfig.py", line 140, in load_from_path
    exec(compile(data, path, "exec"), cfg_globals, None)
  File "/home/buildbot/buildbot-root/debug-iterators/build/runtimes/runtimes-bins/compiler-rt/unittests/lit.common.unit.configured", line 23
    config.aarch64_sme =
                        ^
SyntaxError: invalid syntax

To the new error:

llvm-lit: /home/buildbot/buildbot-root/debug-iterators/llvm-project/llvm/utils/lit/lit/TestingConfig.py:152: fatal: unable to parse config file '/home/buildbot/buildbot-root/debug-iterators/build/runtimes/runtimes-bins/compiler-rt/test/lit.common.configured', traceback: Traceback (most recent call last):
  File "/home/buildbot/buildbot-root/debug-iterators/llvm-project/llvm/utils/lit/lit/TestingConfig.py", line 140, in load_from_path
    exec(compile(data, path, "exec"), cfg_globals, None)
  File "/home/buildbot/buildbot-root/debug-iterators/build/runtimes/runtimes-bins/compiler-rt/test/lit.common.configured", line 66, in <module>
    set_default("aarch64_sme", )
TypeError: set_default() missing 1 required positional argument: 'value'

jmorse · 2024-01-23T17:02:17Z

NB, I've taken a shot at fixing this with 5176df5 , I think using pythonize_bool is adding "_PYBOOL" to the end of the variable name that gets defined, like the nearby config flags (COMPILER_RT_TEST_STANDALONE_BUILD_LIBS for example gets _PYBOOL on the end).

jmorse · 2024-01-23T17:58:14Z

Looks like that cleared the bots -- @dtemirbulatov you should be able to make the same fix to the line you deleted in lit.common.unit.configured.in to restore coverage.

This is a follow-up to 3112578 -- it looks like passing the added cmake flag to pythonize_bool also appends "_PYBOOL" to the flag name, which this lit config file is missing. This trips up builds such as: https://lab.llvm.org/buildbot/#/builders/275/builds/3661 https://lab.llvm.org/buildbot/#/builders/184/builds/9811 Where COMPILER_RT_HAS_AARCH64_SME ends up expanding to nothing.

echristo · 2024-01-24T20:14:57Z

Add naive implementation of memcpy, memset, memmove, memchr for SME targets. Co-authored-by: David Sherwood david.sherwood@arm.com

Do we have an idea of the expected performance differences here in the presence of SME?

peterwaller-arm · 2024-01-25T14:43:41Z

compiler-rt/lib/builtins/aarch64/sme-libc-routines.c

@@ -0,0 +1,87 @@
+#include <stdlib.h>


I've proposed #79454 to switch this out for stddef.h which is provided by the compiler resource headers rather than the libc (since this is inside builtins).

sdesmalen-arm · 2024-01-26T13:19:06Z

Add naive implementation of memcpy, memset, memmove, memchr for SME targets.

Do we have an idea of the expected performance differences here in the presence of SME?

This PR adds the __arm_sc_ (prefixed) streaming-compatible variants to be compliant with the SME ACLE and are written with a focus on functional correctness as opposed to performance. They're more of a starting point right now, but we definitely want to optimise these routines going forward.

The reason to have streaming-compatible variants of these routines is because regular memcpy/memset/etc routines are not streaming-compatible. In practice this means they might be implemented using NEON instructions and so the compiler has to ensure that every call to it will result in a smstop sm -> bl memcpy -> smstart sm sequence when the caller is executed in Streaming-SVE mode. By adding the __arm_streaming_compatible attribute to these functions, the compiler will ensure the implementation generates no incompatible (e.g. NEON) instructions and also ensures that calls to functions with a streaming-compatible interface (e.g. __arm_sc_memcpy) don't result in toggling the streaming mode around the call.

davemgreen · 2024-01-31T09:06:24Z

I dont usually follow compiler-rt commits, so didnt see this. memcpy and friends can be used a lot - performance for them can be critical and a lot of time has been spent over the years trying to optimize them.

Is the idea that these are auto-vectorized? And so might use the SME engine for the copy? I'm not sure that is happening in the current version I have, and single byte memcpys sounds pretty slow. And that's before other considerations like alignment and systems wanting platform specific routines. I would add Wilco (I'm not sure what his username is), he has done a lot of work on these kinds of routines and it would be good to get his input.

sdesmalen-arm · 2024-01-31T09:14:25Z

The point of this implementation is that it doesn't use NEON instructions, as those are not valid in Streaming-SVE mode. We'll definitely want to optimise these routines by using (streaming-compatible) SVE instructions or by using the MOPS instructions, if available. This PR is merely a starting point to guarantee functional correctness.

