use asm_offsets with vp8_fast_quantize_b_sse3

on the same order as the sse2 fast quantize change: ~2%
except for 32bit. only a slight improvment there.

Change-Id: Iff80e5f1ce7e646eebfdc8871405458ff911986b

3 files changed, 82 insertions, 61 deletions

diff --git a/vp8/encoder/x86/quantize_ssse3.asm b/vp8/encoder/x86/quantize_ssse3.asm
index 2f33199e5..912007e02 100644
--- a/vp8/encoder/x86/quantize_ssse3.asm
+++ b/vp8/encoder/x86/quantize_ssse3.asm
@@ -9,38 +9,62 @@
 
 
 %include "vpx_ports/x86_abi_support.asm"
+%include "asm_enc_offsets.asm"
 
 
-;int vp8_fast_quantize_b_impl_ssse3(short *coeff_ptr
-;               short *qcoeff_ptr,short *dequant_ptr,
-;               short *round_ptr,
-;               short *quant_ptr, short *dqcoeff_ptr);
+; void vp8_fast_quantize_b_ssse3 | arg
+;  (BLOCK  *b,                   |  0
+;   BLOCKD *d)                   |  1
 ;
-global sym(vp8_fast_quantize_b_impl_ssse3)
-sym(vp8_fast_quantize_b_impl_ssse3):
+
+global sym(vp8_fast_quantize_b_ssse3)
+sym(vp8_fast_quantize_b_ssse3):
     push        rbp
     mov         rbp, rsp
-    SHADOW_ARGS_TO_STACK 6
     GET_GOT     rbx
+
+%if ABI_IS_32BIT
+    push        rdi
     push        rsi
+%else
+  %ifidn __OUTPUT_FORMAT__,x64
     push        rdi
+    push        rsi
+  %endif
+%endif
     ; end prolog
 
-    mov         rdx, arg(0)                 ;coeff_ptr
-    mov         rdi, arg(3)                 ;round_ptr
-    mov         rsi, arg(4)                 ;quant_ptr
-
-    movdqa      xmm0, [rdx]
-    movdqa      xmm4, [rdx + 16]
-
-    movdqa      xmm2, [rdi]                 ;round lo
-    movdqa      xmm3, [rdi + 16]            ;round hi
+%if ABI_IS_32BIT
+    mov         rdi, arg(0)                 ; BLOCK *b
+    mov         rsi, arg(1)                 ; BLOCKD *d
+%else
+  %ifidn __OUTPUT_FORMAT__,x64
+    mov         rdi, rcx                    ; BLOCK *b
+    mov         rsi, rdx                    ; BLOCKD *d
+  %else
+    ;mov         rdi, rdi                    ; BLOCK *b
+    ;mov         rsi, rsi                    ; BLOCKD *d
+  %endif
+%endif
+
+    mov         rax, [rdi + vp8_block_coeff]
+    mov         rcx, [rdi + vp8_block_round]
+    mov         rdx, [rdi + vp8_block_quant_fast]
+
+    ; coeff
+    movdqa      xmm0, [rax]
+    movdqa      xmm4, [rax + 16]
+
+    ; round
+    movdqa      xmm2, [rcx]
+    movdqa      xmm3, [rcx + 16]
 
     movdqa      xmm1, xmm0
     movdqa      xmm5, xmm4
 
-    psraw       xmm0, 15                    ;sign of z (aka sz)
-    psraw       xmm4, 15                    ;sign of z (aka sz)
+    ; sz = z >> 15
+    psraw       xmm0, 15
+    psraw       xmm4, 15
 
     pabsw       xmm1, xmm1
     pabsw       xmm5, xmm5
@@ -48,23 +72,24 @@ sym(vp8_fast_quantize_b_impl_ssse3):
     paddw       xmm1, xmm2
     paddw       xmm5, xmm3
 
-    pmulhw      xmm1, [rsi]
-    pmulhw      xmm5, [rsi + 16]
+    ; quant_fast
+    pmulhw      xmm1, [rdx]
+    pmulhw      xmm5, [rdx + 16]
 
-    mov         rdi, arg(1)                 ;qcoeff_ptr
-    mov         rcx, arg(2)                 ;dequant_ptr
-    mov         rsi, arg(5)                 ;dqcoeff_ptr
+    mov         rax, [rsi + vp8_blockd_qcoeff]
+    mov         rdi, [rsi + vp8_blockd_dequant]
+    mov         rcx, [rsi + vp8_blockd_dqcoeff]
 
     pxor        xmm1, xmm0
     pxor        xmm5, xmm4
     psubw       xmm1, xmm0
     psubw       xmm5, xmm4
 
-    movdqa      [rdi], xmm1
-    movdqa      [rdi + 16], xmm5
+    movdqa      [rax], xmm1
+    movdqa      [rax + 16], xmm5
 
