p010→nv12変換を追加し、avswで10bitをhwデコードするとエラー終了する問題を修正。

rigaya · May 12, 2024 · 51e1594 · 51e1594
1 parent b232794
commit 51e1594
Show file tree

Hide file tree

Showing 4 changed files with 107 additions and 0 deletions.
diff --git a/QSVEnc/QSVEnc_readme.txt b/QSVEnc/QSVEnc_readme.txt
@@ -313,6 +313,21 @@ API v1.1  … Intel Media SDK v2.0
 
 
 【どうでもいいメモ】
+2024.05.12 (7.64)
+- ffmpeg 7.0に更新。(Windows版)
+  - ffmpeg     6.1    -> 7.0
+  - libpng     1.4.0  -> 1.4.3
+  - expat      2.5.0  -> 2.6.2
+  - opus       1.4    -> 1.5.2
+  - libxml2    2.12.0 -> 2.12.6
+  - dav1d      1.3.0  -> 1.4.1
+  - libvpl     2.11.0 (new!)
+  - nv-codec-headers 12.2.72.0 (new!)
+- avswで使用するデコーダを指定可能に。
+- --audio-bitrateの指定がないとき、デフォルトのビットレートを設定するのではなく、コーデックに任せるように。
+- --audio-bitrateあるいは--audio-copyで指定のない音声/字幕/データトラックは処理しないように。
+- QSVEnc 7.62以降のside_dataの扱いが誤っており、--master-display  copy/--max-cll copyが正常に行われていなかった問題を修正。
+
 2024.04.28 (7.63)
 - 新たなノイズ除去フィルタを追加。(--vpp-nlmeans)
 - --audio-resamplerを拡張し、文字列でパラメータ設定できるように。

diff --git a/QSVPipeline/convert_csp.cpp b/QSVPipeline/convert_csp.cpp
@@ -41,6 +41,9 @@ void copy_p010_to_p010_sse2(void **dst, const void **src, int width, int src_y_p
 void copy_nv12_to_nv12_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
 void copy_p010_to_p010_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
 
+void copy_p010_to_nv12_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
+void copy_nv12_to_p010_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
+
 void convert_yuy2_to_nv12(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
 void convert_yuy2_to_nv12_sse2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
 void convert_yuy2_to_nv12_avx(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop);
@@ -240,6 +243,36 @@ void copy_p010_to_p010_c(void **dst, const void **src, int width, int src_y_pitc
     return copy_nv12_to_nv12_c_internal<true>(dst, src, width, src_y_pitch_byte, src_uv_pitch_byte, dst_y_pitch_byte, height, dst_height, thread_id, thread_n, crop);
 }
 
