avcodec/nvenc: add 4:2:2 encoding and H.264 10-bit support

This commit adds support for 4:2:2 encoding for HEVC and H.264 on NVIDIA Blackwell GPUs. Additionally, it supports 10-bit encoding for H.264 on Blackwell GPUs.

Diego de Souza committed 1 year ago

Commit 2cfef291 parent 7e965583 files · +83 −6Browse files

Showing the changes to libavcodec (3 of 3 files). Show all

Unified · Split

3 files

libavcodec/nvenc.c +62 −6
libavcodec/nvenc.h +12 −0
libavcodec/nvenc_h264.c +9 −0

libavcodec/nvenc.c+62−6

     AV_PIX_FMT_P010,
     AV_PIX_FMT_YUV444P,
     AV_PIX_FMT_P016,      // Truncated to 10bits
+#ifdef NVENC_HAVE_422_SUPPORT
+    AV_PIX_FMT_NV16,
+    AV_PIX_FMT_P210,
+    AV_PIX_FMT_P216,
+#endif
     AV_PIX_FMT_YUV444P16, // Truncated to 10bits
     AV_PIX_FMT_0RGB32,
     AV_PIX_FMT_RGB32,
 #define IS_10BIT(pix_fmt)  (pix_fmt == AV_PIX_FMT_P010      || \
                             pix_fmt == AV_PIX_FMT_P016      || \
+                            pix_fmt == AV_PIX_FMT_P210      || \
+                            pix_fmt == AV_PIX_FMT_P216      || \
                             pix_fmt == AV_PIX_FMT_YUV444P16 || \
                             pix_fmt == AV_PIX_FMT_X2RGB10   || \
                             pix_fmt == AV_PIX_FMT_X2BGR10   || \
                             pix_fmt == AV_PIX_FMT_GBRP16    || \
                             (ctx->rgb_mode == NVENC_RGB_MODE_444 && IS_RGB(pix_fmt)))
+#define IS_YUV422(pix_fmt) (pix_fmt == AV_PIX_FMT_NV16 || \
+                            pix_fmt == AV_PIX_FMT_P210 || \
+                            pix_fmt == AV_PIX_FMT_P216)
 #define IS_GBRP(pix_fmt) (pix_fmt == AV_PIX_FMT_GBRP || \
                           pix_fmt == AV_PIX_FMT_GBRP16)
         return AVERROR(ENOSYS);
     }
+#ifdef NVENC_HAVE_422_SUPPORT
+    ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_YUV422_ENCODE);
+#else
+    ret = 0;
+#endif
+    if (IS_YUV422(ctx->data_pix_fmt) && ret <= 0) {
+        av_log(avctx, AV_LOG_WARNING, "YUV422P not supported\n");
+        return AVERROR(ENOSYS);
+    }
     ret = nvenc_check_cap(avctx, NV_ENC_CAPS_SUPPORT_LOSSLESS_ENCODE);
     if (ctx->flags & NVENC_LOSSLESS && ret <= 0) {
         av_log(avctx, AV_LOG_WARNING, "Lossless encoding not supported\n");
             cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_GUID;
             avctx->profile = AV_PROFILE_H264_HIGH;
             break;
+#ifdef NVENC_HAVE_H264_10BIT_SUPPORT
+        case NV_ENC_H264_PROFILE_HIGH_10:
+            cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_10_GUID;
+            avctx->profile = AV_PROFILE_H264_HIGH_10;
+            break;
+#endif
+#ifdef NVENC_HAVE_422_SUPPORT
+        case NV_ENC_H264_PROFILE_HIGH_422:
+            cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_422_GUID;
+            avctx->profile = AV_PROFILE_H264_HIGH_422;
+            break;
+#endif
         case NV_ENC_H264_PROFILE_HIGH_444P:
             cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
             avctx->profile = AV_PROFILE_H264_HIGH_444_PREDICTIVE;
         }
     }
+#ifdef NVENC_HAVE_H264_10BIT_SUPPORT
+    // force setting profile as high10 if input is 10 bit or if it should be encoded as 10 bit
+    if (IS_10BIT(ctx->data_pix_fmt) || ctx->highbitdepth) {
+        cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_10_GUID;
+        avctx->profile = AV_PROFILE_H264_HIGH_10;
+    }
+#endif
     // force setting profile as high444p if input is AV_PIX_FMT_YUV444P
     if (IS_YUV444(ctx->data_pix_fmt)) {
         cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_444_GUID;
         avctx->profile = AV_PROFILE_H264_HIGH_444_PREDICTIVE;
     }
+#ifdef NVENC_HAVE_422_SUPPORT
+    // force setting profile as high422p if input is AV_PIX_FMT_YUV422P
+    if (IS_YUV422(ctx->data_pix_fmt)) {
+        cc->profileGUID = NV_ENC_H264_PROFILE_HIGH_422_GUID;
+        avctx->profile = AV_PROFILE_H264_HIGH_422;
+    }
+#endif
     vui->bitstreamRestrictionFlag = cc->gopLength != 1 || avctx->profile < AV_PROFILE_H264_HIGH;
-    h264->chromaFormatIDC = avctx->profile == AV_PROFILE_H264_HIGH_444_PREDICTIVE ? 3 : 1;
+    h264->chromaFormatIDC = IS_YUV444(ctx->data_pix_fmt) ? 3 : IS_YUV422(ctx->data_pix_fmt) ? 2 : 1;
     h264->level = ctx->level;
 #ifdef NVENC_HAVE_NEW_BIT_DEPTH_API
-    h264->inputBitDepth = h264->outputBitDepth =
+    h264->inputBitDepth = IS_10BIT(ctx->data_pix_fmt) ? NV_ENC_BIT_DEPTH_10 : NV_ENC_BIT_DEPTH_8;
-        IS_10BIT(ctx->data_pix_fmt) ? NV_ENC_BIT_DEPTH_10 : NV_ENC_BIT_DEPTH_8;
+    h264->outputBitDepth = (IS_10BIT(ctx->data_pix_fmt) || ctx->highbitdepth) ? NV_ENC_BIT_DEPTH_10 : NV_ENC_BIT_DEPTH_8;
 #endif
     if (ctx->coder >= 0)
         avctx->profile = AV_PROFILE_HEVC_MAIN_10;
     }
-    // force setting profile as rext if input is yuv444
+    // force setting profile as rext if input is yuv444 or yuv422
-    if (IS_YUV444(ctx->data_pix_fmt)) {
+    if (IS_YUV444(ctx->data_pix_fmt) || IS_YUV422(ctx->data_pix_fmt)) {
         cc->profileGUID = NV_ENC_HEVC_PROFILE_FREXT_GUID;
         avctx->profile = AV_PROFILE_HEVC_REXT;
     }
-    hevc->chromaFormatIDC = IS_YUV444(ctx->data_pix_fmt) ? 3 : 1;
+    hevc->chromaFormatIDC = IS_YUV444(ctx->data_pix_fmt) ? 3 : IS_YUV422(ctx->data_pix_fmt) ? 2 : 1;
 #ifdef NVENC_HAVE_NEW_BIT_DEPTH_API
     hevc->inputBitDepth = IS_10BIT(ctx->data_pix_fmt) ? NV_ENC_BIT_DEPTH_10 : NV_ENC_BIT_DEPTH_8;
         return NV_ENC_BUFFER_FORMAT_ARGB10;
     case AV_PIX_FMT_X2BGR10:
         return NV_ENC_BUFFER_FORMAT_ABGR10;
+#ifdef NVENC_HAVE_422_SUPPORT
+    case AV_PIX_FMT_NV16:
+        return NV_ENC_BUFFER_FORMAT_NV16;
+    case AV_PIX_FMT_P210:
+    case AV_PIX_FMT_P216:
+        return NV_ENC_BUFFER_FORMAT_P210;
+#endif
     default:
         return NV_ENC_BUFFER_FORMAT_UNDEFINED;
     }

