Commit ff4a71f4 authored by Scott LaVarnway's avatar Scott LaVarnway
Browse files

SSSE3 version of fast quantizer

(test clip: tulip)
For good quality mode with speed=1, this gave the encoder
a small (2 - 3%) performance boost.

Change-Id: I8a1d4269465944ac0819986c2f0be4b0a2ee0b35
parent dcee88ea
;
; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
;
; Use of this source code is governed by a BSD-style license and patent
; grant that can be found in the LICENSE file in the root of the source
; tree. All contributing project authors may be found in the AUTHORS
; file in the root of the source tree.
;
%include "vpx_ports/x86_abi_support.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);
;
global sym(vp8_fast_quantize_b_impl_ssse3)
sym(vp8_fast_quantize_b_impl_ssse3):
push rbp
mov rbp, rsp
SHADOW_ARGS_TO_STACK 6
GET_GOT rbx
push rsi
push rdi
; 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
movdqa xmm1, xmm0
movdqa xmm5, xmm4
psraw xmm0, 15 ;sign of z (aka sz)
psraw xmm4, 15 ;sign of z (aka sz)
pabsw xmm1, xmm1
pabsw xmm5, xmm5
paddw xmm1, xmm2
paddw xmm5, xmm3
pmulhw xmm1, [rsi]
pmulhw xmm5, [rsi + 16]
mov rdi, arg(1) ;qcoeff_ptr
mov rcx, arg(2) ;dequant_ptr
mov rsi, arg(5) ;dqcoeff_ptr
pxor xmm1, xmm0
pxor xmm5, xmm4
psubw xmm1, xmm0
psubw xmm5, xmm4
movdqa [rdi], xmm1
movdqa [rdi + 16], xmm5
movdqa xmm2, [rcx]
movdqa xmm3, [rcx + 16]
pxor xmm4, xmm4
pmullw xmm2, xmm1
pmullw xmm3, xmm5
pcmpeqw xmm1, xmm4 ;non zero mask
pcmpeqw xmm5, xmm4 ;non zero mask
packsswb xmm1, xmm5
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
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
; begin epilog
pop rdi
pop rsi
RESTORE_GOT
UNSHADOW_ARGS
pop rbp
ret
SECTION_RODATA
align 16
zz_shuf:
db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15
......@@ -179,6 +179,25 @@ void vp8_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);
void vp8_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,
d->dqcoeff
);
}
#endif
void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
{
#if CONFIG_RUNTIME_CPU_DETECT
......@@ -333,6 +352,9 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
{
cpi->rtcd.variance.sad16x16x3 = vp8_sad16x16x3_ssse3;
cpi->rtcd.variance.sad16x8x3 = vp8_sad16x8x3_ssse3;
cpi->rtcd.quantize.fastquantb = vp8_fast_quantize_b_ssse3;
}
#endif
......
......@@ -109,6 +109,7 @@ VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/quantize_sse2.asm
VP8_CX_SRCS-$(HAVE_SSE2) += encoder/x86/subtract_sse2.asm
VP8_CX_SRCS-$(HAVE_SSE3) += encoder/x86/sad_sse3.asm
VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/sad_ssse3.asm
VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/quantize_ssse3.asm
VP8_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/sad_sse4.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/encodeopt.asm
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment