eessppeelllloo/ffmpeg
1
0
Fork 0
mirror of https://git.ffmpeg.org/ffmpeg.git synced 2026-06-16 04:32:47 +02:00
Code Issues Packages Projects Releases Wiki Activity

vulkan_prores_raw: use compile-time SPIR-V generation

Browse Source
This commit is contained in:
Lynne
2026-01-02 13:12:44 +01:00
parent f3b0ca4f2c
commit 026e94e339
7 changed files with 134 additions and 100 deletions
Download Patch File Download Diff File Expand all files Collapse all files
configure
Vendored
+2 -2
View File
@@ -3396,7 +3396,7 @@ mpeg4_videotoolbox_hwaccel_deps="videotoolbox"
mpeg4_videotoolbox_hwaccel_select="mpeg4_decoder"
prores_videotoolbox_hwaccel_deps="videotoolbox"
prores_videotoolbox_hwaccel_select="prores_decoder"
prores_raw_vulkan_hwaccel_deps="vulkan spirv_library"
prores_raw_vulkan_hwaccel_deps="vulkan"
prores_raw_vulkan_hwaccel_select="prores_raw_decoder"
prores_vulkan_hwaccel_deps="vulkan spirv_library"
prores_vulkan_hwaccel_select="prores_decoder"
@@ -4239,7 +4239,7 @@ cws2fws_extralibs="zlib_extralibs"
# libraries, in any order
avcodec_deps="avutil"
avcodec_suggest="libm stdatomic spirv_library"
avcodec_suggest="libm stdatomic zlib spirv_library"
avdevice_deps="avformat avcodec avutil"
avdevice_suggest="libm stdatomic"
avfilter_deps="avutil"
libavcodec/vulkan/Makefile
+2 -3
View File
@@ -12,9 +12,8 @@ OBJS-$(CONFIG_FFV1_VULKAN_HWACCEL) += vulkan/common.o \
vulkan/ffv1_common.o vulkan/ffv1_reset.o \
vulkan/ffv1_dec_setup.o vulkan/ffv1_dec.o
OBJS-$(CONFIG_PRORES_RAW_VULKAN_HWACCEL) += vulkan/common.o vulkan/dct.o \
vulkan/prores_raw_decode.o \
vulkan/prores_raw_idct.o
OBJS-$(CONFIG_PRORES_RAW_VULKAN_HWACCEL) += vulkan/prores_raw_decode.comp.spv.o \
vulkan/prores_raw_idct.comp.spv.o
OBJS-$(CONFIG_PRORES_VULKAN_HWACCEL) += vulkan/common.o vulkan/dct.o \
vulkan/prores_vld.o \
libavcodec/vulkan/common.comp
+34
View File
@@ -18,6 +18,35 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#ifndef VULKAN_COMMON_H
#define VULKAN_COMMON_H
#pragma use_vulkan_memory_model
layout (local_size_x_id = 253, local_size_y_id = 254, local_size_z_id = 255) in;
#extension GL_EXT_shader_explicit_arithmetic_types : require
#extension GL_EXT_shader_explicit_arithmetic_types_int8 : require
#extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
#extension GL_EXT_shader_explicit_arithmetic_types_int32 : require
#extension GL_EXT_shader_explicit_arithmetic_types_int64 : require
#extension GL_EXT_shader_explicit_arithmetic_types_float16 : require
#extension GL_EXT_shader_explicit_arithmetic_types_float32 : require
#extension GL_EXT_shader_explicit_arithmetic_types_float64 : require
#extension GL_EXT_shader_8bit_storage : require
#extension GL_EXT_shader_16bit_storage : require
#extension GL_EXT_shader_image_load_formatted : require
#extension GL_EXT_nonuniform_qualifier : require
#extension GL_EXT_scalar_block_layout : require
#extension GL_EXT_buffer_reference : require
