libavcodec/blockdsp : add AVX version

Also modify the required alignment, to 32 instead of 16
for several codecs

Signed-off-by: James Almer <jamrial@gmail.com>

16 files changed, 32 insertions, 24 deletions

diff --git a/libavcodec/asv.h b/libavcodec/asv.h
index e2cdc81300..a1366b6fe4 100644
--- a/libavcodec/asv.h
+++ b/libavcodec/asv.h
@@ -54,7 +54,7 @@ typedef struct ASV1Context {
     int mb_height;
     int mb_width2;
     int mb_height2;
-    DECLARE_ALIGNED(16, int16_t, block)[6][64];
+    DECLARE_ALIGNED(32, int16_t, block)[6][64];
     uint16_t intra_matrix[64];
     int q_intra_matrix[64];
     uint8_t *bitstream_buffer;
diff --git a/libavcodec/bink.c b/libavcodec/bink.c
index cc55870114..346b6cda9d 100644
--- a/libavcodec/bink.c
+++ b/libavcodec/bink.c
@@ -813,7 +813,7 @@ static int binkb_decode_plane(BinkContext *c, AVFrame *frame, GetBitContext *gb,
     int v, col[2];
     const uint8_t *scan;
     int xoff, yoff;
-    LOCAL_ALIGNED_16(int16_t, block, [64]);
+    LOCAL_ALIGNED_32(int16_t, block, [64]);
     LOCAL_ALIGNED_16(int32_t, dctblock, [64]);
     int coordmap[64];
     int ybias = is_key ? -15 : 0;
@@ -976,7 +976,7 @@ static int bink_decode_plane(BinkContext *c, AVFrame *frame, GetBitContext *gb,
     uint8_t *dst, *prev, *ref_start, *ref_end;
     int v, col[2];
     const uint8_t *scan;
-    LOCAL_ALIGNED_16(int16_t, block, [64]);
+    LOCAL_ALIGNED_32(int16_t, block, [64]);
     LOCAL_ALIGNED_16(uint8_t, ublock, [64]);
     LOCAL_ALIGNED_16(int32_t, dctblock, [64]);
     int coordmap[64];
diff --git a/libavcodec/dnxhdenc.h b/libavcodec/dnxhdenc.h
index 26c3eec695..963821ac81 100644
--- a/libavcodec/dnxhdenc.h
+++ b/libavcodec/dnxhdenc.h
@@ -74,7 +74,7 @@ typedef struct DNXHDEncContext {
     unsigned min_padding;
     int intra_quant_bias;
 
-    DECLARE_ALIGNED(16, int16_t, blocks)[12][64];
+    DECLARE_ALIGNED(32, int16_t, blocks)[12][64];
     DECLARE_ALIGNED(16, uint8_t, edge_buf_y)[512]; // has to hold 16x16 uint16 when depth=10
     DECLARE_ALIGNED(16, uint8_t, edge_buf_uv)[2][512]; // has to hold 16x16 uint16_t when depth=10
 
diff --git a/libavcodec/eamad.c b/libavcodec/eamad.c
index 753dee06c3..7f28abbafe 100644
--- a/libavcodec/eamad.c
+++ b/libavcodec/eamad.c
@@ -54,7 +54,7 @@ typedef struct MadContext {
     GetBitContext gb;
     void *bitstream_buf;
     unsigned int bitstream_buf_size;
-    DECLARE_ALIGNED(16, int16_t, block)[64];
+    DECLARE_ALIGNED(32, int16_t, block)[64];
     ScanTable scantable;
     uint16_t quant_matrix[64];
     int mb_x;
diff --git a/libavcodec/eatqi.c b/libavcodec/eatqi.c
index 725289448a..1a847a35da 100644
--- a/libavcodec/eatqi.c
+++ b/libavcodec/eatqi.c
@@ -51,7 +51,7 @@ typedef struct TqiContext {
     uint16_t intra_matrix[64];
     int last_dc[3];
 
-    DECLARE_ALIGNED(16, int16_t, block)[6][64];
+    DECLARE_ALIGNED(32, int16_t, block)[6][64];
 } TqiContext;
 
 static av_cold int tqi_decode_init(AVCodecContext *avctx)
diff --git a/libavcodec/g2meet.c b/libavcodec/g2meet.c
index 10b6808f81..842095ba3b 100644
--- a/libavcodec/g2meet.c
+++ b/libavcodec/g2meet.c
@@ -122,7 +122,7 @@ typedef struct JPGContext {
 
