diracdec: Template DSP functions adding 10-bit versions

libavcodec/dirac_dwt.h

@@ -30,21 +30,21 @@ typedef short IDWTELEM;
 #define MAX_DECOMPOSITIONS 8
 
 typedef struct DWTCompose {
-    IDWTELEM *b[MAX_DWT_SUPPORT];
+    uint8_t *b[MAX_DWT_SUPPORT];
     int y;
 } DWTCompose;
 
 struct DWTContext;
 
 // Possible prototypes for vertical_compose functions
-typedef void (*vertical_compose_2tap)(IDWTELEM *b0, IDWTELEM *b1, int width);
-typedef void (*vertical_compose_3tap)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width);
-typedef void (*vertical_compose_5tap)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, int width);
-typedef void (*vertical_compose_9tap)(IDWTELEM *dst, IDWTELEM *b[8], int width);
+typedef void (*vertical_compose_2tap)(uint8_t *b0, uint8_t *b1, int width);
+typedef void (*vertical_compose_3tap)(uint8_t *b0, uint8_t *b1, uint8_t *b2, int width);
+typedef void (*vertical_compose_5tap)(uint8_t *b0, uint8_t *b1, uint8_t *b2, uint8_t *b3, uint8_t *b4, int width);
+typedef void (*vertical_compose_9tap)(uint8_t *dst, uint8_t *b[8], int width);
 
 typedef struct DWTContext {
-    IDWTELEM *buffer;
-    IDWTELEM *temp;
+    uint8_t *buffer;
+    uint8_t *temp;
     int width;
     int height;
     int stride;
@@ -57,7 +57,7 @@ typedef struct DWTContext {
     void (*vertical_compose_l1)(void);
     void (*vertical_compose_h1)(void);
     void (*vertical_compose)(void);     ///< one set of lowpass and highpass combined
-    void (*horizontal_compose)(IDWTELEM *b, IDWTELEM *tmp, int width);
+    void (*horizontal_compose)(uint8_t *b, uint8_t *tmp, int width);
 
     DWTCompose cs[MAX_DECOMPOSITIONS];
 } DWTContext;
@@ -76,9 +76,9 @@ enum dwt_type {
 };
 
 // -1 if an error occurred, e.g. the dwt_type isn't recognized
-int ff_spatial_idwt_init2(DWTContext *d, IDWTELEM *buffer, int width, int height,
+int ff_spatial_idwt_init2(DWTContext *d, uint8_t *buffer, int width, int height,
                           int stride, enum dwt_type type, int decomposition_count,
-                          IDWTELEM *temp);
+                          uint8_t *temp, int bit_depth);
 
 void ff_spatial_idwt_slice2(DWTContext *d, int y);
 

libavcodec/diracdec.c

@@ -1687,16 +1687,16 @@ static int dirac_decode_frame_internal(DiracContext *s)
             memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height);
             decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */
         }
-        ret = ff_spatial_idwt_init2(&d, (int16_t*)p->idwt_buf, p->idwt_width, p->idwt_height, p->idwt_stride >> 1,
-                                    s->wavelet_idx+2, s->wavelet_depth, (int16_t*)p->idwt_tmp);
+        ret = ff_spatial_idwt_init2(&d, p->idwt_buf, p->idwt_width, p->idwt_height, p->idwt_stride,
+                                    s->wavelet_idx+2, s->wavelet_depth, p->idwt_tmp, s->bit_depth);
         if (ret < 0)
             return ret;
 
         if (!s->num_refs) { /* intra */
             for (y = 0; y < p->height; y += 16) {
                 ff_spatial_idwt_slice2(&d, y+16); /* decode */
-                s->diracdsp.put_signed_rect_clamped(frame + y*p->stride, p->stride,
-                                                   (int16_t*)(p->idwt_buf) + y*(p->idwt_stride >> 1), (p->idwt_stride >> 1), p->width, 16);
+                s->diracdsp.put_signed_rect_clamped[s->pshift](frame + y*p->stride, p->stride,
+                                                               p->idwt_buf + y*p->idwt_stride, p->idwt_stride, p->width, 16);
             }
         } else { /* inter */
             int rowheight = p->ybsep*p->stride;

libavcodec/diracdsp.c

@@ -135,9 +135,10 @@ ADD_OBMC(8)
 ADD_OBMC(16)
 ADD_OBMC(32)
 
