[x265] [PATCH 01/12] AArch64: Fix costCoeffNxN test on Apple Silicon

[chen]

Hi Hari Limaye,




Thank you fix AARCH64 build issues, these 12 patches looks good for me.




Regards,

Chen

At 2024-05-03 05:19:36, "Hari Limaye" <hari.limaye at arm.com> wrote:
>The assembly routine x265_costCoeffNxN_neon is buggy and produces an
>incorrect result on Apple Silicon, causing the pixel testbench to fail
>on these platforms.
>
>x265_costCoeffNxN assumes that parameter `int subPosBase`, the second
>parameter of type `int` passed on the stack, is at position `sp + 8`;
>this assumption is consistent with the AArch64 PCS, as arguments smaller
>than 8 bytes are widened to 8 bytes (aapcs64 6.8.2 C.16).
>However arm64e diverges from AAPCS64: 'Function arguments may consume
>slots on the stack that are not multiples of 8 bytes'.
>---
> source/common/aarch64/asm.S        | 12 +++++++++++-
> source/common/aarch64/pixel-util.S |  4 ++--
> 2 files changed, 13 insertions(+), 3 deletions(-)
>
>diff --git a/source/common/aarch64/asm.S b/source/common/aarch64/asm.S
>index ce0668103..742978631 100644
>--- a/source/common/aarch64/asm.S
>+++ b/source/common/aarch64/asm.S
>@@ -72,6 +72,16 @@
>
> #define PFX_C(name)        JOIN(JOIN(JOIN(EXTERN_ASM, X265_NS), _), name)
>
>+// Alignment of stack arguments of size less than 8 bytes.
>+#ifdef __APPLE__
>+#define STACK_ARG_ALIGNMENT 4
>+#else
>+#define STACK_ARG_ALIGNMENT 8
>+#endif
>+
>+// Get offset from SP of stack argument at index `idx`.
>+#define STACK_ARG_OFFSET(idx) (idx * STACK_ARG_ALIGNMENT)
>+
> #ifdef __APPLE__
> .macro endfunc
> ELF .size \name, . - \name
>@@ -184,4 +194,4 @@ ELF     .size   \name, . - \name
>     vtrn            \t3, \t4, \s3, \s4
> .endm
>
>-#endif
>\ No newline at end of file
>+#endif
>diff --git a/source/common/aarch64/pixel-util.S b/source/common/aarch64/pixel-util.S
>index 9b3c11504..378c6891c 100644
>--- a/source/common/aarch64/pixel-util.S
>+++ b/source/common/aarch64/pixel-util.S
>@@ -2311,7 +2311,7 @@ endfunc
> //    uint8_t *baseCtx,      // x6
> //    int offset,            // x7
> //    int scanPosSigOff,     // sp
>-//    int subPosBase)        // sp + 8
>+//    int subPosBase)        // sp + 8, or sp + 4 on APPLE
> function PFX(costCoeffNxN_neon)
>     // abs(coeff)
>     add             x2, x2, x2
>@@ -2410,7 +2410,7 @@ function PFX(costCoeffNxN_neon)
>     add             x4, x4, x15
>     str             h2, [x13]              // absCoeff[numNonZero] = tmpCoeff[blkPos]
>
>-    ldr             x9, [sp, #8]           // subPosBase
>+    ldr             x9, [sp, #STACK_ARG_OFFSET(1)]           // subPosBase
>     uxth            w9, w9
>     cmp             w9, #0
>     cset            x2, eq
>--
>2.42.1
>
>IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
>_______________________________________________
>x265-devel mailing list
>x265-devel at videolan.org
>https://mailman.videolan.org/listinfo/x265-devel
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20240507/9f2710bf/attachment.htm>