It is also possible to use regular memcpy routines, but those will incur a smstop/smstart around the call-site when called from a streaming function.

davemgreen · 2024-01-31T10:30:54Z

I had forgotten that vectorization was disabled in streaming functions. That would make sense why we don't see SVE instructions so far. It sounds like there is plenty of work yet to go.

Hopefully there will be a lot of overlap with mops instructions, which would help simplify things a lot.

dtemirbulatov requested review from SamTebbs33, sdesmalen-arm, kmclaughlin-arm, david-arm and MDevereau January 9, 2024 16:44

llvmbot added the compiler-rt label Jan 9, 2024

[AArch64][compiler-rt] Add memcpy, memset, memmove, memchr simple imp…

2db7921

…lementation RT builtins. Add naive implementation of memcpy, memset, memmove, memchr for SME targets. Patch co-authored by David Sherwood <david.sherwood@arm.com>

dtemirbulatov force-pushed the compiler-rt-sme-libc-routines branch from 1845737 to 2db7921 Compare January 9, 2024 17:00

sdesmalen-arm reviewed Jan 10, 2024

View reviewed changes

Addressed comments.

7e2df5b

sdesmalen-arm reviewed Jan 12, 2024

View reviewed changes

Addressed comments.

2504e89

sdesmalen-arm reviewed Jan 16, 2024

View reviewed changes

Addressed comments.

5324c3f

sdesmalen-arm reviewed Jan 19, 2024

View reviewed changes

Addressed comments.

d31dcb1

sdesmalen-arm approved these changes Jan 22, 2024

View reviewed changes

MaskRay reviewed Jan 22, 2024

View reviewed changes

MaskRay requested changes Jan 22, 2024

View reviewed changes

MaskRay reviewed Jan 22, 2024

View reviewed changes

Resolved comments.

5d696f7

dtemirbulatov requested a review from MaskRay January 22, 2024 14:32

MaskRay approved these changes Jan 22, 2024

View reviewed changes

dtemirbulatov changed the title ~~[AArch64][compiler-rt] Add memcpy, memset, memmove, memchr simple imp…~~ [AArch64][compiler-rt] Add memcpy, memset, memmove, memchr builtins. Jan 22, 2024

Remove memmove check if source and destination pointers are equal

942a0d8

or length parameter to move is zero.

dtemirbulatov merged commit 3ab8d2a into llvm:main Jan 22, 2024
4 checks passed

dtemirbulatov deleted the compiler-rt-sme-libc-routines branch January 22, 2024 23:40

DavidSpickett mentioned this pull request Jan 23, 2024

[AArch64] compiler-rt tests fail due to blank @COMPILER_RT_HAS_AARCH64_SME@ variable #79138

Closed

peterwaller-arm reviewed Jan 25, 2024

View reviewed changes

		@@ -561,7 +561,7 @@ set(aarch64_SOURCES
		)

		if(COMPILER_RT_HAS_ASM_SME AND (COMPILER_RT_HAS_AUXV OR COMPILER_RT_BAREMETAL_BUILD))

		@@ -0,0 +1,108 @@
		// REQUIRES: linux, aarch64-target-arch, aarch64-sme-available

	void foo(int a) __arm_streaming_compatible {
	void foo(void) __arm_streaming_compatible {

[AArch64][compiler-rt] Add memcpy, memset, memmove, memchr builtins. #77496

[AArch64][compiler-rt] Add memcpy, memset, memmove, memchr builtins. #77496

Conversation

dtemirbulatov commented Jan 9, 2024 • edited Loading

github-actions bot commented Jan 9, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dtemirbulatov Jan 19, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sdesmalen-arm left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dtemirbulatov Jan 19, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sdesmalen-arm left a comment

Choose a reason for hiding this comment

MaskRay commented Jan 22, 2024

MaskRay Jan 22, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

MaskRay commented Jan 22, 2024

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dyung commented Jan 23, 2024

nico commented Jan 23, 2024

dyung commented Jan 23, 2024

nico commented Jan 23, 2024

SLTozer commented Jan 23, 2024 • edited Loading

dtemirbulatov commented Jan 23, 2024

SLTozer commented Jan 23, 2024

jmorse commented Jan 23, 2024

jmorse commented Jan 23, 2024

echristo commented Jan 24, 2024

peterwaller-arm Jan 25, 2024 • edited Loading

dtemirbulatov commented Jan 9, 2024 •

edited

Loading

github-actions bot commented Jan 9, 2024 •

edited

Loading

dtemirbulatov Jan 19, 2024 •

edited

Loading

dtemirbulatov Jan 19, 2024 •

edited

Loading

MaskRay Jan 22, 2024 •

edited

Loading

SLTozer commented Jan 23, 2024 •

edited

Loading

peterwaller-arm Jan 25, 2024 •

edited

Loading