#extension GL_EXT_buffer_reference2 : require
#extension GL_KHR_memory_scope_semantics : require
#extension GL_EXT_null_initializer : require
#extension GL_EXT_expect_assume : enable
#extension GL_EXT_control_flow_attributes : enable
layout(buffer_reference, buffer_reference_align = 1) buffer u8buf {
uint8_t v;
};
@@ -78,6 +107,9 @@ layout(buffer_reference, buffer_reference_align = 8) buffer u64buf {
#define ceil_rshift(a, b) \
(-((-(a)) >> (b)))
#define IS_WITHIN(v1, v2) \
((v1.x < v2.x) && (v1.y < v2.y))
/* TODO: optimize */
uint align(uint src, uint a)
{
@@ -336,3 +368,5 @@ int left_bits(in GetBitContext gb)
{
return int(gb.buf_end - gb.buf) * 8 + gb.bits_valid;
}
#endif /* VULKAN_COMMON_H */
libavcodec/vulkan/dct.comp
+13 -1
View File
@@ -31,12 +31,22 @@
* IEEE Transactions on Communications, Vol. 25, No. 9, pp 1004-1009, Sept. 1977
*/
#ifndef VULKAN_DCT_H
#define VULKAN_DCT_H
#ifndef NB_BLOCKS
#define NB_BLOCKS 1
#endif
#ifndef NB_COMPONENTS
#define NB_COMPONENTS 1
#endif
layout (constant_id = 16) const uint32_t nb_blocks = NB_BLOCKS;
layout (constant_id = 17) const uint32_t nb_components = NB_COMPONENTS;
/* Padded by 1 row to avoid bank conflicts */
shared float blocks[NB_BLOCKS][NB_COMPONENTS*8*(8 + 1)];
shared float blocks[nb_blocks][nb_components*8*(8 + 1)];
const float idct_scale[64] = {
0.1250000000000000, 0.1733799806652684, 0.1633203706095471, 0.1469844503024199,
@@ -117,3 +127,5 @@ void idct8(uint block, uint offset, uint stride)
blocks[block][6*stride + offset] = u6;
blocks[block][7*stride + offset] = u7;
}
#endif /* VULKAN_DCT_H */
libavcodec/vulkan/prores_raw_decode.comp → libavcodec/vulkan/prores_raw_decode.comp.glsl
+24
View File
@@ -20,6 +20,30 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#version 460
#pragma shader_stage(compute)
#extension GL_GOOGLE_include_directive : require
#include "common.comp"
struct TileData {
ivec2 pos;
uint offset;
uint size;
};
layout (set = 0, binding = 0) uniform writeonly uimage2D dst;
layout (set = 0, binding = 1, scalar) readonly buffer frame_data_buf {
TileData tile_data[];
};
layout (push_constant, scalar) uniform pushConstants {
u8buf pkt_data;
ivec2 frame_size;
ivec2 tile_size;
uint8_t qmat[64];
};
#define COMP_ID (gl_LocalInvocationID.x)
GetBitContext gb;
libavcodec/vulkan/prores_raw_idct.comp → libavcodec/vulkan/prores_raw_idct.comp.glsl
+25
View File
@@ -20,6 +20,31 @@
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#version 460
#pragma shader_stage(compute)
#extension GL_GOOGLE_include_directive : require
#include "common.comp"
#include "dct.comp"
struct TileData {
ivec2 pos;
uint offset;
uint size;
};
layout (set = 0, binding = 0) uniform uimage2D dst;
layout (set = 0, binding = 1, scalar) readonly buffer frame_data_buf {
TileData tile_data[];
};
layout (push_constant, scalar) uniform pushConstants {
u8buf pkt_data;
ivec2 frame_size;
ivec2 tile_size;
uint8_t qmat[64];
};
