* DEALINGS IN THE SOFTWARE.
*/
+#include <float.h>
#include <stdio.h>
#include <string.h>
#include "scale_eval.h"
#include "video.h"
+#include "vf_scale_cuda.h"
+
static const enum AVPixelFormat supported_formats[] = {
AV_PIX_FMT_YUV420P,
AV_PIX_FMT_NV12,
AV_PIX_FMT_P010,
AV_PIX_FMT_P016,
AV_PIX_FMT_YUV444P16,
+ AV_PIX_FMT_0RGB32,
+ AV_PIX_FMT_0BGR32,
};
#define DIV_UP(a, b) ( ((a) + (b) - 1) / (b) )
enum {
INTERP_ALGO_DEFAULT,
+ INTERP_ALGO_NEAREST,
INTERP_ALGO_BILINEAR,
INTERP_ALGO_BICUBIC,
+ INTERP_ALGO_LANCZOS,
INTERP_ALGO_COUNT
};
int interp_algo;
int interp_use_linear;
int interp_as_integer;
+
+ float param;
} CUDAScaleContext;
static av_cold int cudascale_init(AVFilterContext *ctx)
extern char vf_scale_cuda_bicubic_ptx[];
switch(s->interp_algo) {
+ case INTERP_ALGO_NEAREST:
+ scaler_ptx = vf_scale_cuda_ptx;
+ function_infix = "_Nearest";
+ s->interp_use_linear = 0;
+ s->interp_as_integer = 1;
+ break;
case INTERP_ALGO_BILINEAR:
scaler_ptx = vf_scale_cuda_ptx;
function_infix = "_Bilinear";
s->interp_use_linear = 0;
s->interp_as_integer = 0;
break;
+ case INTERP_ALGO_LANCZOS:
+ scaler_ptx = vf_scale_cuda_bicubic_ptx;
+ function_infix = "_Lanczos";
+ s->interp_use_linear = 0;
+ s->interp_as_integer = 0;
+ break;
default:
av_log(ctx, AV_LOG_ERROR, "Unknown interpolation algorithm\n");
return AVERROR_BUG;
CudaFunctions *cu = s->hwctx->internal->cuda_dl;
CUdeviceptr dst_devptr = (CUdeviceptr)dst_dptr;
CUtexObject tex = 0;
- void *args_uchar[] = { &tex, &dst_devptr, &dst_width, &dst_height, &dst_pitch, &src_width, &src_height, &bit_depth };
+ void *args_uchar[] = { &tex, &dst_devptr, &dst_width, &dst_height, &dst_pitch,
+ &src_width, &src_height, &bit_depth, &s->param };
int ret;
CUDA_TEXTURE_DESC tex_desc = {
.res.pitch2D.numChannels = channels,
.res.pitch2D.width = src_width,
.res.pitch2D.height = src_height,
- .res.pitch2D.pitchInBytes = src_pitch * pixel_size,
+ .res.pitch2D.pitchInBytes = src_pitch,
.res.pitch2D.devPtr = (CUdeviceptr)src_dptr,
};
+ // Handling of channels is done via vector-types in cuda, so their size is implicitly part of the pitch
+ // Same for pixel_size, which is represented via datatypes on the cuda side of things.
+ dst_pitch /= channels * pixel_size;
+
ret = CHECK_CU(cu->cuTexObjectCreate(&tex, &res_desc, &tex_desc, NULL));
if (ret < 0)
goto exit;
break;
case AV_PIX_FMT_YUV444P16:
call_resize_kernel(ctx, s->cu_func_ushort, 1,
- in->data[0], in->width, in->height, in->linesize[0] / 2,
- out->data[0], out->width, out->height, out->linesize[0] / 2,
+ in->data[0], in->width, in->height, in->linesize[0],
+ out->data[0], out->width, out->height, out->linesize[0],
2, 16);
call_resize_kernel(ctx, s->cu_func_ushort, 1,
- in->data[1], in->width, in->height, in->linesize[1] / 2,
- out->data[1], out->width, out->height, out->linesize[1] / 2,
+ in->data[1], in->width, in->height, in->linesize[1],
+ out->data[1], out->width, out->height, out->linesize[1],
2, 16);
call_resize_kernel(ctx, s->cu_func_ushort, 1,
- in->data[2], in->width, in->height, in->linesize[2] / 2,
- out->data[2], out->width, out->height, out->linesize[2] / 2,
+ in->data[2], in->width, in->height, in->linesize[2],
+ out->data[2], out->width, out->height, out->linesize[2],
2, 16);
break;
case AV_PIX_FMT_NV12:
1, 8);
call_resize_kernel(ctx, s->cu_func_uchar2, 2,
in->data[1], in->width / 2, in->height / 2, in->linesize[1],
- out->data[1], out->width / 2, out->height / 2, out->linesize[1] / 2,
+ out->data[1], out->width / 2, out->height / 2, out->linesize[1],
1, 8);
break;
case AV_PIX_FMT_P010LE:
call_resize_kernel(ctx, s->cu_func_ushort, 1,
- in->data[0], in->width, in->height, in->linesize[0] / 2,
- out->data[0], out->width, out->height, out->linesize[0] / 2,
+ in->data[0], in->width, in->height, in->linesize[0],
+ out->data[0], out->width, out->height, out->linesize[0],
2, 10);
