Better x86 high bit depth predict_8x8c_p

Avoid the need to check for corner cases by reordering arithmetic.
Also make a minor optimization to high bit depth predict_16x16_p.

diff --git a/common/x86/predict-a.asm b/common/x86/predict-a.asm
index 1876d96a982c2f507495780ee62e56b43f91063d..4b955180cf40f163f3990c692680a1e604573425 100644 (file)
--- a/common/x86/predict-a.asm
+++ b/common/x86/predict-a.asm
@@ -31,10 +31,12 @@
 SECTION_RODATA
 
 pw_76543210:
-pw_3210:    dw 0, 1, 2, 3, 4, 5, 6, 7
-pb_00s_ff:  times 8 db 0
-pb_0s_ff:   times 7 db 0
-            db 0xff
+pw_3210:     dw 0, 1, 2, 3, 4, 5, 6, 7
+pw_43210123: dw -3, -2, -1, 0, 1, 2, 3, 4
+pw_m3:       times 8 dw -3
+pb_00s_ff:   times 8 db 0
+pb_0s_ff:    times 7 db 0
+             db 0xff
 
 SECTION .text
 
@@ -45,6 +47,7 @@ cextern pw_1
 cextern pw_2
 cextern pw_4
 cextern pw_8
+cextern pw_16
 cextern pw_ff00
 cextern pb_reverse
 cextern pw_pixel_max
@@ -1045,32 +1048,42 @@ ALIGN 4
     REP_RET
 %endif ; !ARCH_X86_64
 
-%macro PREDICT_8x8C_P 1
-cglobal predict_8x8c_p_core_%1, 1,1
+INIT_XMM
+%ifdef HIGH_BIT_DEPTH
+cglobal predict_8x8c_p_core_sse2, 1,1,7
     movd        m0, r1m
     movd        m2, r2m
     movd        m4, r3m
-%ifdef HIGH_BIT_DEPTH
     mova        m3, [pw_pixel_max]
     pxor        m1, m1
-%endif
     SPLATW      m0, m0, 0
     SPLATW      m2, m2, 0
     SPLATW      m4, m4, 0
-    pmullw      m2, [pw_76543210]
-%ifdef HIGH_BIT_DEPTH
+    pmullw      m2, [pw_43210123] ; b
+    pmullw      m5, m4, [pw_m3]   ; c
+    paddw       m5, [pw_16]
     mov        r1d, 8
 .loop:
-    paddsw      m5, m0, m2
-    psraw       m5, 5
-    CLIPW       m5, m1, m3
-    mova      [r0], m5
-    paddw       m2, m4
+    paddsw      m6, m2, m5
+    paddsw      m6, m0
+    psraw       m6, 5
+    CLIPW       m6, m1, m3
+    mova      [r0], m6
+    paddw       m5, m4
     add         r0, FDEC_STRIDEB
     dec r1d
     jg .loop
-%else ;!HIGH_BIT_DEPTH
-    paddsw      m0, m2        ; m0 = {i+0*b, i+1*b, i+2*b, i+3*b, i+4*b, i+5*b, i+6*b, i+7*b}
+    REP_RET
+%else ; !HIGH_BIT_DEPTH
+cglobal predict_8x8c_p_core_sse2, 1,1
+    movd        m0, r1m
+    movd        m2, r2m
+    movd        m4, r3m
+    SPLATW      m0, m0, 0
+    SPLATW      m2, m2, 0
+    SPLATW      m4, m4, 0
+    pmullw      m2, [pw_76543210]
+    paddsw      m0, m2            ; m0 = {i+0*b, i+1*b, i+2*b, i+3*b, i+4*b, i+5*b, i+6*b, i+7*b}
     paddsw      m3, m0, m4
     paddsw      m4, m4
 call .loop
@@ -1090,12 +1103,8 @@ call .loop
     packuswb    m5, m1
     movq        [r0+FDEC_STRIDE*2], m5
     movhps      [r0+FDEC_STRIDE*3], m5
-%endif ;!HIGH_BIT_DEPTH
     RET
-%endmacro ; PREDICT_8x8C_P
-
-INIT_XMM
-PREDICT_8x8C_P sse2
+%endif ; HIGH_BIT_DEPTH
 
 ;-----------------------------------------------------------------------------
 ; void predict_16x16_p_core( uint8_t *src, int i00, int b, int c )
@@ -1146,16 +1155,13 @@ cglobal predict_16x16_p_core_%1, 1,2,8
     movd     m0, r1m
     movd     m1, r2m
     movd     m2, r3m
-%ifdef HIGH_BIT_DEPTH
-    pxor     m6, m6
-    pxor     m7, m7
-%endif
     SPLATW   m0, m0, 0
     SPLATW   m1, m1, 0
     SPLATW   m2, m2, 0
     pmullw   m3, m1, [pw_76543210]
     psllw    m1, 3
 %ifdef HIGH_BIT_DEPTH
+    pxor     m6, m6
     mov     r1d, 16
 .loop:
     mova     m4, m0

diff --git a/common/x86/predict-c.c b/common/x86/predict-c.c
index 24192a2c9b258ebd2764e80612e3a95b60e61164..2766618eb8eb71ce77f2469437fc2017f3dbf38f 100644 (file)
--- a/common/x86/predict-c.c
+++ b/common/x86/predict-c.c
@@ -315,12 +315,11 @@ static void x264_predict_8x8c_p_ssse3( uint8_t *src )
     b = ( 17 * H + 16 ) >> 5;
     c = ( 17 * V + 16 ) >> 5;
     i00 = a -3*b -3*c + 16;
-    /* b*7 + c*7 can overflow: it's easier to just branch away in this rare case
-     * than to try to consider it in the asm. */
-    if( BIT_DEPTH > 8 && (i00 > 0x7fff || abs(b) > 2340 || abs(c) > 2340) )
-        x264_predict_8x8c_p_c( src );
-    else
-        x264_predict_8x8c_p_core_sse2( src, i00, b, c );
+#if HIGH_BIT_DEPTH
+    x264_predict_8x8c_p_core_sse2( src, a, b, c );
+#else
+    x264_predict_8x8c_p_core_sse2( src, i00, b, c );
+#endif
 }
 #endif
 #if !HIGH_BIT_DEPTH