[FFmpeg-devel] [PATCH v6] Mark C globals with small code model
Pranav Kant
prka at google.com
Wed May 7 02:27:07 EEST 2025
Patch version 6:
- Added support for all `cextern` symbols.
- Added attribute_mcmodel_small attribute to other DECLARE_ASM* macros.
- Used DECLARE_ASM* macros for symbols that are being accessed from
assembly but still marked with non-ASM DECLARE macros.
On Tue, May 6, 2025 at 4:25 PM Pranav Kant <prka at google.com> wrote:
> By default, all globals in C/C++ compiled by clang are allocated
> in non-large data sections. See [1] for background on code models.
> For PIC (Position independent code), this is fine as long as binary is
> small but as binary size increases, users maybe want to use medium/large
> code models (-mcmodel=medium) which moves data in to large sections.
> As data in these large sections cannot be accessed using PIC code
> anymore (as it may be too far away), compiler ends up using a different
> instruction sequence when building C/C++ code -- using GOT to access
> these globals (which can be relaxed by linker at link time if binary
> ends up being smaller).
>
> However, hardcoded assembly (external asm files, as well as inline
> assembly) continue to access these globals defined in C/C++ files using
> older (and invalid instruction sequence). So, we mark all such globals
> with an attribute that forces them to be allocated in small sections
> allowing them to validly be accessed from the assembly code.
>
> This patch should not have any affect on builds that use small code
> model, which is the default mode.
>
> [1]
> https://eli.thegreenplace.net/2012/01/03/understanding-the-x64-code-models
> ---
> libavcodec/ac3dsp.h | 4 +-
> libavcodec/cabac.h | 4 +-
> libavcodec/h263dsp.h | 3 +-
> libavcodec/sbrdsp.h | 3 +-
> libavcodec/x86/constants.c | 4 +-
> libavcodec/x86/constants.h | 79 +++++++++++++++++----------------
> libavutil/attributes.h | 6 +++
> libavutil/attributes_internal.h | 16 +++++++
> libavutil/mem_internal.h | 16 ++++---
> 9 files changed, 84 insertions(+), 51 deletions(-)
>
> diff --git a/libavcodec/ac3dsp.h b/libavcodec/ac3dsp.h
> index b1b2bced8f..914824025f 100644
> --- a/libavcodec/ac3dsp.h
> +++ b/libavcodec/ac3dsp.h
> @@ -25,11 +25,13 @@
> #include <stddef.h>
> #include <stdint.h>
>
> +#include "libavutil/mem_internal.h"
> +
> /**
> * Number of mantissa bits written for each bap value.
> * bap values with fractional bits are set to 0 and are calculated
> separately.
> */
> -extern const uint16_t ff_ac3_bap_bits[16];
> +EXTERN_ASM_VAR(const uint16_t, ff_ac3_bap_bits)[16];
>
> typedef struct AC3DSPContext {
> /**
> diff --git a/libavcodec/cabac.h b/libavcodec/cabac.h
> index 38d06b2842..70730e4059 100644
> --- a/libavcodec/cabac.h
> +++ b/libavcodec/cabac.h
> @@ -29,7 +29,9 @@
>
> #include <stdint.h>
>
> -extern const uint8_t ff_h264_cabac_tables[512 + 4*2*64 + 4*64 + 63];
> +#include "libavutil/mem_internal.h"
> +
> +EXTERN_ASM_VAR(const uint8_t, ff_h264_cabac_tables)[512 + 4*2*64 + 4*64 +
> 63];
> #define H264_NORM_SHIFT_OFFSET 0
> #define H264_LPS_RANGE_OFFSET 512
> #define H264_MLPS_STATE_OFFSET 1024
> diff --git a/libavcodec/h263dsp.h b/libavcodec/h263dsp.h
> index 2dccd23392..fd107e7546 100644
> --- a/libavcodec/h263dsp.h
> +++ b/libavcodec/h263dsp.h
> @@ -20,8 +20,9 @@
> #define AVCODEC_H263DSP_H
>
> #include <stdint.h>
> +#include "libavutil/mem_internal.h"
>
> -extern const uint8_t ff_h263_loop_filter_strength[32];
> +EXTERN_ASM_VAR(const uint8_t, ff_h263_loop_filter_strength)[32];
>
> typedef struct H263DSPContext {
> void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
> diff --git a/libavcodec/sbrdsp.h b/libavcodec/sbrdsp.h
> index 09b2cbfc10..8ce467c0fb 100644
> --- a/libavcodec/sbrdsp.h
> +++ b/libavcodec/sbrdsp.h
> @@ -23,6 +23,7 @@
>
> #include <stdint.h>
