Support NV12->YV12 output in DXVA2-CB Direct Mode

1 files changed, 93 insertions, 0 deletions

diff --git a/decoder/LAVVideo/pixconv/convert_direct.cpp b/decoder/LAVVideo/pixconv/convert_direct.cpp
index b2857317..a8017e81 100644
--- a/decoder/LAVVideo/pixconv/convert_direct.cpp
+++ b/decoder/LAVVideo/pixconv/convert_direct.cpp
@@ -89,6 +89,99 @@ DECLARE_CONV_FUNC_IMPL(plane_copy_direct_sse4)
   return S_OK;
 }
 
+DECLARE_CONV_FUNC_IMPL(convert_nv12_yv12_direct_sse4)
+{
+  const ptrdiff_t inStride = srcStride[0];
+  const ptrdiff_t outStride = dstStride[0];
+  const ptrdiff_t outChromaStride = dstStride[1];
+  const ptrdiff_t chromaHeight = (height >> 1);
+
+  const ptrdiff_t stride = min(FFALIGN(width, 64), min(inStride, outStride));
+
+  __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm7;
+  xmm7 = _mm_set1_epi16(0x00FF);
+
+  _mm_sfence();
+
+  ptrdiff_t line, i;
+
+  for (line = 0; line < height; line++) {
+    const uint8_t *y = (src[0] + line * inStride);
+    uint8_t *dy = (dst[0] + line * outStride);
+    for (i = 0; i < (stride - 63); i += 64) {
+      PIXCONV_STREAM_LOAD(xmm0, y + i + 0);
+      PIXCONV_STREAM_LOAD(xmm1, y + i + 16);
+      PIXCONV_STREAM_LOAD(xmm2, y + i + 32);
+      PIXCONV_STREAM_LOAD(xmm3, y + i + 48);
+
+      _ReadWriteBarrier();
+
+      PIXCONV_PUT_STREAM(dy + i + 0, xmm0);
+      PIXCONV_PUT_STREAM(dy + i + 16, xmm1);
+      PIXCONV_PUT_STREAM(dy + i + 32, xmm2);
+      PIXCONV_PUT_STREAM(dy + i + 48, xmm3);
+    }
+
+    for (; i < width; i += 16) {
+      PIXCONV_LOAD_ALIGNED(xmm0, y + i);
+      PIXCONV_PUT_STREAM(dy + i, xmm0);
+    }
+  }
+
+  for (line = 0; line < chromaHeight; line++) {
+    const uint8_t *uv = (src[1] + line * inStride);
+    uint8_t *dv = (dst[1] + line * outChromaStride);
+    uint8_t *du = (dst[2] + line * outChromaStride);
+    for (i = 0; i < (stride - 63); i += 64) {
+      PIXCONV_STREAM_LOAD(xmm0, uv + i + 0);
+      PIXCONV_STREAM_LOAD(xmm1, uv + i + 16);
+      PIXCONV_STREAM_LOAD(xmm2, uv + i + 32);
+      PIXCONV_STREAM_LOAD(xmm3, uv + i + 48);
+
+      _ReadWriteBarrier();
+
+      // process first pair
+      xmm4 = _mm_srli_epi16(xmm0, 8);
+      xmm5 = _mm_srli_epi16(xmm1, 8);
+      xmm0 = _mm_and_si128(xmm0, xmm7);
+      xmm1 = _mm_and_si128(xmm1, xmm7);
+      xmm0 = _mm_packus_epi16(xmm0, xmm1);
+      xmm4 = _mm_packus_epi16(xmm4, xmm5);
+
+      PIXCONV_PUT_STREAM(du + (i >> 1) + 0, xmm0);
+      PIXCONV_PUT_STREAM(dv + (i >> 1) + 0, xmm4);
+
+      // and second pair
+      xmm4 = _mm_srli_epi16(xmm2, 8);
+      xmm5 = _mm_srli_epi16(xmm3, 8);
+      xmm2 = _mm_and_si128(xmm2, xmm7);
+      xmm3 = _mm_and_si128(xmm3, xmm7);
+      xmm2 = _mm_packus_epi16(xmm2, xmm3);
+      xmm4 = _mm_packus_epi16(xmm4, xmm5);
+
+      PIXCONV_PUT_STREAM(du + (i >> 1) + 16, xmm2);
+      PIXCONV_PUT_STREAM(dv + (i >> 1) + 16, xmm4);
+    }
+
+    for (; i < width; i += 32) {
+      PIXCONV_LOAD_ALIGNED(xmm0, uv + i);
+      PIXCONV_LOAD_ALIGNED(xmm1, uv + i);
+
+      xmm4 = _mm_srli_epi16(xmm0, 8);
+      xmm5 = _mm_srli_epi16(xmm1, 8);
+      xmm0 = _mm_and_si128(xmm0, xmm7);
+      xmm1 = _mm_and_si128(xmm1, xmm7);
+      xmm0 = _mm_packus_epi16(xmm0, xmm1);
+      xmm4 = _mm_packus_epi16(xmm4, xmm5);
+
+      PIXCONV_PUT_STREAM(du + (i >> 1), xmm0);
+      PIXCONV_PUT_STREAM(dv + (i >> 1), xmm4);
+    }
+  }
+
+  return S_OK;
+}
+
 DECLARE_CONV_FUNC_IMPL(convert_p010_nv12_direct_sse4)
 {
   const ptrdiff_t inStride = srcStride[0];