[FFmpeg-devel] [PATCH 2/2] ffv1enc_vulkan: add support for RCT coefficient searching
Lynne
dev at lynne.ee
Wed Nov 20 09:27:02 EET 2024
---
libavcodec/ffv1enc_vulkan.c | 268 ++++++++++++++++++++-
libavcodec/vulkan/Makefile | 4 +-
libavcodec/vulkan/ffv1_enc_rct_search.comp | 139 +++++++++++
libavcodec/vulkan/ffv1_enc_setup.comp | 16 ++
4 files changed, 422 insertions(+), 5 deletions(-)
create mode 100644 libavcodec/vulkan/ffv1_enc_rct_search.comp
diff --git a/libavcodec/ffv1enc_vulkan.c b/libavcodec/ffv1enc_vulkan.c
index 3c1db9fd14..55b17f9784 100644
--- a/libavcodec/ffv1enc_vulkan.c
+++ b/libavcodec/ffv1enc_vulkan.c
@@ -36,6 +36,9 @@
#define LG_ALIGN_W 32
#define LG_ALIGN_H 32
+/* Level 4 and higher */
+#define RCT_MODEMAP_FMT AV_PIX_FMT_RGBA128
+
typedef struct VulkanEncodeFFv1Context {
FFV1Context ctx;
@@ -48,6 +51,7 @@ typedef struct VulkanEncodeFFv1Context {
FFVulkanShader setup;
FFVulkanShader reset;
+ FFVulkanShader rct_search;
FFVulkanShader rct;
FFVulkanShader enc;
@@ -73,6 +77,9 @@ typedef struct VulkanEncodeFFv1Context {
/* Intermediate frame pool */
AVBufferRef *intermediate_frames_ref;
+ /* Frame pool for RCT mode images */
+ AVBufferRef *rct_mode_frames_ref;
+
/* Representation mode */
enum FFVkShaderRepFormat rep_fmt;
@@ -92,6 +99,7 @@ extern const char *ff_source_ffv1_common_comp;
extern const char *ff_source_ffv1_reset_comp;
extern const char *ff_source_ffv1_enc_common_comp;
extern const char *ff_source_ffv1_enc_rct_comp;
+extern const char *ff_source_ffv1_enc_rct_search_comp;
extern const char *ff_source_ffv1_enc_vlc_comp;
extern const char *ff_source_ffv1_enc_ac_comp;
extern const char *ff_source_ffv1_enc_setup_comp;
@@ -122,6 +130,7 @@ typedef struct FFv1VkParameters {
int32_t sar[2];
uint32_t chroma_shift[2];
+ int32_t modemap_basis[2];
uint32_t plane_state_size;
uint32_t context_count;
@@ -154,6 +163,7 @@ static void add_push_data(FFVulkanShader *shd)
GLSLC(0, );
GLSLC(1, ivec2 sar; );
GLSLC(1, uvec2 chroma_shift; );
+ GLSLC(1, ivec2 modemap_basis; );
GLSLC(0, );
GLSLC(1, uint plane_state_size; );
GLSLC(1, uint context_count; );
@@ -179,6 +189,83 @@ static void add_push_data(FFVulkanShader *shd)
VK_SHADER_STAGE_COMPUTE_BIT);
}
+static int run_rct_search(AVCodecContext *avctx, FFVkExecContext *exec,
+ AVFrame *enc_in, VkImageView *enc_in_views,
+ AVFrame **rct_modemap_frame, VkImageView *rct_modemap_views,
+ VkImageMemoryBarrier2 *img_bar, int *nb_img_bar)
+{
+ int err;
+ VulkanEncodeFFv1Context *fv = avctx->priv_data;
+ FFV1Context *f = &fv->ctx;
+ FFVulkanFunctions *vk = &fv->s.vkfn;
+ AVHWFramesContext *src_hwfc = (AVHWFramesContext *)enc_in->hw_frames_ctx->data;
+ FFv1VkRCTParameters pd;