#define COMP_ID (gl_LocalInvocationID.z)
#define BLOCK_ID (gl_LocalInvocationID.y)
#define ROW_ID (gl_LocalInvocationID.x)
libavcodec/vulkan_prores_raw.c
+34 -94
View File
@@ -22,13 +22,13 @@
#include "hwaccel_internal.h"
#include "prores_raw.h"
#include "libavutil/vulkan_spirv.h"
#include "libavutil/mem.h"
extern const char *ff_source_common_comp;
extern const char *ff_source_dct_comp;
extern const char *ff_source_prores_raw_decode_comp;
extern const char *ff_source_prores_raw_idct_comp;
extern const unsigned char ff_prores_raw_decode_comp_spv_data[];
extern const unsigned int ff_prores_raw_decode_comp_spv_len;
extern const unsigned char ff_prores_raw_idct_comp_spv_data[];
extern const unsigned int ff_prores_raw_idct_comp_spv_len;
const FFVulkanDecodeDescriptor ff_vk_dec_prores_raw_desc = {
.codec_id = AV_CODEC_ID_PRORES_RAW,
@@ -287,46 +287,19 @@ fail:
static int add_common_data(AVCodecContext *avctx, FFVulkanContext *s,
FFVulkanShader *shd, int writeonly)
{
AVHWFramesContext *dec_frames_ctx;
dec_frames_ctx = (AVHWFramesContext *)avctx->hw_frames_ctx->data;
/* Common codec header */
GLSLD(ff_source_common_comp);
GLSLC(0, struct TileData { );
GLSLC(1, ivec2 pos; );
GLSLC(1, uint offset; );
GLSLC(1, uint size; );
GLSLC(0, }; );
GLSLC(0, );
GLSLC(0, layout(push_constant, scalar) uniform pushConstants { );
GLSLC(1, u8buf pkt_data; );
GLSLC(1, ivec2 frame_size; );
GLSLC(1, ivec2 tile_size; );
GLSLC(1, uint8_t qmat[64]; );
GLSLC(0, }; );
GLSLC(0, );
ff_vk_shader_add_push_const(shd, 0, sizeof(DecodePushData),
VK_SHADER_STAGE_COMPUTE_BIT);
FFVulkanDescriptorSetBinding *desc_set;
desc_set = (FFVulkanDescriptorSetBinding []) {
FFVulkanDescriptorSetBinding desc_set[] = {
{
.name = "dst",
.type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.mem_layout = ff_vk_shader_rep_fmt(dec_frames_ctx->sw_format,
FF_VK_REP_NATIVE),
.mem_quali = writeonly ? "writeonly" : NULL,
.dimensions = 2,
.stages = VK_SHADER_STAGE_COMPUTE_BIT,
.name = "dst",
.type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.stages = VK_SHADER_STAGE_COMPUTE_BIT,
},
{
.name = "frame_data_buf",
.type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.stages = VK_SHADER_STAGE_COMPUTE_BIT,
.mem_layout = "scalar",
.mem_quali = "readonly",
.buf_content = "TileData tile_data[];",
.name = "frame_data_buf",
.type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.stages = VK_SHADER_STAGE_COMPUTE_BIT,
},
};
@@ -334,74 +307,49 @@ static int add_common_data(AVCodecContext *avctx, FFVulkanContext *s,
}
static int init_decode_shader(AVCodecContext *avctx, FFVulkanContext *s,
FFVkExecPool *pool, FFVkSPIRVCompiler *spv,
FFVulkanShader *shd, int version)
FFVkExecPool *pool, FFVulkanShader *shd,
int version)
{
int err;
uint8_t *spv_data;
size_t spv_len;
void *spv_opaque = NULL;
RET(ff_vk_shader_init(s, shd, "prores_raw",
VK_SHADER_STAGE_COMPUTE_BIT,
(const char *[]) { "GL_EXT_buffer_reference",
