Merge "Added vp8_fast_quantize_b_sse2"

3 files changed, 142 insertions, 160 deletions

diff --git a/vp8/encoder/x86/quantize_mmx.asm b/vp8/encoder/x86/quantize_mmx.asm
index 51cd94078..f29a54ecd 100644
--- a/vp8/encoder/x86/quantize_mmx.asm
+++ b/vp8/encoder/x86/quantize_mmx.asm
@@ -284,156 +284,3 @@ sym(vp8_fast_quantize_b_impl_mmx):
     UNSHADOW_ARGS
     pop         rbp
     ret
-
-
-;int vp8_fast_quantize_b_impl_sse(short *coeff_ptr, short *zbin_ptr,
-;                           short *qcoeff_ptr,short *dequant_ptr,
-;                           short *scan_mask, short *round_ptr,
-;                           short *quant_ptr, short *dqcoeff_ptr);
-global sym(vp8_fast_quantize_b_impl_sse)
-sym(vp8_fast_quantize_b_impl_sse):
-    push        rbp
-    mov         rbp, rsp
-    SHADOW_ARGS_TO_STACK 8
-    push rsi
-    push rdi
-    ; end prolog
-
-
-        mov             rsi,        arg(0) ;coeff_ptr
-        movdqa          xmm0,       [rsi]
-
-        mov             rax,        arg(1) ;zbin_ptr
-        movdqa          xmm1,       [rax]
-
-        movdqa          xmm3,       xmm0
-        psraw           xmm0,       15
-
-        pxor            xmm3,       xmm0
-        psubw           xmm3,       xmm0            ; abs
-
-        movdqa          xmm2,       xmm3
-        pcmpgtw         xmm1,       xmm2
-
-        pandn           xmm1,       xmm2
-        movdqa          xmm3,       xmm1
-
-        mov             rdx,        arg(6) ; quant_ptr
-        movdqa          xmm1,       [rdx]
-
-        mov             rcx,        arg(5) ; round_ptr
-        movdqa          xmm2,       [rcx]
-
-        paddw           xmm3,       xmm2
-        pmulhuw         xmm3,       xmm1
-
-        pxor            xmm3,       xmm0
-        psubw           xmm3,       xmm0        ;gain the sign back
-
-        mov             rdi,        arg(2) ;qcoeff_ptr
-        movdqa          xmm0,       xmm3
-
-        movdqa          [rdi],      xmm3
-
-        mov             rax,        arg(3) ;dequant_ptr
-        movdqa          xmm2,       [rax]
-
-        pmullw          xmm3,       xmm2
-        mov             rax,        arg(7) ;dqcoeff_ptr
-
-        movdqa          [rax],      xmm3
-
-        ; next 8
-        movdqa          xmm4,       [rsi+16]
-
-        mov             rax,        arg(1) ;zbin_ptr
-        movdqa          xmm5,       [rax+16]
-
-        movdqa          xmm7,       xmm4
-        psraw           xmm4,       15
-
-        pxor            xmm7,       xmm4
-        psubw           xmm7,       xmm4            ; abs
-
-        movdqa          xmm6,       xmm7
-        pcmpgtw         xmm5,       xmm6
-
-        pandn           xmm5,       xmm6
-        movdqa          xmm7,       xmm5
-
-        movdqa          xmm5,       [rdx+16]
-        movdqa          xmm6,       [rcx+16]
-
-
-        paddw           xmm7,       xmm6
-        pmulhuw         xmm7,       xmm5
-
-        pxor            xmm7,       xmm4
-        psubw           xmm7,       xmm4;gain the sign back
-
-        mov             rdi,        arg(2) ;qcoeff_ptr
-
-        movdqa          xmm1,       xmm7
-        movdqa          [rdi+16],   xmm7
-
-        mov             rax,        arg(3) ;dequant_ptr
-        movdqa          xmm6,       [rax+16]
-
-        pmullw          xmm7,       xmm6
-        mov             rax,        arg(7) ;dqcoeff_ptr
-
-        movdqa          [rax+16],   xmm7
-        mov             rdi,        arg(4) ;scan_mask
-
-        pxor            xmm7,       xmm7
-        movdqa          xmm2,       [rdi]
-
-        movdqa          xmm3,       [rdi+16];
-        pcmpeqw         xmm0,       xmm7
-
-        pcmpeqw         xmm1,       xmm7
-        pcmpeqw         xmm6,       xmm6
-
-        pxor            xmm0,       xmm6
-        pxor            xmm1,       xmm6
-
-        psrlw           xmm0,       15
-        psrlw           xmm1,       15
-
-        pmaddwd         xmm0,       xmm2
-        pmaddwd         xmm1,       xmm3
-
-        movq            xmm2,       xmm0
-        movq            xmm3,       xmm1
-
-        psrldq          xmm0,       8
-        psrldq          xmm1,       8
-
-        paddd           xmm0,       xmm1
-        paddd           xmm2,       xmm3
-
-        paddd           xmm0,       xmm2
-        movq            xmm1,       xmm0
-
-        psrldq          xmm0,       4
-        paddd           xmm1,       xmm0
-
-        movq            rcx,        xmm1
-        and             rcx,        0xffff
-
