Merge "Change use of eob in the encoder"

14 files changed, 64 insertions, 57 deletions

diff --git a/vp8/common/blockd.h b/vp8/common/blockd.h
index 5012edd61..a90c1c0b6 100644
--- a/vp8/common/blockd.h
+++ b/vp8/common/blockd.h
@@ -187,7 +187,7 @@ typedef struct
     int dst;
     int dst_stride;
 
-    int eob;
+    char *eob;
 
     union b_mode_info bmi;
 } BLOCKD;
diff --git a/vp8/common/invtrans.c b/vp8/common/invtrans.c
index 7712b59b7..090c19cca 100644
--- a/vp8/common/invtrans.c
+++ b/vp8/common/invtrans.c
@@ -15,7 +15,7 @@
 void vp8_inverse_transform_b(const vp8_idct_rtcd_vtable_t *rtcd, BLOCKD *b,
                              int pitch)
 {
-    if (b->eob > 1)
+    if (*b->eob > 1)
     {
         IDCT_INVOKE(rtcd, idct16)(b->dqcoeff, b->predictor, pitch,
               *(b->base_dst) + b->dst, b->dst_stride);
@@ -65,6 +65,3 @@ void vp8_inverse_transform_mbuv(const vp8_idct_rtcd_vtable_t *rtcd, MACROBLOCKD
     }
 
 }
-
-
-
diff --git a/vp8/common/mbpitch.c b/vp8/common/mbpitch.c
index 054042c0b..11fa3ffa7 100644
--- a/vp8/common/mbpitch.c
+++ b/vp8/common/mbpitch.c
@@ -118,6 +118,7 @@ void vp8_setup_block_dptrs(MACROBLOCKD *x)
     {
         x->block[r].qcoeff  = x->qcoeff  + r * 16;
         x->block[r].dqcoeff = x->dqcoeff + r * 16;
+        x->block[r].eob     = x->eobs + r;
     }
 }
 
diff --git a/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm b/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm
index ae2f6030d..d61f5d94d 100644
--- a/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm
+++ b/vp8/encoder/arm/armv6/vp8_fast_quantize_b_armv6.asm
@@ -102,9 +102,10 @@ loop
     bne     loop
 
     ; PART 2: check position for eob...
+    ldr     r11, [sp, #0]           ; restore BLOCKD pointer
     mov     lr, #0                  ; init eob
     cmp     r1, #0                  ; coeffs after quantization?
-    ldr     r11, [sp, #0]           ; restore BLOCKD pointer
+    ldr     r12, [r11, #vp8_blockd_eob]
     beq     end                     ; skip eob calculations if all zero
 
     ldr     r0, [r11, #vp8_blockd_qcoeff]
@@ -212,7 +213,7 @@ quant_coeff_1_0
     mov     lr, #1              ; rc=0,  i=0
 
 end
-    str     lr, [r11, #vp8_blockd_eob]
+    strb    lr, [r12]
     ldmfd   sp!, {r1, r4-r11, pc}
 
     ENDP
diff --git a/vp8/encoder/arm/neon/fastquantizeb_neon.asm b/vp8/encoder/arm/neon/fastquantizeb_neon.asm
index dcf3c5090..c970cb73e 100644
--- a/vp8/encoder/arm/neon/fastquantizeb_neon.asm
+++ b/vp8/encoder/arm/neon/fastquantizeb_neon.asm
@@ -135,17 +135,16 @@
     vmovl.u16       q0, d0
     vmovl.u16       q10, d20
 
-
     vmax.u32        d0, d0, d1
     vmax.u32        d20, d20, d21
     vpmax.u32       d0, d0, d0
     vpmax.u32       d20, d20, d20
 
-    add             r4, r2, #vp8_blockd_eob
-    add             r5, r3, #vp8_blockd_eob
+    ldr             r4, [r2, #vp8_blockd_eob]
+    ldr             r5, [r3, #vp8_blockd_eob]
 
