[compiler-rt] Add initial ARM64EC builtins support #139279
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
You can test this locally with the following command:git-clang-format --diff HEAD~1 HEAD --extensions h,inc,c -- compiler-rt/lib/builtins/clear_cache.c compiler-rt/lib/builtins/cpu_model/aarch64.c compiler-rt/lib/builtins/cpu_model/aarch64.h compiler-rt/lib/builtins/fp_compare_impl.inc compiler-rt/lib/builtins/fp_lib.h compiler-rt/lib/builtins/udivmodti4.c compiler-rt/test/builtins/Unit/enable_execute_stack_test.c compiler-rt/test/builtins/Unit/fixunstfdi_test.c compiler-rt/test/builtins/Unit/multc3_test.cView the diff from clang-format here.diff --git a/compiler-rt/test/builtins/Unit/fixunstfdi_test.c b/compiler-rt/test/builtins/Unit/fixunstfdi_test.c
index 982f3a462..0a66602fa 100644
--- a/compiler-rt/test/builtins/Unit/fixunstfdi_test.c
+++ b/compiler-rt/test/builtins/Unit/fixunstfdi_test.c
@@ -6,7 +6,7 @@
#if _ARCH_PPC || __aarch64__ || __arm64ec__
-#include "int_lib.h"
+# include "int_lib.h"
// Returns: convert a to a unsigned long long, rounding toward zero.
// Negative values all become zero.
@@ -36,77 +36,77 @@ char assumption_3[sizeof(long double)*CHAR_BIT == 128] = {0};
int main()
{
#if _ARCH_PPC || __aarch64__ || __arm64ec__
- if (test__fixunstfdi(0.0, 0))
- return 1;
-
- if (test__fixunstfdi(0.5, 0))
- return 1;
- if (test__fixunstfdi(0.99, 0))
- return 1;
- if (test__fixunstfdi(1.0, 1))
- return 1;
- if (test__fixunstfdi(1.5, 1))
- return 1;
- if (test__fixunstfdi(1.99, 1))
- return 1;
- if (test__fixunstfdi(2.0, 2))
- return 1;
- if (test__fixunstfdi(2.01, 2))
- return 1;
- if (test__fixunstfdi(-0.5, 0))
- return 1;
- if (test__fixunstfdi(-0.99, 0))
- return 1;
- if (test__fixunstfdi(-1.0, 0))
- return 1;
- if (test__fixunstfdi(-1.5, 0))
- return 1;
- if (test__fixunstfdi(-1.99, 0))
- return 1;
- if (test__fixunstfdi(-2.0, 0))
- return 1;
- if (test__fixunstfdi(-2.01, 0))
- return 1;
-
- if (test__fixunstfdi(0x1.FFFFFEp+62, 0x7FFFFF8000000000LL))
- return 1;
- if (test__fixunstfdi(0x1.FFFFFCp+62, 0x7FFFFF0000000000LL))
- return 1;
-
- if (test__fixunstfdi(-0x1.FFFFFEp+62, 0))
- return 1;
- if (test__fixunstfdi(-0x1.FFFFFCp+62, 0))
- return 1;
-
- if (test__fixunstfdi(0x1.FFFFFFFFFFFFFp+62, 0x7FFFFFFFFFFFFC00LL))
- return 1;
- if (test__fixunstfdi(0x1.FFFFFFFFFFFFEp+62, 0x7FFFFFFFFFFFF800LL))
- return 1;
-
- if (test__fixunstfdi(-0x1.FFFFFFFFFFFFFp+62, 0))
- return 1;
- if (test__fixunstfdi(-0x1.FFFFFFFFFFFFEp+62, 0))
- return 1;
-
- if (test__fixunstfdi(0x1.FFFFFFFFFFFFFFFEp+63L, 0xFFFFFFFFFFFFFFFFLL))
- return 1;
- if (test__fixunstfdi(0x1.0000000000000002p+63L, 0x8000000000000001LL))
- return 1;
- if (test__fixunstfdi(0x1.0000000000000000p+63L, 0x8000000000000000LL))
- return 1;
- if (test__fixunstfdi(0x1.FFFFFFFFFFFFFFFCp+62L, 0x7FFFFFFFFFFFFFFFLL))
- return 1;
- if (test__fixunstfdi(0x1.FFFFFFFFFFFFFFF8p+62L, 0x7FFFFFFFFFFFFFFELL))
- return 1;
- if (test__fixunstfdi(0x1.p+64L, 0xFFFFFFFFFFFFFFFFLL))
- return 1;
-
- if (test__fixunstfdi(-0x1.0000000000000000p+63L, 0))
- return 1;
- if (test__fixunstfdi(-0x1.FFFFFFFFFFFFFFFCp+62L, 0))
- return 1;