-static void put_signed_rect_clamped_c(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height)
+static void put_signed_rect_clamped_8bit_c(uint8_t *dst, int dst_stride, const uint8_t *_src, int src_stride, int width, int height)
 {
     int x, y;
+    int16_t *src = (int16_t *)_src;
     for (y = 0; y < height; y++) {
         for (x = 0; x < width; x+=4) {
             dst[x  ] = av_clip_uint8(src[x  ] + 128);
@@ -146,7 +147,24 @@ static void put_signed_rect_clamped_c(uint8_t *dst, int dst_stride, const int16_
             dst[x+3] = av_clip_uint8(src[x+3] + 128);
         }
         dst += dst_stride;
-        src += src_stride;
+        src += src_stride >> 1;
+    }
+}
+
+static void put_signed_rect_clamped_10bit_c(uint8_t *_dst, int dst_stride, const uint8_t *_src, int src_stride, int width, int height)
+{
+    int x, y;
+    uint16_t *dst = (uint16_t *)_dst;
+    int32_t *src = (int32_t *)_src;
+    for (y = 0; y < height; y++) {
+        for (x = 0; x < width; x+=4) {
+            dst[x  ] = av_clip(src[x  ] + 512, 0, (1 << 10) - 1);
+            dst[x+1] = av_clip(src[x+1] + 512, 0, (1 << 10) - 1);
+            dst[x+2] = av_clip(src[x+2] + 512, 0, (1 << 10) - 1);
+            dst[x+3] = av_clip(src[x+3] + 512, 0, (1 << 10) - 1);
+        }
+        dst += dst_stride >> 1;
+        src += src_stride >> 2;
     }
 }
 
@@ -177,7 +195,8 @@ av_cold void ff_diracdsp_init(DiracDSPContext *c)
 {
     c->dirac_hpel_filter = dirac_hpel_filter;
     c->add_rect_clamped = add_rect_clamped_c;
-    c->put_signed_rect_clamped = put_signed_rect_clamped_c;
+    c->put_signed_rect_clamped[0] = put_signed_rect_clamped_8bit_c;
+    c->put_signed_rect_clamped[1] = put_signed_rect_clamped_10bit_c;
 
     c->add_dirac_obmc[0] = add_obmc8_c;
     c->add_dirac_obmc[1] = add_obmc16_c;

libavcodec/diracdsp.h

@@ -41,8 +41,8 @@ typedef struct {
     void (*put_dirac_pixels_tab[3][4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
     void (*avg_dirac_pixels_tab[3][4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);
 
-    void (*put_signed_rect_clamped)(uint8_t *dst/*align 16*/, int dst_stride, const int16_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
-    void (*put_rect_clamped)(uint8_t *dst/*align 16*/, int dst_stride, const int16_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
+    void (*put_signed_rect_clamped[3])(uint8_t *dst/*align 16*/, int dst_stride, const uint8_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
+    void (*put_rect_clamped)(uint8_t *dst/*align 16*/, int dst_stride, const uint8_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
     void (*add_rect_clamped)(uint8_t *dst/*align 16*/, const uint16_t *src/*align 16*/, int stride, const int16_t *idwt/*align 16*/, int idwt_stride, int width, int height/*mod 2*/);
     void (*add_dirac_obmc[3])(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);
 

libavcodec/x86/dirac_dwt.c

@@ -25,17 +25,20 @@
 #include "dirac_dwt.h"
 