-    vst1.32         {d0[0]}, [r4@32]
-    vst1.32         {d20[0]}, [r5@32]
+    vst1.8          {d0[0]}, [r4]       ; store eob
+    vst1.8          {d20[0]}, [r5]      ; store eob
 
     vldmia          sp!, {q4-q7}
     ldmfd           sp!, {r4-r9}
@@ -196,6 +195,8 @@
     vshr.s16        q12, #1             ; right shift 1 after vqdmulh
     vshr.s16        q13, #1
 
+    ldr             r5, [r1, #vp8_blockd_eob]
+
     orr             r2, r2, r3          ; check if all zero (step 4)
     cmp             r2, #0              ; check if all zero (step 5)
     beq             zero_output         ; check if all zero (step 6)
@@ -230,14 +231,13 @@
 
     vst1.s16        {q2, q3}, [r7@128]  ; store dqcoeff = x * Dequant
 
-    add             r4, r1, #vp8_blockd_eob
-    vst1.32         {d0[0]}, [r4@32]
+    vst1.8          {d0[0]}, [r5]       ; store eob
 
     ldmfd           sp!, {r4-r7}
     bx              lr
 
 zero_output
-    str             r2, [r1, #vp8_blockd_eob]
+    strb            r2, [r5]            ; store eob
     vst1.s16        {q0, q1}, [r6@128]  ; qcoeff = 0
     vst1.s16        {q0, q1}, [r7@128]  ; dqcoeff = 0
 
diff --git a/vp8/encoder/arm/quantize_arm.c b/vp8/encoder/arm/quantize_arm.c
index 52d84013e..5b3a0275f 100644
--- a/vp8/encoder/arm/quantize_arm.c
+++ b/vp8/encoder/arm/quantize_arm.c
@@ -46,7 +46,7 @@ void vp8_quantize_mb_neon(MACROBLOCK *x)
                            &x->e_mbd.block[i], &x->e_mbd.block[i+1]);
 
     if (has_2nd_order)
-        x->quantize_b(&x->block[i], &x->e_mbd.block[i]);
+        x->quantize_b(&x->block[24], &x->e_mbd.block[24]);
 }
 
 
diff --git a/vp8/encoder/encodemb.c b/vp8/encoder/encodemb.c
index 16f6875b1..faa1a8e33 100644
--- a/vp8/encoder/encodemb.c
+++ b/vp8/encoder/encodemb.c
@@ -274,7 +274,7 @@ static void optimize_b(MACROBLOCK *mb, int ib, int type,
     qcoeff_ptr = d->qcoeff;
     dqcoeff_ptr = d->dqcoeff;
     i0 = !type;
-    eob = d->eob;
+    eob = *d->eob;
 
     /* Now set up a Viterbi trellis to evaluate alternative roundings. */
     rdmult = mb->rdmult * err_mult;
@@ -466,8 +466,8 @@ static void optimize_b(MACROBLOCK *mb, int ib, int type,
     }
     final_eob++;
 
-    d->eob = final_eob;
-    *a = *l = (d->eob != !type);
+    *a = *l = (final_eob != !type);
+    *d->eob = (char)final_eob;
 }
 static void check_reset_2nd_coeffs(MACROBLOCKD *x, int type,
                                    ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l)
@@ -650,7 +650,7 @@ static void inverse_transform_mb(const vp8_idct_rtcd_vtable_t *rtcd,
     {
         BLOCKD *b = &x->block[i];
 
-        if (b->eob > 1)
+        if (*b->eob > 1)
         {
             IDCT_INVOKE(rtcd, idct16)(b->dqcoeff, b->predictor, 16,
                   *(b->base_dst) + b->dst, b->dst_stride);
@@ -667,7 +667,7 @@ static void inverse_transform_mb(const vp8_idct_rtcd_vtable_t *rtcd,
     {
         BLOCKD *b = &x->block[i];
 
