Bump dates to 2016

[x264] / common / x86 / trellis-64.asm

diff --git a/common/x86/trellis-64.asm b/common/x86/trellis-64.asm
index f0683bd8b363e5ea628a7a561b87f0c1e5acd6ba..bb1282df2a764be565fec1554207b4adee4d3432 100644 (file)
--- a/common/x86/trellis-64.asm
+++ b/common/x86/trellis-64.asm
@@ -1,7 +1,7 @@
 ;*****************************************************************************
 ;* trellis-64.asm: x86_64 trellis quantization
 ;*****************************************************************************
-;* Copyright (C) 2012 x264 project
+;* Copyright (C) 2012-2016 x264 project
 ;*
 ;* Authors: Loren Merritt <lorenm@u.washington.edu>
 ;*
@@ -86,7 +86,7 @@ SECTION .text
 %if cpuflag(ssse3)
     pabsd   m%1, m%1
     pmuludq m%1, m%1
-%elifdef HIGH_BIT_DEPTH
+%elif HIGH_BIT_DEPTH
     ABSD    m%2, m%1
     SWAP     %1, %2
     pmuludq m%1, m%1
@@ -96,6 +96,15 @@ SECTION .text
 %endif
 %endmacro
 
+%macro LOAD_DUP 2 ; dst, src
+%if cpuflag(ssse3)
+    movddup    %1, %2
+%else
+    movd       %1, %2
+    punpcklqdq %1, %1
+%endif
+%endmacro
+
 ;-----------------------------------------------------------------------------
 ; int trellis_cabac_4x4_psy(
 ;     const int *unquant_mf, const uint8_t *zigzag, int lambda2,
@@ -113,7 +122,7 @@ cglobal %1, 4,15,9
     %assign pad 96 + level_tree_size + 16*SIZEOF_NODE + 16-gprsize-(stack_offset&15)
     SUB  rsp, pad
     DEFINE_ARGS unquant_mf, zigzag, lambda2, ii, orig_coefs, quant_coefs, dct, cabac_state_sig, cabac_state_last
-%ifdef WIN64
+%if WIN64
     %define level_statem rsp+stack_offset+80 ; r9m, except that we need to index into it (and r10m) as an array
 %else
     %define level_statem rsp+stack_offset+32
@@ -137,7 +146,7 @@ cglobal %1, 4,15,9
     %define zigzagm   [stack+8]
     mov     last_nnzm, iid
     mov     zigzagm,   zigzagq
-%ifndef WIN64
+%if WIN64 == 0
     %define orig_coefsm  [stack+16]
     %define quant_coefsm [stack+24]
     mov     orig_coefsm,  orig_coefsq
@@ -186,12 +195,11 @@ cglobal %1, 4,15,9
     mov dword levelgt1_ctxm, 9
 %endif
 %if psy
-    movd    m6, psy_trellism
+    LOAD_DUP m6, psy_trellism
     %define psy_trellis m6
 %elif dc
-    movd       m6, [unquant_mfq]
+    LOAD_DUP   m6, [unquant_mfq]
     paddd      m6, m6
-    punpcklqdq m6, m6
     %define unquant_mf m6
 %endif
 %ifdef PIC
@@ -232,7 +240,7 @@ cglobal %1, 4,15,9
     movzx   r0, word [level_tree + r0*4]
     psrld   m0, 16
     movd    m1, [dctq + r2*SIZEOF_DCTCOEF]
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
     psignd  m0, m1
     movd [dctq + r2*SIZEOF_DCTCOEF], m0
 %else
@@ -242,7 +250,7 @@ cglobal %1, 4,15,9
 %endif
 %else
     mov    r5d, [level_tree + r0*4]
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
     mov    r4d, dword [dctq + r2*SIZEOF_DCTCOEF]
 %else
     movsx  r4d, word [dctq + r2*SIZEOF_DCTCOEF]
@@ -252,7 +260,7 @@ cglobal %1, 4,15,9
     shr    r5d, 16
     xor    r5d, r4d
     sub    r5d, r4d
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
     mov  [dctq + r2*SIZEOF_DCTCOEF], r5d
 %else
     mov  [dctq + r2*SIZEOF_DCTCOEF], r5w
@@ -261,9 +269,7 @@ cglobal %1, 4,15,9
     inc    iiq
     jle .writeback_loop
 