     VLC        dc_vlc[2], ac_vlc[2];
     int        prev_dc[3];
-    DECLARE_ALIGNED(16, int16_t, block)[6][64];
+    DECLARE_ALIGNED(32, int16_t, block)[6][64];
 
     uint8_t    *buf;
 } JPGContext;
diff --git a/libavcodec/ituh263dec.c b/libavcodec/ituh263dec.c
index edb68861ac..fc95a532ce 100644
--- a/libavcodec/ituh263dec.c
+++ b/libavcodec/ituh263dec.c
@@ -574,7 +574,7 @@ not_coded:
 
 static int h263_skip_b_part(MpegEncContext *s, int cbp)
 {
-    LOCAL_ALIGNED_16(int16_t, dblock, [64]);
+    LOCAL_ALIGNED_32(int16_t, dblock, [64]);
     int i, mbi;
     int bli[6];
 
diff --git a/libavcodec/mdec.c b/libavcodec/mdec.c
index 59658b331d..330b761279 100644
--- a/libavcodec/mdec.c
+++ b/libavcodec/mdec.c
@@ -48,7 +48,7 @@ typedef struct MDECContext {
     int mb_width;
     int mb_height;
     int mb_x, mb_y;
-    DECLARE_ALIGNED(16, int16_t, block)[6][64];
+    DECLARE_ALIGNED(32, int16_t, block)[6][64];
     DECLARE_ALIGNED(16, uint16_t, quant_matrix)[64];
     uint8_t *bitstream_buffer;
     unsigned int bitstream_buffer_size;
diff --git a/libavcodec/mimic.c b/libavcodec/mimic.c
index 02d8b30e31..1d463e9962 100644
--- a/libavcodec/mimic.c
+++ b/libavcodec/mimic.c
@@ -49,7 +49,7 @@ typedef struct MimicContext {
 
     ThreadFrame     frames     [16];
 
-    DECLARE_ALIGNED(16, int16_t, dct_block)[64];
+    DECLARE_ALIGNED(32, int16_t, dct_block)[64];
 
     GetBitContext   gb;
     ScanTable       scantable;
diff --git a/libavcodec/mjpegdec.h b/libavcodec/mjpegdec.h
index 2bc69fa930..c84a40aa6e 100644
--- a/libavcodec/mjpegdec.h
+++ b/libavcodec/mjpegdec.h
@@ -98,7 +98,7 @@ typedef struct MJpegDecodeContext {
     int got_picture;                                ///< we found a SOF and picture is valid, too.
     int linesize[MAX_COMPONENTS];                   ///< linesize << interlaced
     int8_t *qscale_table;
-    DECLARE_ALIGNED(16, int16_t, block)[64];
+    DECLARE_ALIGNED(32, int16_t, block)[64];
     int16_t (*blocks[MAX_COMPONENTS])[64]; ///< intermediate sums (progressive mode)
     uint8_t *last_nnz[MAX_COMPONENTS];
     uint64_t coefs_finished[MAX_COMPONENTS]; ///< bitmask of which coefs have been completely decoded (progressive mode)
diff --git a/libavcodec/proresdec2.c b/libavcodec/proresdec2.c
index a7cea2b3b8..0f791de97b 100644
--- a/libavcodec/proresdec2.c
+++ b/libavcodec/proresdec2.c
@@ -368,7 +368,7 @@ static int decode_slice_luma(AVCodecContext *avctx, SliceContext *slice,
                              const int16_t *qmat)
 {
     ProresContext *ctx = avctx->priv_data;
-    LOCAL_ALIGNED_16(int16_t, blocks, [8*4*64]);
+    LOCAL_ALIGNED_32(int16_t, blocks, [8*4*64]);
     int16_t *block;
     GetBitContext gb;
     int i, blocks_per_slice = slice->mb_count<<2;
@@ -402,7 +402,7 @@ static int decode_slice_chroma(AVCodecContext *avctx, SliceContext *slice,
                                const int16_t *qmat, int log2_blocks_per_mb)
 {
     ProresContext *ctx = avctx->priv_data;
-    LOCAL_ALIGNED_16(int16_t, blocks, [8*4*64]);
+    LOCAL_ALIGNED_32(int16_t, blocks, [8*4*64]);
     int16_t *block;
     GetBitContext gb;
     int i, j, blocks_per_slice = slice->mb_count << log2_blocks_per_mb;
@@ -485,7 +485,7 @@ static void decode_slice_alpha(ProresContext *ctx,
 {
     GetBitContext gb;
     int i;
-    LOCAL_ALIGNED_16(int16_t, blocks, [8*4*64]);
+    LOCAL_ALIGNED_32(int16_t, blocks, [8*4*64]);
     int16_t *block;
 
     for (i = 0; i < blocks_per_slice<<2; i++)
diff --git a/libavcodec/speedhq.c b/libavcodec/speedhq.c
index 47b1e4dc7a..6d3487ca19 100644
--- a/libavcodec/speedhq.c
+++ b/libavcodec/speedhq.c
@@ -224,7 +224,7 @@ static inline int decode_dct_block(const SHQContext *s, GetBitContext *gb, int l
 {
     const int *quant_matrix = s->quant_matrix;
     const uint8_t *scantable = s->intra_scantable.permutated;
-    LOCAL_ALIGNED_16(int16_t, block, [64]);
+    LOCAL_ALIGNED_32(int16_t, block, [64]);
     int dc_offset;
 
     s->bdsp.clear_block(block);
diff --git a/libavcodec/wmv2.h b/libavcodec/wmv2.h
index 31593b8c38..0f459ae5ae 100644
--- a/libavcodec/wmv2.h
+++ b/libavcodec/wmv2.h
@@ -51,7 +51,7 @@ typedef struct Wmv2Context {
     int hshift;
 