+
+ /* Create a temporaty frame */
+ *rct_modemap_frame = av_frame_alloc();
+ if (!(*rct_modemap_frame))
+ return AVERROR(ENOMEM);
+
+ RET(av_hwframe_get_buffer(fv->rct_mode_frames_ref,
+ *rct_modemap_frame, 0));
+ RET(ff_vk_exec_add_dep_frame(&fv->s, exec, *rct_modemap_frame,
+ VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
+ VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT));
+ RET(ff_vk_create_imageviews(&fv->s, exec, rct_modemap_views,
+ *rct_modemap_frame,
+ FF_VK_REP_UINT));
+ ff_vk_frame_barrier(&fv->s, exec, *rct_modemap_frame, img_bar, nb_img_bar,
+ VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
+ VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
+ VK_ACCESS_SHADER_WRITE_BIT,
+ VK_IMAGE_LAYOUT_GENERAL,
+ VK_QUEUE_FAMILY_IGNORED);
+
+ /* Update descriptors */
+ ff_vk_shader_update_img_array(&fv->s, exec, &fv->rct_search,
+ enc_in, enc_in_views,
+ 1, 0,
+ VK_IMAGE_LAYOUT_GENERAL,
+ VK_NULL_HANDLE);
+ ff_vk_shader_update_img_array(&fv->s, exec, &fv->rct_search,
+ *rct_modemap_frame, rct_modemap_views,
+ 1, 1,
+ VK_IMAGE_LAYOUT_GENERAL,
+ VK_NULL_HANDLE);
+
+ vk->CmdPipelineBarrier2(exec->buf, &(VkDependencyInfo) {
+ .sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO,
+ .pImageMemoryBarriers = img_bar,
+ .imageMemoryBarrierCount = *nb_img_bar,
+ });
+
+ /* Run the shader */
+ ff_vk_exec_bind_shader(&fv->s, exec, &fv->rct_search);
+ pd = (FFv1VkRCTParameters) {
+ .offset = 1 << f->bits_per_raw_sample,
+ .planar_rgb = ff_vk_mt_is_np_rgb(src_hwfc->sw_format) &&
+ (ff_vk_count_images((AVVkFrame *)enc_in->data[0]) > 1),
+ .transparency = f->transparency,
+ };
+ ff_vk_shader_update_push_const(&fv->s, exec, &fv->rct_search,
+ VK_SHADER_STAGE_COMPUTE_BIT,
+ 0, sizeof(pd), &pd);
+
+ vk->CmdDispatch(exec->buf,
+ (*rct_modemap_frame)->width,
+ (*rct_modemap_frame)->height, 1);
+
+ ff_vk_frame_barrier(&fv->s, exec, *rct_modemap_frame, img_bar, nb_img_bar,
+ VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
+ VK_PIPELINE_STAGE_2_COMPUTE_SHADER_BIT,
+ VK_ACCESS_SHADER_READ_BIT,
+ VK_IMAGE_LAYOUT_GENERAL,
+ VK_QUEUE_FAMILY_IGNORED);
+
+fail:
+ return err;
+}
+
static int run_rct(AVCodecContext *avctx, FFVkExecContext *exec,
AVFrame *enc_in, VkImageView *enc_in_views,
AVFrame **intermediate_frame, VkImageView *intermediate_views,
@@ -285,6 +372,7 @@ static int vulkan_encode_ffv1_frame(AVCodecContext *avctx, AVPacket *pkt,
FFv1VkParameters pd;
AVFrame *intermediate_frame = NULL;
+ AVFrame *rct_modemap_frame = NULL;
/* Temporary data */
size_t tmp_data_size;
@@ -317,6 +405,7 @@ static int vulkan_encode_ffv1_frame(AVCodecContext *avctx, AVPacket *pkt,
VkImageView in_views[AV_NUM_DATA_POINTERS];