libavcodec/nvenc.h+12−0

 #define NVENC_HAVE_UNIDIR_B
 #endif
+// SDK 13.0 compile time feature checks
+#if NVENCAPI_CHECK_VERSION(13, 0)
+#define NVENC_HAVE_H264_10BIT_SUPPORT
+#define NVENC_HAVE_422_SUPPORT
+#endif
 typedef struct NvencSurface
 {
     NV_ENC_INPUT_PTR input_surface;
     NV_ENC_H264_PROFILE_BASELINE,
     NV_ENC_H264_PROFILE_MAIN,
     NV_ENC_H264_PROFILE_HIGH,
+#ifdef NVENC_HAVE_H264_10BIT_SUPPORT
+    NV_ENC_H264_PROFILE_HIGH_10,
+#endif
+#ifdef NVENC_HAVE_422_SUPPORT
+    NV_ENC_H264_PROFILE_HIGH_422,
+#endif
     NV_ENC_H264_PROFILE_HIGH_444P,
 };

libavcodec/nvenc_h264.c+9−0

     { "baseline",     "",                                   0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_H264_PROFILE_BASELINE },  0, 0, VE, .unit = "profile" },
     { "main",         "",                                   0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_H264_PROFILE_MAIN },      0, 0, VE, .unit = "profile" },
     { "high",         "",                                   0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_H264_PROFILE_HIGH },      0, 0, VE, .unit = "profile" },
+#ifdef NVENC_HAVE_H264_10BIT_SUPPORT
+    { "high10",       "",                                   0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_H264_PROFILE_HIGH_10 },   0, 0, VE, .unit = "profile" },
+#endif
+#ifdef NVENC_HAVE_422_SUPPORT
+    { "high422",      "",                                   0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_H264_PROFILE_HIGH_422 },  0, 0, VE, .unit = "profile" },
+#endif
     { "high444p",     "",                                   0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_H264_PROFILE_HIGH_444P }, 0, 0, VE, .unit = "profile" },
 #ifdef NVENC_HAVE_H264_LVL6
     { "level",        "Set the encoding level restriction", OFFSET(level),        AV_OPT_TYPE_INT,   { .i64 = NV_ENC_LEVEL_AUTOSELECT }, NV_ENC_LEVEL_AUTOSELECT, NV_ENC_LEVEL_H264_62, VE, .unit = "level" },
     { "fullres",      "Two Pass encoding is enabled where first Pass is full resolution",
                                                             0,                    AV_OPT_TYPE_CONST, { .i64 = NV_ENC_TWO_PASS_FULL_RESOLUTION },    0,                          0,                               VE, .unit = "multipass" },
 #endif
+#ifdef NVENC_HAVE_H264_10BIT_SUPPORT
+    { "highbitdepth", "Enable 10 bit encode for 8 bit input",OFFSET(highbitdepth),AV_OPT_TYPE_BOOL,  { .i64 = 0 }, 0, 1, VE },
+#endif
 #ifdef NVENC_HAVE_LDKFS
     { "ldkfs",        "Low delay key frame scale; Specifies the Scene Change frame size increase allowed in case of single frame VBV and CBR",
                                                             OFFSET(ldkfs),        AV_OPT_TYPE_INT,   { .i64 = 0 }, 0, UCHAR_MAX, VE },

avcodec/nvenc: add 4:2:2 encoding and H.264 10-bit support

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code