- if (test__fixunstfdi(-0x1.FFFFFFFFFFFFFFF8p+62L, 0))
- return 1;
+ if (test__fixunstfdi(0.0, 0))
+ return 1;
+
+ if (test__fixunstfdi(0.5, 0))
+ return 1;
+ if (test__fixunstfdi(0.99, 0))
+ return 1;
+ if (test__fixunstfdi(1.0, 1))
+ return 1;
+ if (test__fixunstfdi(1.5, 1))
+ return 1;
+ if (test__fixunstfdi(1.99, 1))
+ return 1;
+ if (test__fixunstfdi(2.0, 2))
+ return 1;
+ if (test__fixunstfdi(2.01, 2))
+ return 1;
+ if (test__fixunstfdi(-0.5, 0))
+ return 1;
+ if (test__fixunstfdi(-0.99, 0))
+ return 1;
+ if (test__fixunstfdi(-1.0, 0))
+ return 1;
+ if (test__fixunstfdi(-1.5, 0))
+ return 1;
+ if (test__fixunstfdi(-1.99, 0))
+ return 1;
+ if (test__fixunstfdi(-2.0, 0))
+ return 1;
+ if (test__fixunstfdi(-2.01, 0))
+ return 1;
+
+ if (test__fixunstfdi(0x1.FFFFFEp+62, 0x7FFFFF8000000000LL))
+ return 1;
+ if (test__fixunstfdi(0x1.FFFFFCp+62, 0x7FFFFF0000000000LL))
+ return 1;
+
+ if (test__fixunstfdi(-0x1.FFFFFEp+62, 0))
+ return 1;
+ if (test__fixunstfdi(-0x1.FFFFFCp+62, 0))
+ return 1;
+
+ if (test__fixunstfdi(0x1.FFFFFFFFFFFFFp+62, 0x7FFFFFFFFFFFFC00LL))
+ return 1;
+ if (test__fixunstfdi(0x1.FFFFFFFFFFFFEp+62, 0x7FFFFFFFFFFFF800LL))
+ return 1;
+
+ if (test__fixunstfdi(-0x1.FFFFFFFFFFFFFp+62, 0))
+ return 1;
+ if (test__fixunstfdi(-0x1.FFFFFFFFFFFFEp+62, 0))
+ return 1;
+
+ if (test__fixunstfdi(0x1.FFFFFFFFFFFFFFFEp+63L, 0xFFFFFFFFFFFFFFFFLL))
+ return 1;
+ if (test__fixunstfdi(0x1.0000000000000002p+63L, 0x8000000000000001LL))
+ return 1;
+ if (test__fixunstfdi(0x1.0000000000000000p+63L, 0x8000000000000000LL))
+ return 1;
+ if (test__fixunstfdi(0x1.FFFFFFFFFFFFFFFCp+62L, 0x7FFFFFFFFFFFFFFFLL))
+ return 1;
+ if (test__fixunstfdi(0x1.FFFFFFFFFFFFFFF8p+62L, 0x7FFFFFFFFFFFFFFELL))
+ return 1;
+ if (test__fixunstfdi(0x1.p+64L, 0xFFFFFFFFFFFFFFFFLL))
+ return 1;
+
+ if (test__fixunstfdi(-0x1.0000000000000000p+63L, 0))
+ return 1;
+ if (test__fixunstfdi(-0x1.FFFFFFFFFFFFFFFCp+62L, 0))
+ return 1;
+ if (test__fixunstfdi(-0x1.FFFFFFFFFFFFFFF8p+62L, 0))
+ return 1;
#else
printf("skipped\n");
diff --git a/compiler-rt/test/builtins/Unit/multc3_test.c b/compiler-rt/test/builtins/Unit/multc3_test.c
index e9c99a72b..44707c75d 100644
--- a/compiler-rt/test/builtins/Unit/multc3_test.c
+++ b/compiler-rt/test/builtins/Unit/multc3_test.c
@@ -6,9 +6,9 @@
#if _ARCH_PPC || __aarch64__ || __arm64ec__
-#include "int_lib.h"
-#include <math.h>
-#include <complex.h>
+# include "int_lib.h"
+# include <math.h>
+# include <complex.h>
// Returns: the product of a + ib and c + id
@@ -349,16 +349,14 @@ long double x[][2] =
int main()
{
#if _ARCH_PPC || __aarch64__ || __arm64ec__
- const unsigned N = sizeof(x) / sizeof(x[0]);
- unsigned i, j;
- for (i = 0; i < N; ++i)
- {
- for (j = 0; j < N; ++j)
- {
- if (test__multc3(x[i][0], x[i][1], x[j][0], x[j][1]))
- return 1;
- }
+ const unsigned N = sizeof(x) / sizeof(x[0]);
+ unsigned i, j;
+ for (i = 0; i < N; ++i) {
+ for (j = 0; j < N; ++j) {
+ if (test__multc3(x[i][0], x[i][1], x[j][0], x[j][1]))
+ return 1;
}
+ }
#else
printf("skipped\n");
#endif
|
|
The code formatting errors relate to existing formatting around my changes. Including those in the diff would be misleading, in my opinion. |