VkImageView intermediate_views[AV_NUM_DATA_POINTERS];
+ VkImageView rct_modemap_views[AV_NUM_DATA_POINTERS];
AVFrame *enc_in = (AVFrame *)pict;
VkImageView *enc_in_views = in_views;
@@ -475,6 +564,19 @@ static int vulkan_encode_ffv1_frame(AVCodecContext *avctx, AVPacket *pkt,
};
}
+ if (fv->is_rgb && f->version >= 4) {
+ RET(run_rct_search(avctx, exec,
+ enc_in, enc_in_views,
+ &rct_modemap_frame, rct_modemap_views,
+ img_bar, &nb_img_bar));
+
+ ff_vk_shader_update_img_array(&fv->s, exec, &fv->setup,
+ rct_modemap_frame, rct_modemap_views,
+ 1, 2,
+ VK_IMAGE_LAYOUT_GENERAL,
+ VK_NULL_HANDLE);
+ }
+
vk->CmdPipelineBarrier2(exec->buf, &(VkDependencyInfo) {
.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO,
.pImageMemoryBarriers = img_bar,
@@ -501,6 +603,8 @@ static int vulkan_encode_ffv1_frame(AVCodecContext *avctx, AVPacket *pkt,
.sar[1] = pict->sample_aspect_ratio.den,
.chroma_shift[0] = f->chroma_h_shift,
.chroma_shift[1] = f->chroma_v_shift,
+ .modemap_basis[0] = fv->rct_search.lg_size[0],
+ .modemap_basis[1] = fv->rct_search.lg_size[1],
.plane_state_size = plane_state_size,
.context_count = context_count,
.crcref = f->crcref,
@@ -652,6 +756,7 @@ static int vulkan_encode_ffv1_frame(AVCodecContext *avctx, AVPacket *pkt,
/* We need the encoded data immediately */
ff_vk_exec_wait(&fv->s, exec);
av_frame_free(&intermediate_frame);
+ av_frame_free(&rct_modemap_frame);
/* Invalidate slice/output data if needed */
if (!(results_data_buf->flags & VK_MEMORY_PROPERTY_HOST_COHERENT_BIT)) {
@@ -741,6 +846,7 @@ fail:
/* Frames added as a dep are always referenced, so we only need to
* clean this up. */
av_frame_free(&intermediate_frame);
+ av_frame_free(&rct_modemap_frame);
return 0;
}
@@ -752,6 +858,10 @@ static int init_indirect(AVCodecContext *avctx, enum AVPixelFormat sw_format)
AVHWFramesContext *frames_ctx;
AVVulkanFramesContext *vk_frames;
+ int subgroup_size = fv->s.props_11.subgroupSize;
+ int lg_rows = fv->s.props.properties.limits.maxComputeWorkGroupInvocations /
+ subgroup_size;
+
fv->intermediate_frames_ref = av_hwframe_ctx_alloc(fv->s.device_ref);
if (!fv->intermediate_frames_ref)
return AVERROR(ENOMEM);
@@ -759,8 +869,8 @@ static int init_indirect(AVCodecContext *avctx, enum AVPixelFormat sw_format)
frames_ctx = (AVHWFramesContext *)fv->intermediate_frames_ref->data;
frames_ctx->format = AV_PIX_FMT_VULKAN;
frames_ctx->sw_format = sw_format;
- frames_ctx->width = FFALIGN(fv->s.frames->width, 32);
- frames_ctx->height = FFALIGN(fv->s.frames->height, 32);
+ frames_ctx->width = FFALIGN(fv->s.frames->width, FFMAX(subgroup_size, 32));
+ frames_ctx->height = FFALIGN(fv->s.frames->height, FFMAX(lg_rows, 32));