     ScanTable abt_scantable[2];
-    DECLARE_ALIGNED(16, int16_t, abt_block2)[6][64];
+    DECLARE_ALIGNED(32, int16_t, abt_block2)[6][64];
 } Wmv2Context;
 
 void ff_wmv2_common_init(Wmv2Context *w);
diff --git a/libavcodec/x86/blockdsp.asm b/libavcodec/x86/blockdsp.asm
index 7cbfa3a843..2498bd40b3 100644
--- a/libavcodec/x86/blockdsp.asm
+++ b/libavcodec/x86/blockdsp.asm
@@ -4,6 +4,8 @@
 ;* Copyright (c) 2008 Loren Merritt
 ;* Copyright (c) 2009 Fiona Glaser
 ;*
+;* AVX version by Jokyo Images
+;*
 ;* This file is part of FFmpeg.
 ;*
 ;* FFmpeg is free software; you can redistribute it and/or
@@ -39,20 +41,18 @@ cglobal clear_block, 1, 1, %1, blocks
     mova  [blocksq+mmsize*(1+%%i)], m0
     mova  [blocksq+mmsize*(2+%%i)], m0
     mova  [blocksq+mmsize*(3+%%i)], m0
-    mova  [blocksq+mmsize*(4+%%i)], m0
-    mova  [blocksq+mmsize*(5+%%i)], m0
-    mova  [blocksq+mmsize*(6+%%i)], m0
-    mova  [blocksq+mmsize*(7+%%i)], m0
-%assign %%i %%i+8
+%assign %%i %%i+4
 %endrep
     RET
 %endmacro
 
 INIT_MMX mmx
 %define ZERO pxor
-CLEAR_BLOCK 0, 2
+CLEAR_BLOCK 0, 4
 INIT_XMM sse
 %define ZERO xorps
+CLEAR_BLOCK 1, 2
+INIT_YMM avx
 CLEAR_BLOCK 1, 1
 
 ;-----------------------------------------
@@ -84,3 +84,5 @@ CLEAR_BLOCKS 0
 INIT_XMM sse
 %define ZERO xorps
 CLEAR_BLOCKS 1
+INIT_YMM avx
+CLEAR_BLOCKS 1
diff --git a/libavcodec/x86/blockdsp_init.c b/libavcodec/x86/blockdsp_init.c
index afd25e1cbb..8b01a447cd 100644
--- a/libavcodec/x86/blockdsp_init.c
+++ b/libavcodec/x86/blockdsp_init.c
@@ -28,8 +28,10 @@
 
 void ff_clear_block_mmx(int16_t *block);
 void ff_clear_block_sse(int16_t *block);
+void ff_clear_block_avx(int16_t *block);
 void ff_clear_blocks_mmx(int16_t *blocks);
 void ff_clear_blocks_sse(int16_t *blocks);
+void ff_clear_blocks_avx(int16_t *blocks);
 
 av_cold void ff_blockdsp_init_x86(BlockDSPContext *c,
                                   AVCodecContext *avctx)
@@ -50,5 +52,9 @@ av_cold void ff_blockdsp_init_x86(BlockDSPContext *c,
         c->clear_block  = ff_clear_block_sse;
         c->clear_blocks = ff_clear_blocks_sse;
     }
+    if (EXTERNAL_AVX_FAST(cpu_flags)) {
+        c->clear_block  = ff_clear_block_avx;
+        c->clear_blocks = ff_clear_blocks_avx;
+    }
 #endif /* HAVE_X86ASM */
 }
diff --git a/tests/checkasm/blockdsp.c b/tests/checkasm/blockdsp.c
index 153699b632..c753506b3c 100644
--- a/tests/checkasm/blockdsp.c
+++ b/tests/checkasm/blockdsp.c
@@ -53,8 +53,8 @@ do {                                                                \
 
 void checkasm_check_blockdsp(void)
 {
-    LOCAL_ALIGNED_16(uint16_t, buf0, [6 * 8 * 8]);
-    LOCAL_ALIGNED_16(uint16_t, buf1, [6 * 8 * 8]);
+    LOCAL_ALIGNED_32(uint16_t, buf0, [6 * 8 * 8]);
+    LOCAL_ALIGNED_32(uint16_t, buf1, [6 * 8 * 8]);
 
     AVCodecContext avctx = { 0 };
     BlockDSPContext h;