-%if dc
     mov eax, 1
-%endif
 .return:
     ADD rsp, pad
     RET
@@ -273,7 +279,7 @@ cglobal %1, 4,15,9
     pxor       m0, m0
     mova [r10+ 0], m0
     mova [r10+16], m0
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
     mova [r10+32], m0
     mova [r10+48], m0
 %endif
@@ -287,7 +293,7 @@ cglobal %1, 4,15,9
 .i_loop%1:
     ; if( !quant_coefs[i] )
     mov   r6, quant_coefsm
-%ifdef HIGH_BIT_DEPTH
+%if HIGH_BIT_DEPTH
     mov   abs_leveld, dword [r6 + iiq*SIZEOF_DCTCOEF]
 %else
     movsx abs_leveld, word [r6 + iiq*SIZEOF_DCTCOEF]
@@ -334,14 +340,13 @@ cglobal %1, 4,15,9
     movzx   zigzagid, byte [zigzagq+iiq]
     movd    m0, abs_leveld
     mov     r6, orig_coefsm
-%ifdef HIGH_BIT_DEPTH
-    movd    m1, [r6 + zigzagiq*SIZEOF_DCTCOEF]
+%if HIGH_BIT_DEPTH
+    LOAD_DUP m1, [r6 + zigzagiq*SIZEOF_DCTCOEF]
 %else
-    movd    m1, [r6 + zigzagiq*SIZEOF_DCTCOEF - 2]
-    psrad   m1, 16
+    LOAD_DUP m1, [r6 + zigzagiq*SIZEOF_DCTCOEF - 2]
+    psrad    m1, 16     ; sign_coef
 %endif
     punpcklqdq m0, m0 ; quant_coef
-    punpcklqdq m1, m1 ; sign_coef
 %if cpuflag(ssse3)
     pabsd   m0, m0
     pabsd   m2, m1 ; abs_coef
@@ -405,11 +410,10 @@ cglobal %1, 4,15,9
 %else
 %ifdef PIC
     mov    r10, unquant_mfm
-    movd    m3, [r10 + zigzagiq*4]
+    LOAD_DUP m3, [r10 + zigzagiq*4]
 %else
-    movd    m3, [unquant_mfq + zigzagiq*4]
+    LOAD_DUP m3, [unquant_mfq + zigzagiq*4]
 %endif
-    punpcklqdq m3, m3
     pmuludq m0, m3
 %endif
     paddd   m0, [pq_128]
@@ -422,8 +426,7 @@ cglobal %1, 4,15,9
 %if dc
     psllq   m0, 8
 %else
-    movd    m5, [dct_weight2_tab + zigzagiq*4 GLOBAL]
-    punpcklqdq m5, m5
+    LOAD_DUP m5, [dct_weight2_tab + zigzagiq*4 GLOBAL]
     pmuludq m0, m5
 %endif
 
@@ -435,13 +438,12 @@ cglobal %1, 4,15,9
     ; int psy_weight = dct_weight_tab[zigzag[i]] * h->mb.i_psy_trellis;
     ; ssd1[k] -= psy_weight * psy_value;
     mov     r6, fenc_dctm
-%ifdef HIGH_BIT_DEPTH
-    movd    m3, [r6 + zigzagiq*SIZEOF_DCTCOEF]
+%if HIGH_BIT_DEPTH
+    LOAD_DUP m3, [r6 + zigzagiq*SIZEOF_DCTCOEF]
 %else
-    movd    m3, [r6 + zigzagiq*SIZEOF_DCTCOEF - 2]
+    LOAD_DUP m3, [r6 + zigzagiq*SIZEOF_DCTCOEF - 2]
     psrad   m3, 16 ; orig_coef
 %endif
-    punpcklqdq m3, m3
 %if cpuflag(ssse3)
     psignd  m4, m1 ; SIGN(unquant_abs_level, sign_coef)
 %else
@@ -455,9 +457,8 @@ cglobal %1, 4,15,9
     ABSD    m3, m4
     SWAP     4, 3
 %endif
-    movd    m1, [dct_weight1_tab + zigzagiq*4 GLOBAL]
+    LOAD_DUP m1, [dct_weight1_tab + zigzagiq*4 GLOBAL]
     pmuludq m1, psy_trellis
-    punpcklqdq m1, m1
     pmuludq m4, m1
     psubq   m0, m4
 %if %1
@@ -724,6 +725,7 @@ TRELLIS_COEF0 1
     mov [nodes_curq + node_cabac_state(node_ctx) + (coeff_abs_level1_offs>>2)], r11b ; delayed from x264_cabac_size_decision2
 %endif
 %if %1 && node_ctx == 7
+    mov  r6d, levelgt1_ctxm
     mov [nodes_curq + node_cabac_state(node_ctx) + coeff_abs_levelgt1_offs-6], r10b
 %endif
     mov  r6d, [nodes_prevq + node_level_idx(j)]
@@ -786,8 +788,8 @@ TRELLIS_COEF0 1
     add  bitsd, r5d ; bs_size_ue_big from COEFN_SUFFIX
     ; n.cabac_state[levelgt1_ctx]
 %if j == 7 ; && compiling support for 4:2:2
-    mov    r5d, levelgt1_ctxm
-    %define coeff_abs_levelgt1_offs r5
+    mov    r6d, levelgt1_ctxm
+    %define coeff_abs_levelgt1_offs r6
 %endif
 %if j == 7
     movzx  r10, byte [nodes_prevq + node_cabac_state(j) + coeff_abs_levelgt1_offs-6] ; -6 because node only stores ctx 8 and 9