vk_frames = frames_ctx->hwctx;
vk_frames->tiling = VK_IMAGE_TILING_OPTIMAL;
@@ -826,6 +936,39 @@ end:
return fmt;
}
+static int init_modemap(AVCodecContext *avctx, enum AVPixelFormat sw_format,
+ int lg_size0, int lg_size1)
+{
+ int err;
+ VulkanEncodeFFv1Context *fv = avctx->priv_data;
+ AVHWFramesContext *frames_ctx;
+ AVVulkanFramesContext *vk_frames;
+
+ fv->rct_mode_frames_ref = av_hwframe_ctx_alloc(fv->s.device_ref);
+ if (!fv->rct_mode_frames_ref)
+ return AVERROR(ENOMEM);
+
+ frames_ctx = (AVHWFramesContext *)fv->rct_mode_frames_ref->data;
+ frames_ctx->format = AV_PIX_FMT_VULKAN;
+ frames_ctx->sw_format = sw_format;
+ frames_ctx->width = (FFALIGN(fv->s.frames->width, lg_size0)/lg_size0) + 20;
+ frames_ctx->height = (FFALIGN(fv->s.frames->height, lg_size1)/lg_size1) + 20;
+
+ vk_frames = frames_ctx->hwctx;
+ vk_frames->tiling = VK_IMAGE_TILING_OPTIMAL;
+ vk_frames->usage = VK_IMAGE_USAGE_STORAGE_BIT;
+
+ err = av_hwframe_ctx_init(fv->rct_mode_frames_ref);
+ if (err < 0) {
+ av_log(avctx, AV_LOG_ERROR, "Unable to initialize modemap pool with format %s: %s\n",
+ av_get_pix_fmt_name(sw_format), av_err2str(err));
+ av_buffer_unref(&fv->rct_mode_frames_ref);
+ return err;
+ }
+
+ return 0;
+}
+
static void define_shared_code(AVCodecContext *avctx, FFVulkanShader *shd)
{
VulkanEncodeFFv1Context *fv = avctx->priv_data;
@@ -912,8 +1055,18 @@ static int init_setup_shader(AVCodecContext *avctx, FFVkSPIRVCompiler *spv)
.mem_quali = "readonly",
.stages = VK_SHADER_STAGE_COMPUTE_BIT,
},
+ {
+ .name = "modemap",
+ .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .dimensions = 2,
+ .mem_layout = ff_vk_shader_rep_fmt(RCT_MODEMAP_FMT,
+ FF_VK_REP_UINT),
+ .elems = av_pix_fmt_count_planes(RCT_MODEMAP_FMT),
+ .mem_quali = "readonly",
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ },
};
- RET(ff_vk_shader_add_descriptor_set(&fv->s, shd, desc_set, 2, 0, 0));
+ RET(ff_vk_shader_add_descriptor_set(&fv->s, shd, desc_set, 3, 0, 0));
add_push_data(shd);
@@ -1013,6 +1166,105 @@ fail:
return err;
}
+static int init_rct_search_shader(AVCodecContext *avctx, FFVkSPIRVCompiler *spv)
+{
+ int err;
+ VulkanEncodeFFv1Context *fv = avctx->priv_data;
+ FFVulkanShader *shd = &fv->rct_search;
+ FFVulkanDescriptorSetBinding *desc_set;
+ int subgroup_size = fv->s.props_11.subgroupSize;
+ int lg_rows = fv->s.props.properties.limits.maxComputeWorkGroupInvocations /
+ subgroup_size;
+
+ uint8_t *spv_data;
+ size_t spv_len;
+ void *spv_opaque = NULL;
+
+ RET(init_modemap(avctx, RCT_MODEMAP_FMT, subgroup_size, lg_rows));
+
+ RET(ff_vk_shader_init(&fv->s, shd, "ffv1_rct_search",
+ VK_SHADER_STAGE_COMPUTE_BIT,
+ (const char *[]) { "GL_EXT_null_initializer",