> #include "aac_defines.h"
> +#include "libavutil/mem_internal.h"
>
> typedef struct SBRDSPContext {
> void (*sum64x5)(INTFLOAT *z);
> @@ -43,7 +44,7 @@ typedef struct SBRDSPContext {
> int kx, int m_max);
> } SBRDSPContext;
>
> -extern const INTFLOAT AAC_RENAME(ff_sbr_noise_table)[][2];
> +EXTERN_ASM_VAR(const INTFLOAT, AAC_RENAME(ff_sbr_noise_table))[][2];
>
> void AAC_RENAME(ff_sbrdsp_init)(SBRDSPContext *s);
> void ff_sbrdsp_init_arm(SBRDSPContext *s);
> diff --git a/libavcodec/x86/constants.c b/libavcodec/x86/constants.c
> index c5f3c6428e..ed0d456d99 100644
> --- a/libavcodec/x86/constants.c
> +++ b/libavcodec/x86/constants.c
> @@ -46,7 +46,7 @@ DECLARE_ALIGNED(32, const ymm_reg, ff_pw_256) = {
> 0x0100010001000100ULL, 0x010
>
> 0x0100010001000100ULL, 0x0100010001000100ULL };
> DECLARE_ALIGNED(32, const ymm_reg, ff_pw_512) = {
> 0x0200020002000200ULL, 0x0200020002000200ULL,
>
> 0x0200020002000200ULL, 0x0200020002000200ULL };
> -DECLARE_ALIGNED(16, const xmm_reg, ff_pw_1019) = {
> 0x03FB03FB03FB03FBULL, 0x03FB03FB03FB03FBULL };
> +DECLARE_ASM_ALIGNED(16, const xmm_reg, ff_pw_1019) = {
> 0x03FB03FB03FB03FBULL, 0x03FB03FB03FB03FBULL };
> DECLARE_ALIGNED(32, const ymm_reg, ff_pw_1023) = {
> 0x03ff03ff03ff03ffULL, 0x03ff03ff03ff03ffULL,
>
> 0x03ff03ff03ff03ffULL, 0x03ff03ff03ff03ffULL};
> DECLARE_ALIGNED(32, const ymm_reg, ff_pw_1024) = {
> 0x0400040004000400ULL, 0x0400040004000400ULL,
> @@ -70,7 +70,7 @@ DECLARE_ALIGNED(32, const ymm_reg, ff_pb_2) = {
> 0x0202020202020202ULL, 0x020
>
> 0x0202020202020202ULL, 0x0202020202020202ULL };
> DECLARE_ALIGNED(32, const ymm_reg, ff_pb_3) = {
> 0x0303030303030303ULL, 0x0303030303030303ULL,
>
> 0x0303030303030303ULL, 0x0303030303030303ULL };
> -DECLARE_ALIGNED(32, const xmm_reg, ff_pb_15) = {
> 0x0F0F0F0F0F0F0F0FULL, 0x0F0F0F0F0F0F0F0FULL };
> +DECLARE_ASM_ALIGNED(32, const xmm_reg, ff_pb_15) = {
> 0x0F0F0F0F0F0F0F0FULL, 0x0F0F0F0F0F0F0F0FULL };
> DECLARE_ALIGNED(32, const ymm_reg, ff_pb_80) = {
> 0x8080808080808080ULL, 0x8080808080808080ULL,
>
> 0x8080808080808080ULL, 0x8080808080808080ULL };
> DECLARE_ALIGNED(32, const ymm_reg, ff_pb_FE) = {
> 0xFEFEFEFEFEFEFEFEULL, 0xFEFEFEFEFEFEFEFEULL,
> diff --git a/libavcodec/x86/constants.h b/libavcodec/x86/constants.h
> index 4a55adb5b3..c2e0fd8f02 100644
> --- a/libavcodec/x86/constants.h
> +++ b/libavcodec/x86/constants.h
> @@ -23,49 +23,50 @@
>
> #include <stdint.h>
>
> +#include "libavutil/mem_internal.h"
> #include "libavutil/x86/asm.h"
>
> -extern const ymm_reg ff_pw_1;
> -extern const ymm_reg ff_pw_2;
> -extern const xmm_reg ff_pw_3;
> -extern const ymm_reg ff_pw_4;
> -extern const xmm_reg ff_pw_5;
> -extern const xmm_reg ff_pw_8;
> -extern const xmm_reg ff_pw_9;
> -extern const uint64_t ff_pw_15;
> -extern const xmm_reg ff_pw_16;
> -extern const xmm_reg ff_pw_18;
> -extern const xmm_reg ff_pw_20;
> -extern const xmm_reg ff_pw_32;
> -extern const uint64_t ff_pw_53;
> -extern const xmm_reg ff_pw_64;
> -extern const uint64_t ff_pw_128;
> -extern const ymm_reg ff_pw_255;
> -extern const ymm_reg ff_pw_256;
> -extern const ymm_reg ff_pw_512;
> -extern const ymm_reg ff_pw_1023;
> -extern const ymm_reg ff_pw_1024;
> -extern const ymm_reg ff_pw_2048;
> -extern const ymm_reg ff_pw_4095;
> -extern const ymm_reg ff_pw_4096;
> -extern const ymm_reg ff_pw_8192;
> -extern const ymm_reg ff_pw_m1;
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_1);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_2);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_3);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_4);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_5);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_8);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_9);