 #define COMPOSE_VERTICAL(ext, align) \
-void ff_vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width); \
-void ff_vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width); \
-void ff_vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, int width); \
-void ff_vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, int width); \
-void ff_vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width); \
-void ff_horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w);\
-void ff_horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w);\
-\
-static void vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width) \
+void ff_vertical_compose53iL0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int width); \
+void ff_vertical_compose_dirac53iH0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int width); \
+void ff_vertical_compose_dd137iL0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int16_t *b3, int16_t *b4, int width); \
+void ff_vertical_compose_dd97iH0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int16_t *b3, int16_t *b4, int width); \
+void ff_vertical_compose_haar##ext(int16_t *b0, int16_t *b1, int width); \
+void ff_horizontal_compose_haar0i##ext(int16_t *b, int16_t *tmp, int w);\
+void ff_horizontal_compose_haar1i##ext(int16_t *b, int16_t *tmp, int w);\
+\
+static void vertical_compose53iL0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, int width) \
 { \
     int i, width_align = width&~(align-1); \
+    int16_t *b0 = (int16_t *)_b0; \
+    int16_t *b1 = (int16_t *)_b1; \
+    int16_t *b2 = (int16_t *)_b2; \
 \
     for(i=width_align; i<width; i++) \
         b1[i] = COMPOSE_53iL0(b0[i], b1[i], b2[i]); \
@@ -43,9 +46,12 @@ static void vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
     ff_vertical_compose53iL0##ext(b0, b1, b2, width_align); \
 } \
 \
-static void vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width) \
+static void vertical_compose_dirac53iH0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, int width) \
 { \
     int i, width_align = width&~(align-1); \
+    int16_t *b0 = (int16_t *)_b0; \
+    int16_t *b1 = (int16_t *)_b1; \
+    int16_t *b2 = (int16_t *)_b2; \
 \
     for(i=width_align; i<width; i++) \
         b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]); \
@@ -53,10 +59,15 @@ static void vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELE
     ff_vertical_compose_dirac53iH0##ext(b0, b1, b2, width_align); \
 } \
 \
-static void vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, \
-                                           IDWTELEM *b3, IDWTELEM *b4, int width) \
+static void vertical_compose_dd137iL0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, \
+                                           uint8_t *_b3, uint8_t *_b4, int width) \
 { \
     int i, width_align = width&~(align-1); \
+    int16_t *b0 = (int16_t *)_b0; \
+    int16_t *b1 = (int16_t *)_b1; \
+    int16_t *b2 = (int16_t *)_b2; \
+    int16_t *b3 = (int16_t *)_b3; \
+    int16_t *b4 = (int16_t *)_b4; \
 \
     for(i=width_align; i<width; i++) \
         b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
@@ -64,19 +75,26 @@ static void vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
     ff_vertical_compose_dd137iL0##ext(b0, b1, b2, b3, b4, width_align); \
 } \
 \
-static void vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, \
-                                          IDWTELEM *b3, IDWTELEM *b4, int width) \
+static void vertical_compose_dd97iH0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, \
+                                          uint8_t *_b3, uint8_t *_b4, int width) \
 { \
     int i, width_align = width&~(align-1); \
+    int16_t *b0 = (int16_t *)_b0; \
+    int16_t *b1 = (int16_t *)_b1; \
+    int16_t *b2 = (int16_t *)_b2; \
+    int16_t *b3 = (int16_t *)_b3; \
+    int16_t *b4 = (int16_t *)_b4; \
 \
     for(i=width_align; i<width; i++) \
         b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
 \
     ff_vertical_compose_dd97iH0##ext(b0, b1, b2, b3, b4, width_align); \
 } \
-static void vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width) \
+static void vertical_compose_haar##ext(uint8_t *_b0, uint8_t *_b1, int width) \
 { \
     int i, width_align = width&~(align-1); \
+    int16_t *b0 = (int16_t *)_b0; \
+    int16_t *b1 = (int16_t *)_b1; \
 \
     for(i=width_align; i<width; i++) { \
         b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]); \
@@ -85,10 +103,13 @@ static void vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width) \
 \
     ff_vertical_compose_haar##ext(b0, b1, width_align); \
 } \
-static void horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
+static void horizontal_compose_haar0i##ext(uint8_t *_b, uint8_t *_tmp, int w)\
 {\
     int w2= w>>1;\
     int x= w2 - (w2&(align-1));\
+    int16_t *b = (int16_t *)_b; \
+    int16_t *tmp = (int16_t *)_tmp; \
+\
     ff_horizontal_compose_haar0i##ext(b, tmp, w);\
 \
     for (; x < w2; x++) {\
@@ -96,10 +117,13 @@ static void horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
         b[2*x+1] = COMPOSE_HAARiH0(b[x+w2], tmp[x]);\
     }\
 }\
-static void horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
+static void horizontal_compose_haar1i##ext(uint8_t *_b, uint8_t *_tmp, int w)\
 {\
     int w2= w>>1;\
     int x= w2 - (w2&(align-1));\
+    int16_t *b = (int16_t *)_b; \
+    int16_t *tmp = (int16_t *)_tmp; \
+\
     ff_horizontal_compose_haar1i##ext(b, tmp, w);\
 \
     for (; x < w2; x++) {\
@@ -116,12 +140,15 @@ COMPOSE_VERTICAL(_mmx, 4)
 COMPOSE_VERTICAL(_sse2, 8)
 