+ "GL_KHR_shader_subgroup_basic",
+ "GL_KHR_shader_subgroup_arithmetic",
+ "GL_EXT_buffer_reference",
+ "GL_EXT_buffer_reference2" }, 5,
+ subgroup_size, lg_rows, 1,
+ 0));
+
+ av_bprintf(&shd->src, "#define MAX_QUANT_TABLES %i\n", MAX_QUANT_TABLES);
+ av_bprintf(&shd->src, "#define MAX_CONTEXT_INPUTS %i\n", MAX_CONTEXT_INPUTS);
+ av_bprintf(&shd->src, "#define MAX_QUANT_TABLE_SIZE %i\n", MAX_QUANT_TABLE_SIZE);
+
+ desc_set = (FFVulkanDescriptorSetBinding []) {
+ {
+ .name = "rangecoder_static_buf",
+ .type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ .mem_layout = "scalar",
+ .buf_content = "uint8_t zero_one_state[512];",
+ },
+ {
+ .name = "quant_buf",
+ .type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ .mem_layout = "scalar",
+ .buf_content = "int16_t quant_table[MAX_QUANT_TABLES]"
+ "[MAX_CONTEXT_INPUTS][MAX_QUANT_TABLE_SIZE];",
+ },
+ };
+ RET(ff_vk_shader_add_descriptor_set(&fv->s, shd, desc_set, 2, 1, 0));
+
+ define_shared_code(avctx, shd);
+
+ desc_set = (FFVulkanDescriptorSetBinding []) {
+ {
+ .name = "src",
+ .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .dimensions = 2,
+ .mem_layout = ff_vk_shader_rep_fmt(fv->s.frames->sw_format,
+ fv->rep_fmt),
+ .elems = av_pix_fmt_count_planes(fv->s.frames->sw_format),
+ .mem_quali = "readonly",
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ },
+ {
+ .name = "modemap",
+ .type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
+ .dimensions = 2,
+ .mem_layout = ff_vk_shader_rep_fmt(RCT_MODEMAP_FMT,
+ FF_VK_REP_UINT),
+ .elems = av_pix_fmt_count_planes(RCT_MODEMAP_FMT),
+ .mem_quali = "writeonly",
+ .stages = VK_SHADER_STAGE_COMPUTE_BIT,
+ },
+ };
+ RET(ff_vk_shader_add_descriptor_set(&fv->s, shd, desc_set, 2, 0, 0));
+
+ GLSLC(0, layout(push_constant, scalar) uniform pushConstants { );
+ GLSLC(1, int offset; );
+ GLSLC(1, uint8_t planar_rgb; );
+ GLSLC(1, uint8_t transparency; );
+ GLSLC(1, uint8_t padding[2]; );
+ GLSLC(0, }; );
+ ff_vk_shader_add_push_const(shd, 0, sizeof(FFv1VkRCTParameters),
+ VK_SHADER_STAGE_COMPUTE_BIT);
+
+ GLSLD(ff_source_ffv1_enc_rct_search_comp);
+
+ RET(spv->compile_shader(&fv->s, spv, shd, &spv_data, &spv_len, "main",
+ &spv_opaque));
+ RET(ff_vk_shader_link(&fv->s, shd, spv_data, spv_len, "main"));
+
+ RET(ff_vk_shader_register_exec(&fv->s, &fv->exec_pool, shd));
+
+fail:
+ if (spv_opaque)
+ spv->free_shader(spv, &spv_opaque);
+
+ return err;
+}
+
static int init_rct_shader(AVCodecContext *avctx, FFVkSPIRVCompiler *spv)
{
int err;
@@ -1506,6 +1758,14 @@ static av_cold int vulkan_encode_ffv1_init(AVCodecContext *avctx)
/* Init RCT shader */
if (fv->is_rgb) {
+ if (f->version >= 4) {