call_resize_kernel(ctx, s->cu_func_ushort2, 2,
- in->data[1], in->width / 2, in->height / 2, in->linesize[1] / 2,
- out->data[1], out->width / 2, out->height / 2, out->linesize[1] / 4,
+ in->data[1], in->width / 2, in->height / 2, in->linesize[1],
+ out->data[1], out->width / 2, out->height / 2, out->linesize[1],
2, 10);
break;
case AV_PIX_FMT_P016LE:
call_resize_kernel(ctx, s->cu_func_ushort, 1,
- in->data[0], in->width, in->height, in->linesize[0] / 2,
- out->data[0], out->width, out->height, out->linesize[0] / 2,
+ in->data[0], in->width, in->height, in->linesize[0],
+ out->data[0], out->width, out->height, out->linesize[0],
2, 16);
call_resize_kernel(ctx, s->cu_func_ushort2, 2,
- in->data[1], in->width / 2, in->height / 2, in->linesize[1] / 2,
- out->data[1], out->width / 2, out->height / 2, out->linesize[1] / 4,
+ in->data[1], in->width / 2, in->height / 2, in->linesize[1],
+ out->data[1], out->width / 2, out->height / 2, out->linesize[1],
2, 16);
break;
+ case AV_PIX_FMT_0RGB32:
+ case AV_PIX_FMT_0BGR32:
+ call_resize_kernel(ctx, s->cu_func_uchar4, 4,
+ in->data[0], in->width, in->height, in->linesize[0],
+ out->data[0], out->width, out->height, out->linesize[0],
+ 1, 8);
+ break;
default:
return AVERROR_BUG;
}
#define OFFSET(x) offsetof(CUDAScaleContext, x)
#define FLAGS (AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM)
static const AVOption options[] = {
- { "w", "Output video width", OFFSET(w_expr), AV_OPT_TYPE_STRING, { .str = "iw" }, .flags = FLAGS },
- { "h", "Output video height", OFFSET(h_expr), AV_OPT_TYPE_STRING, { .str = "ih" }, .flags = FLAGS },
+ { "w", "Output video width", OFFSET(w_expr), AV_OPT_TYPE_STRING, { .str = "iw" }, .flags = FLAGS },
+ { "h", "Output video height", OFFSET(h_expr), AV_OPT_TYPE_STRING, { .str = "ih" }, .flags = FLAGS },
{ "interp_algo", "Interpolation algorithm used for resizing", OFFSET(interp_algo), AV_OPT_TYPE_INT, { .i64 = INTERP_ALGO_DEFAULT }, 0, INTERP_ALGO_COUNT - 1, FLAGS, "interp_algo" },
+ { "nearest", "nearest neighbour", 0, AV_OPT_TYPE_CONST, { .i64 = INTERP_ALGO_NEAREST }, 0, 0, FLAGS, "interp_algo" },
{ "bilinear", "bilinear", 0, AV_OPT_TYPE_CONST, { .i64 = INTERP_ALGO_BILINEAR }, 0, 0, FLAGS, "interp_algo" },
{ "bicubic", "bicubic", 0, AV_OPT_TYPE_CONST, { .i64 = INTERP_ALGO_BICUBIC }, 0, 0, FLAGS, "interp_algo" },
+ { "lanczos", "lanczos", 0, AV_OPT_TYPE_CONST, { .i64 = INTERP_ALGO_LANCZOS }, 0, 0, FLAGS, "interp_algo" },
{ "passthrough", "Do not process frames at all if parameters match", OFFSET(passthrough), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, FLAGS },
+ { "param", "Algorithm-Specific parameter", OFFSET(param), AV_OPT_TYPE_FLOAT, { .dbl = SCALE_CUDA_PARAM_DEFAULT }, -FLT_MAX, FLT_MAX, FLAGS },
{ "force_original_aspect_ratio", "decrease or increase w/h if necessary to keep the original AR", OFFSET(force_original_aspect_ratio), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 2, FLAGS, "force_oar" },
{ "disable", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, 0, 0, FLAGS, "force_oar" },
{ "decrease", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, 0, 0, FLAGS, "force_oar" },
{ "increase", NULL, 0, AV_OPT_TYPE_CONST, {.i64 = 2 }, 0, 0, FLAGS, "force_oar" },
- { "force_divisible_by", "enforce that the output resolution is divisible by a defined integer when force_original_aspect_ratio is used", OFFSET(force_divisible_by), AV_OPT_TYPE_INT, { .i64 = 1}, 1, 256, FLAGS },
+ { "force_divisible_by", "enforce that the output resolution is divisible by a defined integer when force_original_aspect_ratio is used", OFFSET(force_divisible_by), AV_OPT_TYPE_INT, { .i64 = 1 }, 1, 256, FLAGS },
{ NULL },
};
{ NULL }
};
-AVFilter ff_vf_scale_cuda = {
+const AVFilter ff_vf_scale_cuda = {
.name = "scale_cuda",
.description = NULL_IF_CONFIG_SMALL("GPU accelerated video resizer"),
projects / ffmpeg / blobdiff