 
-void ff_horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w);
+void ff_horizontal_compose_dd97i_ssse3(int16_t *_b, int16_t *_tmp, int w);
 
-static void horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w)
+static void horizontal_compose_dd97i_ssse3(uint8_t *_b, uint8_t *_tmp, int w)
 {
     int w2= w>>1;
     int x= w2 - (w2&7);
+    int16_t *b = (int16_t *)_b;
+    int16_t *tmp = (int16_t *)_tmp;
+
     ff_horizontal_compose_dd97i_ssse3(b, tmp, w);
 
     for (; x < w2; x++) {

libavcodec/x86/diracdsp_mmx.c

@@ -130,7 +130,7 @@ void ff_diracdsp_init_mmx(DiracDSPContext* c)
     c->add_dirac_obmc[2] = ff_add_dirac_obmc32_mmx;
     c->dirac_hpel_filter = dirac_hpel_filter_mmx;
     c->add_rect_clamped = ff_add_rect_clamped_mmx;
-    c->put_signed_rect_clamped = ff_put_signed_rect_clamped_mmx;
+    c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_mmx;
 #endif
     PIXFUNC(put, 0, mmx);
     PIXFUNC(avg, 0, mmx);
@@ -143,7 +143,7 @@ void ff_diracdsp_init_mmx(DiracDSPContext* c)
     if (EXTERNAL_SSE2(mm_flags)) {
         c->dirac_hpel_filter = dirac_hpel_filter_sse2;
         c->add_rect_clamped = ff_add_rect_clamped_sse2;
-        c->put_signed_rect_clamped = ff_put_signed_rect_clamped_sse2;
+        c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_sse2;
 
         c->add_dirac_obmc[1] = ff_add_dirac_obmc16_sse2;
         c->add_dirac_obmc[2] = ff_add_dirac_obmc32_sse2;

libavcodec/x86/diracdsp_yasm.asm

@@ -150,7 +150,7 @@ cglobal put_signed_rect_clamped_%1, 5,9,3, dst, dst_stride, src, src_stride, w,
 %endif
 
 .loopy:
-    lea     src2q, [srcq+src_strideq*2]
+    lea     src2q, [srcq+src_strideq]
     lea     dst2q, [dstq+dst_strideq]
 .loopx:
     sub      wd, mmsize
@@ -164,7 +164,7 @@ cglobal put_signed_rect_clamped_%1, 5,9,3, dst, dst_stride, src, src_stride, w,
     mova    [dst2q+wq], m2
     jg      .loopx
 
-    lea   srcq, [srcq+src_strideq*4]
+    lea   srcq, [srcq+src_strideq*2]
     lea   dstq, [dstq+dst_strideq*2]
     sub     hd, 2
     mov     wd, wspill