+ err = init_rct_search_shader(avctx, spv);
+ if (err < 0) {
+ spv->uninit(&spv);
+ return err;
+ }
+ }
+
err = init_rct_shader(avctx, spv);
if (err < 0) {
spv->uninit(&spv);
@@ -1548,9 +1808,11 @@ static av_cold int vulkan_encode_ffv1_close(AVCodecContext *avctx)
ff_vk_shader_free(&fv->s, &fv->enc);
ff_vk_shader_free(&fv->s, &fv->rct);
+ ff_vk_shader_free(&fv->s, &fv->rct_search);
ff_vk_shader_free(&fv->s, &fv->reset);
ff_vk_shader_free(&fv->s, &fv->setup);
+ av_buffer_unref(&fv->rct_mode_frames_ref);
av_buffer_unref(&fv->intermediate_frames_ref);
av_buffer_pool_uninit(&fv->results_data_pool);
diff --git a/libavcodec/vulkan/Makefile b/libavcodec/vulkan/Makefile
index 351332ee44..6b6eedda4d 100644
--- a/libavcodec/vulkan/Makefile
+++ b/libavcodec/vulkan/Makefile
@@ -6,8 +6,8 @@ clean::
OBJS-$(CONFIG_FFV1_VULKAN_ENCODER) += vulkan/common.o \
vulkan/rangecoder.o vulkan/ffv1_vlc.o \
vulkan/ffv1_common.o vulkan/ffv1_reset.o \
- vulkan/ffv1_enc_common.o \
- vulkan/ffv1_enc_rct.o vulkan/ffv1_enc_setup.o \
+ vulkan/ffv1_enc_common.o vulkan/ffv1_enc_setup.o \
+ vulkan/ffv1_enc_rct.o vulkan/ffv1_enc_rct_search.o \
vulkan/ffv1_enc_vlc.o vulkan/ffv1_enc_ac.o \
vulkan/ffv1_enc.o vulkan/ffv1_enc_rgb.o
diff --git a/libavcodec/vulkan/ffv1_enc_rct_search.comp b/libavcodec/vulkan/ffv1_enc_rct_search.comp
new file mode 100644
index 0000000000..ad251b8588
--- /dev/null
+++ b/libavcodec/vulkan/ffv1_enc_rct_search.comp
@@ -0,0 +1,139 @@
+/*
+ * FFv1 codec
+ *
+ * Copyright (c) 2024 Lynne <dev at lynne.ee>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+ivec4 load_components(void)
+{
+ const ivec2 pos = ivec2(gl_GlobalInvocationID);
+
+ if (planar_rgb == 0)
+ return ivec4(imageLoad(src[0], pos));
+
+ ivec4 pix;
+ for (int i = 0; i < (3 + transparency); i++)
+ pix[i] = int(imageLoad(src[i], pos)[0]);
+
+ /* Swizzle out the difference */
+ if (transparency > 0)
+ return pix.brga;
+ return pix.bgra;
+}
+
+#define NUM_CHECKS 15
+const ivec2 rct_y_coeff[] = {
+ ivec2(0, 0), // 4G
+
+ ivec2(0, 1), // 3G + B
+ ivec2(1, 0), // R + 3G
+ ivec2(1, 1), // R + 2G + B
+
+ ivec2(0, 2), // 2G + 2B
+ ivec2(2, 0), // 2R + 2G
+ ivec2(2, 2), // 2R + 2B
+
+ ivec2(0, 3), // 1G + 3B
+ ivec2(3, 0), // 3R + 1G
+
+ ivec2(0, 4), // 4B
+ ivec2(4, 0), // 4R
+
+ ivec2(1, 2), // R + G + 2B
+ ivec2(2, 1), // 2R + G + B
+
+ ivec2(3, 1), // 3R + B
+ ivec2(1, 3), // R + 3B
+};
+
+shared ivec4 pix_buf[gl_WorkGroupSize.x + 1][gl_WorkGroupSize.y + 1] = { };
+
+ivec4 transform_sample(ivec4 pix, ivec2 rct_coef)
+{
+ pix.b -= pix.g;
+ pix.r -= pix.g;
+ pix.g += (pix.r*rct_coef.x + pix.b*rct_coef.y) >> 2;
+ pix.b += offset;
+ pix.r += offset;
+ return pix;
+}
+