-        if (b->eob > 1)
+        if (*b->eob > 1)
         {
             IDCT_INVOKE(rtcd, idct16)(b->dqcoeff, b->predictor, 8,
                   *(b->base_dst) + b->dst, b->dst_stride);
diff --git a/vp8/encoder/quantize.c b/vp8/encoder/quantize.c
index 503d24123..22cbbee85 100644
--- a/vp8/encoder/quantize.c
+++ b/vp8/encoder/quantize.c
@@ -62,7 +62,7 @@ void vp8_fast_quantize_b_c(BLOCK *b, BLOCKD *d)
             }
         }
     }
-    d->eob = eob + 1;
+    *d->eob = (char)(eob + 1);
 }
 
 #else
@@ -97,7 +97,7 @@ void vp8_fast_quantize_b_c(BLOCK *b, BLOCKD *d)
             eob = i;                                // last nonzero coeffs
         }
     }
-    d->eob = eob + 1;
+    *d->eob = (char)(eob + 1);
 }
 
 #endif
@@ -152,7 +152,7 @@ void vp8_regular_quantize_b(BLOCK *b, BLOCKD *d)
         }
     }
 
-    d->eob = eob + 1;
+    *d->eob = (char)(eob + 1);
 }
 
 /* Perform regular quantization, with unbiased rounding and no zero bin. */
@@ -210,7 +210,7 @@ void vp8_strict_quantize_b(BLOCK *b, BLOCKD *d)
         }
     }
 
-    d->eob = eob + 1;
+    *d->eob = (char)(eob + 1);
 }
 
 #else
@@ -264,7 +264,7 @@ void vp8_regular_quantize_b(BLOCK *b, BLOCKD *d)
         }
     }
 
-    d->eob = eob + 1;
+    *d->eob = (char)(eob + 1);
 }
 
 #endif
@@ -731,4 +731,3 @@ void vp8_set_quantizer(struct VP8_COMP *cpi, int Q)
         vp8cx_init_quantizer(cpi);
 
 }
-
diff --git a/vp8/encoder/rdopt.c b/vp8/encoder/rdopt.c
index fdb519c19..7950960de 100644
--- a/vp8/encoder/rdopt.c
+++ b/vp8/encoder/rdopt.c
@@ -485,7 +485,7 @@ int VP8_UVSSE(MACROBLOCK *x, const vp8_variance_rtcd_vtable_t *rtcd)
 static int cost_coeffs(MACROBLOCK *mb, BLOCKD *b, int type, ENTROPY_CONTEXT *a, ENTROPY_CONTEXT *l)
 {
     int c = !type;              /* start at coef 0, unless Y with Y2 */
-    int eob = b->eob;
+    int eob = (int)(*b->eob);
     int pt ;    /* surrounding block/prev coef predictor */
     int cost = 0;
     short *qcoeff_ptr = b->qcoeff;
@@ -1299,11 +1299,9 @@ static void rd_check_segment(VP8_COMP *cpi, MACROBLOCK *x,
         // store everything needed to come back to this!!
         for (i = 0; i < 16; i++)
         {
-            BLOCKD *bd = &x->e_mbd.block[i];
-
             bsi->mvs[i].as_mv = x->partition_info->bmi[i].mv.as_mv;
             bsi->modes[i] = x->partition_info->bmi[i].mode;
-            bsi->eobs[i] = bd->eob;
+            bsi->eobs[i] = x->e_mbd.eobs[i];
         }
     }
 }
@@ -1432,7 +1430,7 @@ static int vp8_rd_pick_best_mbsegmentation(VP8_COMP *cpi, MACROBLOCK *x,
         BLOCKD *bd = &x->e_mbd.block[i];
 
         bd->bmi.mv.as_int = bsi.mvs[i].as_int;
-        bd->eob = bsi.eobs[i];
+        *bd->eob = bsi.eobs[i];
     }
 