+template<typename Tin, int in_bit_depth, typename Tout, int out_bit_depth>
+void copy_nv12p010_to_nv12p010_c_internal(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
+    const int crop_left = crop[0];
+    const int crop_up = crop[1];
+    const int crop_right = crop[2];
+    const int crop_bottom = crop[3];
+    for (int i = 0; i < 2; i++) {
+        const auto y_range = thread_y_range(crop_up >> i, (height - crop_bottom) >> i, thread_id, thread_n);
+        const uint8_t *srcYLine = ((const uint8_t *)src[i] + src_y_pitch_byte * y_range.start_src + crop_left * sizeof(uint16_t));
+        uint8_t *dstLine = (uint8_t *)dst[i] + dst_y_pitch_byte * y_range.start_dst;
+        const int y_width = width - crop_right - crop_left;
+        for (int y = 0; y < y_range.len; y++, srcYLine += src_y_pitch_byte, dstLine += dst_y_pitch_byte) {
+            const int x_fin = width - crop_right - crop_left;
+            const Tin *ptrSrc = (const Tin *)srcYLine;
+            Tout *ptrDst = (Tout *)dstLine;
+            for (int x = 0; x < x_fin; x++) {
+                ptrDst[x] = (Tout)conv_bit_depth<out_bit_depth, in_bit_depth, 0>(ptrSrc[x]);
+            }
+        }
+    }
+}
+
+void copy_p010_to_nv12_c(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
+    return copy_nv12p010_to_nv12p010_c_internal<uint8_t, 8, uint16_t, 16>(dst, src, width, src_y_pitch_byte, src_uv_pitch_byte, dst_y_pitch_byte, height, dst_height, thread_id, thread_n, crop);
+}
+
+void copy_nv12_to_p010_c(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
+    return copy_nv12p010_to_nv12p010_c_internal<uint16_t, 16, uint8_t, 8>(dst, src, width, src_y_pitch_byte, src_uv_pitch_byte, dst_y_pitch_byte, height, dst_height, thread_id, thread_n, crop);
+}
+
 void convert_yuy2_to_nv12(void **dst_array, const void **src_array, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
     int crop_left   = crop[0];
     int crop_up     = crop[1];
@@ -1989,6 +2022,10 @@ static const ConvertCSP funcList[] = {
     FUNC_AVX2( RGY_CSP_P010,      RGY_CSP_P010,      false,  copy_p010_to_p010_avx2,              copy_p010_to_p010_avx2,              AVX2|AVX)
     FUNC_SSE(  RGY_CSP_P010,      RGY_CSP_P010,      false,  copy_p010_to_p010_sse2,              copy_p010_to_p010_sse2,              SSE2 )
     FUNC__C_(  RGY_CSP_P010,      RGY_CSP_P010,      false,  copy_p010_to_p010_c,                 copy_p010_to_p010_c,                 NONE)
+    FUNC_AVX2( RGY_CSP_NV12,      RGY_CSP_P010,      false,  copy_nv12_to_p010_avx2,              copy_nv12_to_p010_avx2,              AVX2|AVX)
+    FUNC__C_(  RGY_CSP_NV12,      RGY_CSP_P010,      false,  copy_nv12_to_p010_c,                 copy_nv12_to_p010_c,                 NONE)
+    FUNC_AVX2( RGY_CSP_P010,      RGY_CSP_NV12,      false,  copy_p010_to_nv12_avx2,              copy_p010_to_nv12_avx2,              AVX2|AVX)
+    FUNC__C_(  RGY_CSP_P010,      RGY_CSP_NV12,      false,  copy_p010_to_nv12_c,                 copy_p010_to_nv12_c,                 NONE)
 #endif
 #if !CLFILTERS_AUF
     FUNC_AVX2( RGY_CSP_YUY2,      RGY_CSP_NV12,      false,  convert_yuy2_to_nv12_avx2,           convert_yuy2_to_nv12_i_avx2,         AVX2|AVX)