"GL_EXT_buffer_reference2",
"GL_EXT_null_initializer" }, 3,
4, 1, 1, 0));
ff_vk_shader_load(shd, VK_SHADER_STAGE_COMPUTE_BIT, NULL,
(uint32_t []) { 4, 1, 1 }, 0);
RET(add_common_data(avctx, s, shd, 1));
add_common_data(avctx, s, shd, 1);
GLSLD(ff_source_prores_raw_decode_comp);
RET(spv->compile_shader(s, spv, shd, &spv_data, &spv_len, "main",
&spv_opaque));
RET(ff_vk_shader_link(s, shd, spv_data, spv_len, "main"));
RET(ff_vk_shader_link(s, shd,
ff_prores_raw_decode_comp_spv_data,
ff_prores_raw_decode_comp_spv_len, "main"));
RET(ff_vk_shader_register_exec(s, pool, shd));
fail:
if (spv_opaque)
spv->free_shader(spv, &spv_opaque);
return err;
}
static int init_idct_shader(AVCodecContext *avctx, FFVulkanContext *s,
FFVkExecPool *pool, FFVkSPIRVCompiler *spv,
FFVulkanShader *shd, int version)
FFVkExecPool *pool, FFVulkanShader *shd,
int version)
{
int err;
uint8_t *spv_data;
size_t spv_len;
void *spv_opaque = NULL;
SPEC_LIST_CREATE(sl, 2, 2*sizeof(uint32_t))
RET(ff_vk_shader_init(s, shd, "prores_raw",
VK_SHADER_STAGE_COMPUTE_BIT,
(const char *[]) { "GL_EXT_buffer_reference",
"GL_EXT_buffer_reference2" }, 2,
8,
version == 0 ? 8 : 16 /* Horizontal blocks */,
4 /* Components */,
0));
int nb_blocks = version == 0 ? 8 : 16;
SPEC_LIST_ADD(sl, 16, 32, nb_blocks);
SPEC_LIST_ADD(sl, 17, 32, 4); /* nb_components */
RET(add_common_data(avctx, s, shd, 0));
ff_vk_shader_load(shd, VK_SHADER_STAGE_COMPUTE_BIT, sl,
(uint32_t []) { 8, nb_blocks, 4 }, 0);
GLSLC(0, #define NB_BLOCKS 16);
GLSLC(0, #define NB_COMPONENTS 4);
GLSLD(ff_source_dct_comp);
add_common_data(avctx, s, shd, 0);
GLSLD(ff_source_prores_raw_idct_comp);
RET(spv->compile_shader(s, spv, shd, &spv_data, &spv_len, "main",
&spv_opaque));
RET(ff_vk_shader_link(s, shd, spv_data, spv_len, "main"));
RET(ff_vk_shader_link(s, shd,
ff_prores_raw_idct_comp_spv_data,
ff_prores_raw_idct_comp_spv_len, "main"));
RET(ff_vk_shader_register_exec(s, pool, shd));
fail:
if (spv_opaque)
spv->free_shader(spv, &spv_opaque);
return err;
}
@@ -423,12 +371,6 @@ static int vk_decode_prores_raw_init(AVCodecContext *avctx)
FFVulkanDecodeContext *dec = avctx->internal->hwaccel_priv_data;
ProResRAWContext *prr = avctx->priv_data;
FFVkSPIRVCompiler *spv = ff_vk_spirv_init();
if (!spv) {
av_log(avctx, AV_LOG_ERROR, "Unable to initialize SPIR-V compiler!\n");
return AVERROR_EXTERNAL;
}
err = ff_vk_decode_init(avctx);
if (err < 0)
return err;
@@ -443,14 +385,12 @@ static int vk_decode_prores_raw_init(AVCodecContext *avctx)
ctx->sd_ctx_free = &vk_decode_prores_raw_uninit;
/* Setup decode shader */
RET(init_decode_shader(avctx, &ctx->s, &ctx->exec_pool, spv, &prv->decode,
RET(init_decode_shader(avctx, &ctx->s, &ctx->exec_pool, &prv->decode,
prr->version));
RET(init_idct_shader(avctx, &ctx->s, &ctx->exec_pool, spv, &prv->idct,
RET(init_idct_shader(avctx, &ctx->s, &ctx->exec_pool, &prv->idct,
prr->version));
fail:
spv->uninit(&spv);
return err;
}