     *returntotrate = bsi.r;
@@ -2271,7 +2269,7 @@ void vp8_rd_pick_inter_mode(VP8_COMP *cpi, MACROBLOCK *x, int recon_yoffset, int
 
                 for (i = 0; i <= 24; i++)
                 {
-                    tteob += x->e_mbd.block[i].eob;
+                    tteob += x->e_mbd.eobs[i];
                 }
 
                 if (tteob == 0)
diff --git a/vp8/encoder/tokenize.c b/vp8/encoder/tokenize.c
index c8db4f067..e81948567 100644
--- a/vp8/encoder/tokenize.c
+++ b/vp8/encoder/tokenize.c
@@ -108,15 +108,16 @@ static void tokenize2nd_order_b
     ENTROPY_CONTEXT * a;
     ENTROPY_CONTEXT * l;
     int band, rc, v, token;
+    int eob;
 
     b = x->block + 24;
     qcoeff_ptr = b->qcoeff;
     a = (ENTROPY_CONTEXT *)x->above_context + 8;
     l = (ENTROPY_CONTEXT *)x->left_context + 8;
-
+    eob = x->eobs[24];
     VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
 
-    if(!b->eob)
+    if(!eob)
     {
         /* c = band for this case */
         t->Token = DCT_EOB_TOKEN;
@@ -142,7 +143,7 @@ static void tokenize2nd_order_b
     t++;
     c = 1;
 
-    for (; c < b->eob; c++)
+    for (; c < eob; c++)
     {
         rc = vp8_default_zig_zag1d[c];
         band = vp8_coef_bands[c];
@@ -213,7 +214,7 @@ static void tokenize1st_order_b
 
         c = type ? 0 : 1;
 
-        if(c >= b->eob)
+        if(c >= *b->eob)
         {
             /* c = band for this case */
             t->Token = DCT_EOB_TOKEN;
@@ -240,7 +241,7 @@ static void tokenize1st_order_b
         t++;
         c++;
 
-        for (; c < b->eob; c++)
+        for (; c < *b->eob; c++)
         {
             rc = vp8_default_zig_zag1d[c];
             band = vp8_coef_bands[c];
@@ -284,7 +285,7 @@ static void tokenize1st_order_b
 
         VP8_COMBINEENTROPYCONTEXTS(pt, *a, *l);
 
-        if(!b->eob)
+        if(!(*b->eob))
         {
             /* c = band for this case */
             t->Token = DCT_EOB_TOKEN;
@@ -311,7 +312,7 @@ static void tokenize1st_order_b
         t++;
         c = 1;
 
-        for (; c < b->eob; c++)
+        for (; c < *b->eob; c++)
         {
             rc = vp8_default_zig_zag1d[c];
             band = vp8_coef_bands[c];
@@ -356,11 +357,11 @@ static int mb_is_skippable(MACROBLOCKD *x, int has_y2_block)
     if (has_y2_block)
     {
         for (i = 0; i < 16; i++)
-            skip &= (x->block[i].eob < 2);
+            skip &= (x->eobs[i] < 2);
     }
 
     for (; i < 24 + has_y2_block; i++)
-        skip &= (!x->block[i].eob);
+        skip &= (!x->eobs[i]);
 
     return skip;
 }
diff --git a/vp8/encoder/x86/quantize_sse2.asm b/vp8/encoder/x86/quantize_sse2.asm
index c483933df..7c249ff88 100644
--- a/vp8/encoder/x86/quantize_sse2.asm
+++ b/vp8/encoder/x86/quantize_sse2.asm
@@ -194,6 +194,8 @@ ZIGZAG_LOOP 15
     movdqa      [rdi], xmm0        ; store dqcoeff
     movdqa      [rdi + 16], xmm1
 
+    mov         rcx, [rsi + vp8_blockd_eob]
+
     ; select the last value (in zig_zag order) for EOB
     pcmpeqw     xmm2, xmm6
     pcmpeqw     xmm3, xmm6
@@ -214,7 +216,8 @@ ZIGZAG_LOOP 15
     pmaxsw      xmm2, xmm3
     movd        eax, xmm2
     and         eax, 0xff
-    mov         [rsi + vp8_blockd_eob], eax
+
+    mov         BYTE PTR [rcx], al          ; store eob
 