> +EXTERN_ASM_VAR(const uint64_t, ff_pw_15);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_16);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_18);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_20);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_32);
> +EXTERN_ASM_VAR(const uint64_t, ff_pw_53);
> +EXTERN_ASM_VAR(const xmm_reg, ff_pw_64);
> +EXTERN_ASM_VAR(const uint64_t, ff_pw_128);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_255);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_256);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_512);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_1023);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_1024);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_2048);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_4095);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_4096);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_8192);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pw_m1);
>
> -extern const ymm_reg ff_pb_0;
> -extern const ymm_reg ff_pb_1;
> -extern const ymm_reg ff_pb_2;
> -extern const ymm_reg ff_pb_3;
> -extern const ymm_reg ff_pb_80;
> -extern const ymm_reg ff_pb_FE;
> -extern const uint64_t ff_pb_FC;
> +EXTERN_ASM_VAR(const ymm_reg, ff_pb_0);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pb_1);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pb_2);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pb_3);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pb_80);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pb_FE);
> +EXTERN_ASM_VAR(const uint64_t, ff_pb_FC);
>
> -extern const xmm_reg ff_ps_neg;
> +EXTERN_ASM_VAR(const xmm_reg, ff_ps_neg);
>
> -extern const ymm_reg ff_pd_1;
> -extern const ymm_reg ff_pd_16;
> -extern const ymm_reg ff_pd_32;
> -extern const ymm_reg ff_pd_64;
> -extern const ymm_reg ff_pd_8192;
> -extern const ymm_reg ff_pd_65535;
> +EXTERN_ASM_VAR(const ymm_reg, ff_pd_1);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pd_16);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pd_32);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pd_64);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pd_8192);
> +EXTERN_ASM_VAR(const ymm_reg, ff_pd_65535);
>
> #endif /* AVCODEC_X86_CONSTANTS_H */
> diff --git a/libavutil/attributes.h b/libavutil/attributes.h
> index 04c615c952..dfc35fa31e 100644
> --- a/libavutil/attributes.h
> +++ b/libavutil/attributes.h
> @@ -40,6 +40,12 @@
> # define AV_HAS_BUILTIN(x) 0
> #endif
>
> +#ifdef __has_attribute
> +# define AV_HAS_ATTRIBUTE(x) __has_attribute(x)
> +#else
> +# define AV_HAS_ATTRIBUTE(x) 0
> +#endif
> +
> #ifndef av_always_inline
> #if AV_GCC_VERSION_AT_LEAST(3,1)
> # define av_always_inline __attribute__((always_inline)) inline
> diff --git a/libavutil/attributes_internal.h
> b/libavutil/attributes_internal.h
> index bc85ce77ff..c557fa0af0 100644
> --- a/libavutil/attributes_internal.h
> +++ b/libavutil/attributes_internal.h
> @@ -19,6 +19,7 @@
> #ifndef AVUTIL_ATTRIBUTES_INTERNAL_H
> #define AVUTIL_ATTRIBUTES_INTERNAL_H
>
> +#include "config.h"
> #include "attributes.h"
>
> #if (AV_GCC_VERSION_AT_LEAST(4,0) || defined(__clang__)) &&
> (defined(__ELF__) || defined(__MACH__))
> @@ -33,4 +34,19 @@
>
> #define EXTERN extern attribute_visibility_hidden
>
> +/**
> + * Some globals defined in C files are used from hardcoded asm that
> assumes small
> + * code model (that is, accessing these globals without GOT). This is a
> problem
> + * when FFMpeg is built with medium code model (-mcmodel=medium) which
> allocates
> + * all globals in a data section that's unreachable with PC relative
> instructions
> + * (small code model instruction sequence). We mark all such globals with
> this
> + * attribute_mcmodel_small to ensure assembly accessible globals continue
> to be
> + * allocated in sections reachable from PC relative instructions.
