arm: Implement chroma intra deblock

[x264] / common / deblock.c

diff --git a/common/deblock.c b/common/deblock.c
index dcf3b9880b5b8d61c697020df05e1cf845b67d3c..46379eccd8d1408446d17c2b8a38628a4c52d621 100644 (file)
--- a/common/deblock.c
+++ b/common/deblock.c
@@ -739,18 +739,34 @@ void x264_deblock_h_chroma_neon( uint8_t *pix, intptr_t stride, int alpha, int b
 void x264_deblock_strength_neon( uint8_t nnz[X264_SCAN8_SIZE], int8_t ref[2][X264_SCAN8_LUMA_SIZE],
                                  int16_t mv[2][X264_SCAN8_LUMA_SIZE][2], uint8_t bs[2][8][4],
                                  int mvy_limit, int bframe );
-#if ARCH_AARCH64
 void x264_deblock_h_chroma_422_neon( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
 void x264_deblock_h_chroma_mbaff_neon( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
 void x264_deblock_h_chroma_intra_mbaff_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 void x264_deblock_h_chroma_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 void x264_deblock_h_chroma_422_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 void x264_deblock_v_chroma_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
+#if ARCH_AARCH64
 void x264_deblock_h_luma_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 void x264_deblock_v_luma_intra_neon( uint8_t *pix, intptr_t stride, int alpha, int beta );
 #endif
 #endif
 
+#if !HIGH_BIT_DEPTH
+#if HAVE_MSA
+void x264_deblock_v_luma_msa( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
+void x264_deblock_h_luma_msa( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
+void x264_deblock_v_chroma_msa( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
+void x264_deblock_h_chroma_msa( uint8_t *pix, intptr_t stride, int alpha, int beta, int8_t *tc0 );
+void x264_deblock_v_luma_intra_msa( uint8_t *pix, intptr_t stride, int alpha, int beta );
+void x264_deblock_h_luma_intra_msa( uint8_t *pix, intptr_t stride, int alpha, int beta );
+void x264_deblock_v_chroma_intra_msa( uint8_t *pix, intptr_t stride, int alpha, int beta );
+void x264_deblock_h_chroma_intra_msa( uint8_t *pix, intptr_t stride, int alpha, int beta );
+void x264_deblock_strength_msa( uint8_t nnz[X264_SCAN8_SIZE], int8_t ref[2][X264_SCAN8_LUMA_SIZE],
+                                int16_t mv[2][X264_SCAN8_LUMA_SIZE][2], uint8_t bs[2][8][4], int mvy_limit,
+                                int bframe );
+#endif
+#endif
+
 void x264_deblock_init( int cpu, x264_deblock_function_t *pf, int b_mbaff )
 {
     pf->deblock_luma[1] = deblock_v_luma_c;
@@ -857,19 +873,34 @@ void x264_deblock_init( int cpu, x264_deblock_function_t *pf, int b_mbaff )
         pf->deblock_luma[0] = x264_deblock_h_luma_neon;
         pf->deblock_chroma[1] = x264_deblock_v_chroma_neon;
         pf->deblock_h_chroma_420 = x264_deblock_h_chroma_neon;
-#if ARCH_AARCH64
+        pf->deblock_h_chroma_422 = x264_deblock_h_chroma_422_neon;
         pf->deblock_chroma_420_mbaff = x264_deblock_h_chroma_mbaff_neon;
         pf->deblock_chroma_420_intra_mbaff = x264_deblock_h_chroma_intra_mbaff_neon;
         pf->deblock_h_chroma_420_intra = x264_deblock_h_chroma_intra_neon;
-        pf->deblock_h_chroma_422 = x264_deblock_h_chroma_422_neon;
         pf->deblock_h_chroma_422_intra = x264_deblock_h_chroma_422_intra_neon;
         pf->deblock_chroma_intra[1] = x264_deblock_v_chroma_intra_neon;
+#if ARCH_AARCH64
         pf->deblock_luma_intra[0] = x264_deblock_h_luma_intra_neon;
         pf->deblock_luma_intra[1] = x264_deblock_v_luma_intra_neon;
 #endif
         pf->deblock_strength     = x264_deblock_strength_neon;
     }
 #endif
+
+#if HAVE_MSA
+    if( cpu&X264_CPU_MSA )
+    {
+        pf->deblock_luma[1] = x264_deblock_v_luma_msa;
+        pf->deblock_luma[0] = x264_deblock_h_luma_msa;
+        pf->deblock_chroma[1] = x264_deblock_v_chroma_msa;
+        pf->deblock_h_chroma_420 = x264_deblock_h_chroma_msa;
+        pf->deblock_luma_intra[1] = x264_deblock_v_luma_intra_msa;
+        pf->deblock_luma_intra[0] = x264_deblock_h_luma_intra_msa;
+        pf->deblock_chroma_intra[1] = x264_deblock_v_chroma_intra_msa;
+        pf->deblock_h_chroma_420_intra = x264_deblock_h_chroma_intra_msa;
+        pf->deblock_strength = x264_deblock_strength_msa;
+    }
+#endif
 #endif // !HIGH_BIT_DEPTH
 
     /* These functions are equivalent, so don't duplicate them. */