     ; begin epilog
     add         rsp, stack_size
@@ -337,6 +340,8 @@ sym(vp8_fast_quantize_b_sse2):
 
     pmaxsw      xmm1, xmm5
 
+    mov         rcx, [rsi + vp8_blockd_eob]
+
     ; now down to 8
     pshufd      xmm5, xmm1, 00001110b
 
@@ -354,7 +359,8 @@ sym(vp8_fast_quantize_b_sse2):
 
     movd        eax, xmm1
     and         eax, 0xff
-    mov         [rsi + vp8_blockd_eob], eax
+
+    mov         BYTE PTR [rcx], al          ; store eob
 
     ; begin epilog
 %if ABI_IS_32BIT
diff --git a/vp8/encoder/x86/quantize_sse4.asm b/vp8/encoder/x86/quantize_sse4.asm
index 95e1c2074..70eac0c0f 100644
--- a/vp8/encoder/x86/quantize_sse4.asm
+++ b/vp8/encoder/x86/quantize_sse4.asm
@@ -208,6 +208,8 @@ ZIGZAG_LOOP 15, 7, xmm3, xmm7, xmm8
     movdqa      [rdi], xmm0
     movdqa      [rdi + 16], xmm1
 
+    mov         rcx, [rsi + vp8_blockd_eob]
+
     ; select the last value (in zig_zag order) for EOB
     pxor        xmm6, xmm6
     pcmpeqw     xmm4, xmm6
@@ -225,7 +227,7 @@ ZIGZAG_LOOP 15, 7, xmm3, xmm7, xmm8
     add         eax, 1
     and         eax, edi
 
-    mov         [rsi + vp8_blockd_eob], eax
+    mov         BYTE PTR [rcx], al          ; store eob
 
     ; begin epilog
 %if ABI_IS_32BIT
diff --git a/vp8/encoder/x86/quantize_ssse3.asm b/vp8/encoder/x86/quantize_ssse3.asm
index 912007e02..34cc9c3bb 100644
--- a/vp8/encoder/x86/quantize_ssse3.asm
+++ b/vp8/encoder/x86/quantize_ssse3.asm
@@ -110,12 +110,14 @@ sym(vp8_fast_quantize_b_ssse3):
     movdqa      [rcx], xmm2                 ;store dqcoeff
     movdqa      [rcx + 16], xmm3            ;store dqcoeff
 
+    mov         rcx, [rsi + vp8_blockd_eob]
+
     sub         edi, edx                    ;check for all zeros in bit mask
     sar         edi, 31                     ;0 or -1
     add         eax, 1
     and         eax, edi                    ;if the bit mask was all zero,
                                             ;then eob = 0
-    mov         [rsi + vp8_blockd_eob], eax
+    mov         BYTE PTR [rcx], al          ;store eob
 
     ; begin epilog
 %if ABI_IS_32BIT
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index 191d61c60..7f5208461 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -37,17 +37,17 @@ void vp8_fast_quantize_b_mmx(BLOCK *b, BLOCKD *d)
     short *dqcoeff_ptr = d->dqcoeff;
     short *dequant_ptr = d->dequant;
 
-    d->eob = vp8_fast_quantize_b_impl_mmx(
-                 coeff_ptr,
-                 zbin_ptr,
-                 qcoeff_ptr,
-                 dequant_ptr,
-                 scan_mask,
-
-                 round_ptr,
-                 quant_ptr,
-                 dqcoeff_ptr
-             );
+    *d->eob = (char)vp8_fast_quantize_b_impl_mmx(
+                                                 coeff_ptr,
+                                                 zbin_ptr,
+                                                 qcoeff_ptr,
+                                                 dequant_ptr,
+                                                 scan_mask,
+
+                                                 round_ptr,
+                                                 quant_ptr,
+                                                 dqcoeff_ptr
+                                                 );
 }
 
 int vp8_mbblock_error_mmx_impl(short *coeff_ptr, short *dcoef_ptr, int dc);