> + */
> +#if ARCH_X86_64 && defined(__ELF__) && AV_HAS_ATTRIBUTE(model)
> +# define attribute_mcmodel_small __attribute__((model("small")))
> +#else
> +# define attribute_mcmodel_small
> +#endif
> +
> #endif /* AVUTIL_ATTRIBUTES_INTERNAL_H */
> diff --git a/libavutil/mem_internal.h b/libavutil/mem_internal.h
> index c027fa51c3..c621fa861f 100644
> --- a/libavutil/mem_internal.h
> +++ b/libavutil/mem_internal.h
> @@ -29,6 +29,7 @@
> #endif
>
> #include "attributes.h"
> +#include "attributes_internal.h"
> #include "macros.h"
>
> /**
> @@ -78,16 +79,19 @@
>
> #if defined(__DJGPP__)
> #define DECLARE_ALIGNED_T(n,t,v) alignas(FFMIN(n, 16)) t v
> - #define DECLARE_ASM_ALIGNED(n,t,v) alignas(FFMIN(n, 16)) t av_used v
> - #define DECLARE_ASM_CONST(n,t,v) alignas(FFMIN(n, 16)) static
> const t av_used v
> + #define DECLARE_ASM_ALIGNED(n,t,v) alignas(FFMIN(n, 16)) t av_used
> attribute_mcmodel_small v
> + #define DECLARE_ASM_CONST(n,t,v) alignas(FFMIN(n, 16)) static
> const t av_used attribute_mcmodel_small v
> + #define EXTERN_ASM_VAR(t,v) extern t attribute_mcmodel_small v
> #elif defined(_MSC_VER)
> #define DECLARE_ALIGNED_T(n,t,v) __declspec(align(n)) t v
> - #define DECLARE_ASM_ALIGNED(n,t,v) __declspec(align(n)) t v
> - #define DECLARE_ASM_CONST(n,t,v) __declspec(align(n)) static const
> t v
> + #define DECLARE_ASM_ALIGNED(n,t,v) __declspec(align(n)) t
> attribute_mcmodel_small v
> + #define DECLARE_ASM_CONST(n,t,v) __declspec(align(n)) static const
> t attribute_mcmodel_small v
> + #define EXTERN_ASM_VAR(t,v) extern t attribute_mcmodel_small v
> #else
> #define DECLARE_ALIGNED_T(n,t,v) alignas(n) t v
> - #define DECLARE_ASM_ALIGNED(n,t,v) alignas(n) t av_used v
> - #define DECLARE_ASM_CONST(n,t,v) alignas(n) static const t av_used
> v
> + #define DECLARE_ASM_ALIGNED(n,t,v) alignas(n) t av_used
> attribute_mcmodel_small v
> + #define DECLARE_ASM_CONST(n,t,v) alignas(n) static const t av_used
> attribute_mcmodel_small v
> + #define EXTERN_ASM_VAR(t,v) extern t attribute_mcmodel_small v
> #endif
>
> #if HAVE_SIMD_ALIGN_64
> --
> 2.49.0.987.g0cc8ee98dc-goog
>
>
More information about the ffmpeg-devel
mailing list