-        xor             rdx,        rdx
-        sub             rdx,        rcx
-
-        bsr             rax,        rcx
-        inc             rax
-
-        sar             rdx,        31
-        and             rax,        rdx
-
-
-    ; begin epilog
-    pop rdi
-    pop rsi
-    UNSHADOW_ARGS
-    pop         rbp
-    ret
diff --git a/vp8/encoder/x86/quantize_sse2.asm b/vp8/encoder/x86/quantize_sse2.asm
index a1b1c40cb..324881337 100644
--- a/vp8/encoder/x86/quantize_sse2.asm
+++ b/vp8/encoder/x86/quantize_sse2.asm
@@ -252,3 +252,137 @@ rq_zigzag_1c:
     UNSHADOW_ARGS
     pop         rbp
     ret
+
+
+;int vp8_fast_quantize_b_impl_sse2(short *coeff_ptr,
+;                           short *qcoeff_ptr,short *dequant_ptr,
+;                           short *scan_mask, short *round_ptr,
+;                           short *quant_ptr, short *dqcoeff_ptr);
+global sym(vp8_fast_quantize_b_impl_ssse2)
+sym(vp8_fast_quantize_b_impl_ssse2):
+    push        rbp
+    mov         rbp, rsp
+    SHADOW_ARGS_TO_STACK 7
+    push        rsi
+    push        rdi
+    push        rbx
+    ; end prolog
+
+    ALIGN_STACK 16, rax
+
+    %define save_xmm6  0
+    %define save_xmm7 16
+
+    %define vp8_fastquantizeb_stack_size save_xmm7 + 16
+
+    sub         rsp, vp8_fastquantizeb_stack_size
+
+    movdqa      XMMWORD PTR[rsp + save_xmm6], xmm6
+    movdqa      XMMWORD PTR[rsp + save_xmm7], xmm7
+
+    mov         rdx, arg(0)                 ;coeff_ptr
+    mov         rcx, arg(2)                 ;dequant_ptr
+    mov         rax, arg(3)                 ;scan_mask
+    mov         rdi, arg(4)                 ;round_ptr
+    mov         rsi, arg(5)                 ;quant_ptr
+
+    movdqa      xmm0, XMMWORD PTR[rdx]
+    movdqa      xmm4, XMMWORD PTR[rdx + 16]
+
+    movdqa      xmm6, XMMWORD PTR[rdi]      ;round lo
+    movdqa      xmm7, XMMWORD PTR[rdi + 16] ;round hi
+
+    movdqa      xmm1, xmm0
+    movdqa      xmm5, xmm4
+
+    psraw       xmm0, 15                    ;sign of z (aka sz)
+    psraw       xmm4, 15                    ;sign of z (aka sz)
+
+    pxor        xmm1, xmm0
+    pxor        xmm5, xmm4
+    psubw       xmm1, xmm0                  ;x = abs(z)
+    psubw       xmm5, xmm4                  ;x = abs(z)
+
+    paddw       xmm1, xmm6
+    paddw       xmm5, xmm7
+
+    pmulhw      xmm1, XMMWORD PTR[rsi]
+    pmulhw      xmm5, XMMWORD PTR[rsi + 16]
+
+    mov         rdi, arg(1)                 ;qcoeff_ptr
+    mov         rsi, arg(6)                 ;dqcoeff_ptr
+
+    movdqa      xmm6, XMMWORD PTR[rcx]
+    movdqa      xmm7, XMMWORD PTR[rcx + 16]
+
+    pxor        xmm1, xmm0
+    pxor        xmm5, xmm4
+    psubw       xmm1, xmm0
+    psubw       xmm5, xmm4
+
+    movdqa      XMMWORD PTR[rdi], xmm1
+    movdqa      XMMWORD PTR[rdi + 16], xmm5
+
+    pmullw      xmm6, xmm1
+    pmullw      xmm7, xmm5
+
+    movdqa      xmm2, XMMWORD PTR[rax]
+    movdqa      xmm3, XMMWORD PTR[rax+16];
+
+    pxor        xmm4, xmm4            ;clear all bits
+    pcmpeqw     xmm1, xmm4
+    pcmpeqw     xmm5, xmm4
+
+    pcmpeqw     xmm4, xmm4            ;set all bits
+    pxor        xmm1, xmm4
+    pxor        xmm5, xmm4
+
+    psrlw       xmm1, 15
+    psrlw       xmm5, 15
+
+    pmaddwd     xmm1, xmm2
+    pmaddwd     xmm5, xmm3
+
+    movq        xmm2, xmm1
+    movq        xmm3, xmm5
+
+    psrldq      xmm1, 8
+    psrldq      xmm5, 8
+
+    paddd       xmm1, xmm5
+    paddd       xmm2, xmm3
+
+    paddd       xmm1, xmm2
+    movq        xmm5, xmm1
+
+    psrldq      xmm1, 4
+    paddd       xmm5, xmm1
+
+    movq        rcx,  xmm5
+    and         rcx,  0xffff
+
+    xor         rdx,  rdx
+    sub         rdx,  rcx
+
+    bsr         rax,  rcx
+    inc         rax
+
+    sar         rdx,  31
+    and         rax,  rdx
+
+    movdqa      XMMWORD PTR[rsi], xmm6        ;store dqcoeff
+    movdqa      XMMWORD PTR[rsi + 16], xmm7   ;store dqcoeff
+
+    movdqa      xmm6, XMMWORD PTR[rsp + save_xmm6]
+    movdqa      xmm7, XMMWORD PTR[rsp + save_xmm7]
+
+    add         rsp, vp8_fastquantizeb_stack_size
+    pop         rsp
+
+    ; begin epilog
+    pop         rbx
+    pop         rdi
+    pop         rsi
+    UNSHADOW_ARGS
+    pop         rbp
+    ret
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index 18dc49cd4..7490a8add 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -88,24 +88,22 @@ void vp8_short_fdct8x4_sse2(short *input, short *output, int pitch)
     vp8_short_fdct4x4_sse2(input + 4, output + 16, pitch);
 }
 