diff --git a/QSVPipeline/convert_csp_avx2.cpp b/QSVPipeline/convert_csp_avx2.cpp
@@ -166,6 +166,60 @@ void copy_nv12_to_nv12_avx2(void **dst, const void **src, int width, int src_y_p
 void copy_p010_to_p010_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
     return copy_nv12_to_nv12_avx2_internal<true>(dst, src, width, src_y_pitch_byte, src_uv_pitch_byte, dst_y_pitch_byte, height, dst_height, thread_id, thread_n, crop);
 }
+void copy_nv12_to_p010_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
+    const int crop_left = crop[0];
+    const int crop_up = crop[1];
+    const int crop_right = crop[2];
+    const int crop_bottom = crop[3];
+    for (int i = 0; i < 2; i++) {
+        const auto y_range = thread_y_range(crop_up >> i, (height - crop_bottom) >> i, thread_id, thread_n);
+        const uint8_t *srcYLine = (const uint8_t *)src[i] + src_y_pitch_byte * y_range.start_src + crop_left;
+        uint8_t *dstLine = (uint8_t *)dst[i] + dst_y_pitch_byte * y_range.start_dst;
+        const int y_width = width - crop_right - crop_left;
+        for (int y = 0; y < y_range.len; y++, srcYLine += src_y_pitch_byte, dstLine += dst_y_pitch_byte) {
+            const uint8_t *src_ptr = (const uint8_t *)srcYLine;
+            uint16_t *dst_ptr = (uint16_t *)dstLine;
+            for (int x = 0; x < y_width; x += 32, dst_ptr += 32, src_ptr += 32) {
+                __m256i y0, y1;
+                y0 = _mm256_loadu_si256((const __m256i *)src_ptr);
+                y0 = _mm256_permute4x64_epi64(y0, _MM_SHUFFLE(3, 1, 2, 0));
+                y1 = _mm256_unpackhi_epi8(_mm256_setzero_si256(), y0);
+                y0 = _mm256_unpacklo_epi8(_mm256_setzero_si256(), y0);
+                _mm256_storeu_si256((__m256i *)(dst_ptr + 0), y0);
+                _mm256_storeu_si256((__m256i *)(dst_ptr + 16), y1);
+            }
+        }
+    }
+}
+void copy_p010_to_nv12_avx2(void **dst, const void **src, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
+    const int crop_left = crop[0];
+    const int crop_up = crop[1];
+    const int crop_right = crop[2];
+    const int crop_bottom = crop[3];
+    const int in_bit_depth = 16;
+    const __m256i yrsftAdd = _mm256_set1_epi16((short)conv_bit_depth_rsft_add<in_bit_depth, 8, 0>());
+    for (int i = 0; i < 2; i++) {
+        const auto y_range = thread_y_range(crop_up >> i, (height - crop_bottom) >> i, thread_id, thread_n);
+        const uint8_t *srcYLine = (const uint8_t *)src[i] + src_y_pitch_byte * y_range.start_src + crop_left;
+        uint8_t *dstLine = (uint8_t *)dst[i] + dst_y_pitch_byte * y_range.start_dst;
+        const int y_width = width - crop_right - crop_left;
+        for (int y = 0; y < y_range.len; y++, srcYLine += src_y_pitch_byte, dstLine += dst_y_pitch_byte) {
+            const uint16_t *src_ptr = (const uint16_t *)srcYLine;
+            uint8_t *dst_ptr = dstLine;
+            for (int x = 0; x < y_width; x += 32, dst_ptr += 32, src_ptr += 32) {
+                __m256i y0 = _mm256_loadu2_m128i((const __m128i *)(src_ptr + 16), (const __m128i *)(src_ptr + 0));
+                __m256i y1 = _mm256_loadu2_m128i((const __m128i *)(src_ptr + 24), (const __m128i *)(src_ptr + 8));
+                y0 = _mm256_adds_epi16(y0, yrsftAdd);
+                y1 = _mm256_adds_epi16(y1, yrsftAdd);
+                y0 = _mm256_srli_epi16(y0, in_bit_depth - 8);
+                y1 = _mm256_srli_epi16(y1, in_bit_depth - 8);
+                y0 = _mm256_packus_epi16(y0, y1);
+                _mm256_storeu_si256((__m256i *)dst_ptr, y0);
+            }
+
+        }
+    }
+}
 
 void convert_yuy2_to_nv12_avx2(void **dst_array, const void **src_array, int width, int src_y_pitch_byte, int src_uv_pitch_byte, int dst_y_pitch_byte, int height, int dst_height, int thread_id, int thread_n, int *crop) {
     const int crop_left   = crop[0];

diff --git a/QSVPipeline/rgy_avutil.cpp b/QSVPipeline/rgy_avutil.cpp
@@ -840,6 +840,7 @@ static const auto CSP_PIXFMT_RGY = make_array<std::pair<AVPixelFormat, RGY_CSP>>
     std::make_pair(AV_PIX_FMT_YUVJ420P,    RGY_CSP_YV12),
     std::make_pair(AV_PIX_FMT_NV12,        RGY_CSP_NV12),
     std::make_pair(AV_PIX_FMT_NV21,        RGY_CSP_NV12),
+    std::make_pair(AV_PIX_FMT_P010LE,      RGY_CSP_P010),
     std::make_pair(AV_PIX_FMT_YUV422P,     RGY_CSP_YUV422),
     std::make_pair(AV_PIX_FMT_YUVJ422P,    RGY_CSP_YUV422),
     std::make_pair(AV_PIX_FMT_YUYV422,     RGY_CSP_YUY2),