mstorsjo
approved these changes
May 14, 2025
| @@ -122,7 +123,7 @@ void __clear_cache(void *start, void *end) { | |||
| compilerrt_abort(); | |||
| #endif | |||
| } | |||
| #elif defined(__aarch64__) && !defined(__APPLE__) | |||
| #elif (defined(__aarch64__) || defined(__arm64ec__)) && !defined(__APPLE__) | |||
There was a problem hiding this comment.
Isn't this case redundant here - isn't this #elif on the same level as the _WIN32 && (arm || aarch64 || arm64ec) at the start of the function?
On the other hand, it's probably ok on one sense, as I guess this implementation should be usable too - or is it? I'm not entirely sure if this uses some instruction which is not accessible from userspace on Windows (which could be why we're using FlushInstructionCache to being with). Or perhaps that was required for 32 bit arm and we just kept doing the same for 64 bit too?
There was a problem hiding this comment.
Good point, I think this is redundant. This came from Billy's original patch. I remember changing a similar case elsewhere but missed this one.
| @@ -10,9 +10,22 @@ extern void __enable_execute_stack(void* addr); | |||
|
|
|||
| typedef int (*pfunc)(void); | |||
|
|
|||
| #ifdef __x86_64__ | |||
| // On ARM64EC, we need the x86_64 version of this function, but the compiler | |||
There was a problem hiding this comment.
Doing this for all of x86_64 seems a bit far-reaching; wouldn't it be cleaner to restrict this case to arm64ec? Then again, extending it to all of x86_64 does give it more coverage and exposure, but ideally I wouldn't be touching the behaviour for other architectures in a patch like this.
There was a problem hiding this comment.
I changed this to use __arm64ec__.
| @@ -31,22 +44,28 @@ int main() | |||
| { | |||
| #if defined(__ve__) | |||
| unsigned char execution_buffer[128] __attribute__((__aligned__(8))); | |||
| #elif defined(__x86_64__) | |||
| unsigned char execution_buffer[sizeof(func1)]; | |||
There was a problem hiding this comment.
Why do we need to shrink the buffer here?
| pfunc f1 = (pfunc)memcpy_f(execution_buffer, func1, 128); | ||
| __clear_cache(execution_buffer, &execution_buffer[128]); | ||
| pfunc f1 = | ||
| (pfunc)memcpy_f(execution_buffer, func1, sizeof(execution_buffer)); |
There was a problem hiding this comment.
Same here - shouldn't overreads here be equally problematic for arm64ec as for all the existing architectures?
There was a problem hiding this comment.
My thinking was that using an array for the code made it easier to avoid overreads than on other platforms, but there’s no strong reason for it. I’ve reverted that in the new version. Thanks for the review!
Use the aarch64 variants of assembly functions. Based on a patch by Billy Laws.
5d27c3f to
48aa2a7
Compare
mstorsjo
approved these changes
May 15, 2025
|
I believe the meaning is the same (this is a modified version of Billy's patch), but I agree it's better to use the formalized form. I've updated the commit accordingly, thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Use the aarch64 variants of assembly functions.
Based on a patch by Billy Laws.