-int vp8_fast_quantize_b_impl_sse(short *coeff_ptr, short *zbin_ptr,
+int vp8_fast_quantize_b_impl_sse2(short *coeff_ptr,
                                  short *qcoeff_ptr, short *dequant_ptr,
                                  short *scan_mask, short *round_ptr,
                                  short *quant_ptr, short *dqcoeff_ptr);
-void vp8_fast_quantize_b_sse(BLOCK *b, BLOCKD *d)
+void vp8_fast_quantize_b_sse2(BLOCK *b, BLOCKD *d)
 {
     short *scan_mask    = vp8_default_zig_zag_mask;//d->scan_order_mask_ptr;
     short *coeff_ptr  = &b->coeff[0];
-    short *zbin_ptr   = &b->zbin[0][0];
     short *round_ptr  = &b->round[0][0];
     short *quant_ptr  = &b->quant[0][0];
     short *qcoeff_ptr = d->qcoeff;
     short *dqcoeff_ptr = d->dqcoeff;
     short *dequant_ptr = &d->dequant[0][0];
 
-    d->eob = vp8_fast_quantize_b_impl_sse(
+    d->eob = vp8_fast_quantize_b_impl_ssse2(
                  coeff_ptr,
-                 zbin_ptr,
                  qcoeff_ptr,
                  dequant_ptr,
                  scan_mask,
@@ -116,6 +114,7 @@ void vp8_fast_quantize_b_sse(BLOCK *b, BLOCKD *d)
              );
 }
 
+
 int vp8_regular_quantize_b_impl_sse2(short *coeff_ptr, short *zbin_ptr,
                                short *qcoeff_ptr,short *dequant_ptr,
                                const int *default_zig_zag, short *round_ptr,
@@ -285,8 +284,10 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
         cpi->rtcd.encodemb.mbuverr               = vp8_mbuverror_xmm;
         /* cpi->rtcd.encodemb.sub* not implemented for wmt */
 
-        /*cpi->rtcd.quantize.fastquantb            = vp8_fast_quantize_b_sse;
-        cpi->rtcd.quantize.quantb            = vp8_regular_quantize_b_sse2;*/
+        /*cpi->rtcd.quantize.quantb            = vp8_regular_quantize_b_sse2;*/
+
+        cpi->rtcd.quantize.fastquantb            = vp8_fast_quantize_b_sse2;
+
     }
 
 #endif