+uint get_dist(ivec4 cur)
+{
+ ivec4 LL = pix_buf[gl_LocalInvocationID.x + 0][gl_LocalInvocationID.y + 1];
+ ivec4 TL = pix_buf[gl_LocalInvocationID.x + 0][gl_LocalInvocationID.y + 0];
+ ivec4 TT = pix_buf[gl_LocalInvocationID.x + 1][gl_LocalInvocationID.y + 0];
+
+ ivec4 pred = ivec4(predict(LL.r, ivec2(TL.r, TT.r)),
+ predict(LL.g, ivec2(TL.g, TT.g)),
+ predict(LL.b, ivec2(TL.b, TT.b)),
+ predict(LL.a, ivec2(TL.a, TT.a)));
+
+ uvec4 c = abs(cur - pred);
+ return c.r + c.g + c.b + c.a;
+}
+
+shared uint score_cols[gl_WorkGroupSize.y] = { };
+
+void coeff_rating(void)
+{
+ ivec4 pix = load_components();
+ uint min_sum = 0xFFFFFFFF;
+ int best_mode = 1;
+
+ for (int i = 0; i < NUM_CHECKS; i++) {
+ ivec4 tx_pix = transform_sample(pix, rct_y_coeff[i]);
+ pix_buf[gl_LocalInvocationID.x + 1][gl_LocalInvocationID.y + 1] = tx_pix;
+
+ memoryBarrierShared();
+
+ uint dist = get_dist(tx_pix);
+
+ /* Sum from all columns */
+ uint col_sum = subgroupAdd(dist);
+
+ if (gl_LocalInvocationID.x == 0 && gl_LocalInvocationID.y == 0)
+ score_cols[gl_LocalInvocationID.y] = col_sum;
+
+ memoryBarrierShared();
+
+ /* Sum row-wise */
+ uint row_sum = 0;
+ for (uint j = gl_LocalInvocationID.x; j < gl_WorkGroupSize.y; j += gl_WorkGroupSize.x)
+ row_sum += score_cols[j];
+
+ uint block_sum = subgroupAdd(row_sum);
+ if (block_sum < min_sum) {
+ min_sum = block_sum;
+ best_mode = i;
+ }
+ }
+
+ if (gl_LocalInvocationID.x == 0 && gl_LocalInvocationID.y == 0)
+ imageStore(modemap[0], ivec2(gl_WorkGroupID),
+ uvec4(rct_y_coeff[best_mode].x,
+ rct_y_coeff[best_mode].y,
+ min_sum,
+ 0));
+}
+
+void main(void)
+{
+ coeff_rating();
+}
diff --git a/libavcodec/vulkan/ffv1_enc_setup.comp b/libavcodec/vulkan/ffv1_enc_setup.comp
index b861e25f74..d9bc2b453f 100644
--- a/libavcodec/vulkan/ffv1_enc_setup.comp
+++ b/libavcodec/vulkan/ffv1_enc_setup.comp
@@ -53,6 +53,22 @@ void init_slice(out SliceContext sc, const uint slice_idx)
sc.slice_rct_coef = ivec2(1, 1);
sc.slice_coding_mode = int(force_pcm == 1);
+ if (version >= 4) {
+ ivec2 modemap_pos = sc.slice_pos / modemap_basis;
+ ivec2 modemap_end = (sc.slice_pos + sc.slice_dim) / modemap_basis;
+
+ /* Pick the lowest one amongst all blocks within the image */
+ uvec4 res = uvec4(1, 1, 0xFFFFFFFF, 0);
+ for (; modemap_pos.y < modemap_end.y; modemap_pos.y++) {
+ for (; modemap_pos.x < modemap_end.x; modemap_pos.x++) {
+ uvec4 tmp = imageLoad(modemap[0], modemap_pos);
+ if (tmp.z < res.z)
+ res = tmp;
+ }
+ }
+ sc.slice_rct_coef = ivec2(res.xy);
+ }
+
rac_init(sc.c,
OFFBUF(u8buf, out_data, slice_idx * slice_size_max),
slice_size_max);
--
2.45.2.753.g447d99e1c3b
More information about the ffmpeg-devel
mailing list