-    movdqa      xmm2, [rcx]
-    movdqa      xmm3, [rcx + 16]
+    movdqa      xmm2, [rdi]
+    movdqa      xmm3, [rdi + 16]
 
     pxor        xmm4, xmm4
     pmullw      xmm2, xmm1
@@ -73,38 +98,37 @@ sym(vp8_fast_quantize_b_impl_ssse3):
     pcmpeqw     xmm1, xmm4                  ;non zero mask
     pcmpeqw     xmm5, xmm4                  ;non zero mask
     packsswb    xmm1, xmm5
-    pshufb      xmm1, [ GLOBAL(zz_shuf)]
+    pshufb      xmm1, [GLOBAL(zz_shuf)]
 
     pmovmskb    edx, xmm1
 
-;    xor         ecx, ecx
-;    mov         eax, -1
-;find_eob_loop:
-;    shr         edx, 1
-;    jc          fq_skip
-;    mov         eax, ecx
-;fq_skip:
-;    inc         ecx
-;    cmp         ecx, 16
-;    jne         find_eob_loop
     xor         rdi, rdi
     mov         eax, -1
     xor         dx, ax                      ;flip the bits for bsr
     bsr         eax, edx
 
-    movdqa      [rsi], xmm2                 ;store dqcoeff
-    movdqa      [rsi + 16], xmm3            ;store dqcoeff
+    movdqa      [rcx], xmm2                 ;store dqcoeff
+    movdqa      [rcx + 16], xmm3            ;store dqcoeff
 
     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
+
     ; begin epilog
+%if ABI_IS_32BIT
+    pop         rsi
     pop         rdi
+%else
+  %ifidn __OUTPUT_FORMAT__,x64
     pop         rsi
+    pop         rdi
+  %endif
+%endif
+
     RESTORE_GOT
-    UNSHADOW_ARGS
     pop         rbp
     ret
 
diff --git a/vp8/encoder/x86/quantize_x86.h b/vp8/encoder/x86/quantize_x86.h
index df2e0bc39..f09358061 100644
--- a/vp8/encoder/x86/quantize_x86.h
+++ b/vp8/encoder/x86/quantize_x86.h
@@ -19,7 +19,7 @@
  */
 #if HAVE_MMX
 
-#endif
+#endif /* HAVE_MMX */
 
 
 #if HAVE_SSE2
@@ -34,9 +34,21 @@ extern prototype_quantize_block(vp8_fast_quantize_b_sse2);
 #undef vp8_quantize_fastquantb
 #define vp8_quantize_fastquantb vp8_fast_quantize_b_sse2
 
-#endif
+#endif /* !CONFIG_RUNTIME_CPU_DETECT */
 
-#endif
+#endif /* HAVE_SSE2 */
 
 
-#endif
+#if HAVE_SSSE3
+extern prototype_quantize_block(vp8_fast_quantize_b_ssse3);
+
+#if !CONFIG_RUNTIME_CPU_DETECT
+
+#undef vp8_quantize_fastquantb
+#define vp8_quantize_fastquantb vp8_fast_quantize_b_ssse3
+
+#endif /* !CONFIG_RUNTIME_CPU_DETECT */
+
+#endif /* HAVE_SSSE3 */
+
+#endif /* QUANTIZE_X86_H */
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index 8bceecec4..8f2774b7a 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -112,21 +112,6 @@ static void subtract_b_sse2(BLOCK *be, BLOCKD *bd, int pitch)
 #endif
 
 #if HAVE_SSSE3
-int vp8_fast_quantize_b_impl_ssse3(short *coeff_ptr,
-                                 short *qcoeff_ptr, short *dequant_ptr,
-                                 short *round_ptr,
-                                 short *quant_ptr, short *dqcoeff_ptr);
-static void fast_quantize_b_ssse3(BLOCK *b, BLOCKD *d)
-{
-    d->eob = vp8_fast_quantize_b_impl_ssse3(
-                    b->coeff,
-                    d->qcoeff,
-                    d->dequant,
-                    b->round,
-                    b->quant_fast,
-                    d->dqcoeff
-               );
-}
 #if CONFIG_PSNR
 #if ARCH_X86_64
 typedef void ssimpf
@@ -307,7 +292,7 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
         cpi->rtcd.variance.subpixvar16x8         = vp8_sub_pixel_variance16x8_ssse3;
         cpi->rtcd.variance.subpixvar16x16        = vp8_sub_pixel_variance16x16_ssse3;
 
-        cpi->rtcd.quantize.fastquantb            = fast_quantize_b_ssse3;
+        cpi->rtcd.quantize.fastquantb            = vp8_fast_quantize_b_ssse3;
 
 #if CONFIG_PSNR
 #if ARCH_X86_64