diff --git a/vp8/encoder/x86/quantize_sse4.asm b/vp8/encoder/x86/quantize_sse4.asm
new file mode 100644
index 0000000000000000000000000000000000000000..258899eedd988becdf5e99a6e24a91f2df143601
--- /dev/null
+++ b/vp8/encoder/x86/quantize_sse4.asm
@@ -0,0 +1,254 @@
+;
+; 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"
+%include "asm_enc_offsets.asm"
+
+
+; void vp8_regular_quantize_b_sse4 | arg
+; (BLOCK *b, | 0
+; BLOCKD *d) | 1
+
+global sym(vp8_regular_quantize_b_sse4)
+sym(vp8_regular_quantize_b_sse4):
+
+%if ABI_IS_32BIT
+ push rbp
+ mov rbp, rsp
+ GET_GOT rbx
+ push rdi
+ push rsi
+
+ ALIGN_STACK 16, rax
+ %define qcoeff 0 ; 32
+ %define stack_size 32
+ sub rsp, stack_size
+%else
+ %ifidn __OUTPUT_FORMAT__,x64
+ SAVE_XMM 8, u
+ push rdi
+ push rsi
+ %endif
+%endif
+ ; end prolog
+
+%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_zbin]
+ mov rdx, [rdi + vp8_block_round]
+ movd xmm7, [rdi + vp8_block_zbin_extra]
+
+ ; z
+ movdqa xmm0, [rax]
+ movdqa xmm1, [rax + 16]
+
+ ; duplicate zbin_oq_value
+ pshuflw xmm7, xmm7, 0
+ punpcklwd xmm7, xmm7
+
+ movdqa xmm2, xmm0
+ movdqa xmm3, xmm1
+
+ ; sz
+ psraw xmm0, 15
+ psraw xmm1, 15
+
+ ; (z ^ sz)
+ pxor xmm2, xmm0
+ pxor xmm3, xmm1
+
+ ; x = abs(z)
+ psubw xmm2, xmm0
+ psubw xmm3, xmm1
+
+ ; zbin
+ movdqa xmm4, [rcx]
+ movdqa xmm5, [rcx + 16]
+
+ ; *zbin_ptr + zbin_oq_value
+ paddw xmm4, xmm7
+ paddw xmm5, xmm7
+
+ movdqa xmm6, xmm2
+ movdqa xmm7, xmm3
+
+ ; x - (*zbin_ptr + zbin_oq_value)
+ psubw xmm6, xmm4
+ psubw xmm7, xmm5
+
+ ; round
+ movdqa xmm4, [rdx]
+ movdqa xmm5, [rdx + 16]
+
+ mov rax, [rdi + vp8_block_quant_shift]
+ mov rcx, [rdi + vp8_block_quant]
+ mov rdx, [rdi + vp8_block_zrun_zbin_boost]
+
+ ; x + round
+ paddw xmm2, xmm4
+ paddw xmm3, xmm5
+
+ ; quant
+ movdqa xmm4, [rcx]
+ movdqa xmm5, [rcx + 16]
+
+ ; y = x * quant_ptr >> 16
+ pmulhw xmm4, xmm2
+ pmulhw xmm5, xmm3
+
+ ; y += x
+ paddw xmm2, xmm4
+ paddw xmm3, xmm5
+
+ pxor xmm4, xmm4
+%if ABI_IS_32BIT
+ movdqa [rsp + qcoeff], xmm4
+ movdqa [rsp + qcoeff + 16], xmm4
+%else
+ pxor xmm8, xmm8
+%endif
+
+ ; quant_shift
+ movdqa xmm5, [rax]
+
+ ; zrun_zbin_boost
+ mov rax, rdx
+
+%macro ZIGZAG_LOOP 5
+ ; x
+ pextrw ecx, %4, %2
+
+ ; if (x >= zbin)
+ sub cx, WORD PTR[rdx] ; x - zbin
+ lea rdx, [rdx + 2] ; zbin_boost_ptr++
+ jl rq_zigzag_loop_%1 ; x < zbin
+
+ pextrw edi, %3, %2 ; y
+
+ ; downshift by quant_shift[rc]
+ pextrb ecx, xmm5, %1 ; quant_shift[rc]
+ sar edi, cl ; also sets Z bit
+ je rq_zigzag_loop_%1 ; !y
+%if ABI_IS_32BIT
+ mov WORD PTR[rsp + qcoeff + %1 *2], di
+%else
+ pinsrw %5, edi, %2 ; qcoeff[rc]
+%endif
+ mov rdx, rax ; reset to b->zrun_zbin_boost
+rq_zigzag_loop_%1:
+%endmacro
+; in vp8_default_zig_zag1d order: see vp8/common/entropy.c
+ZIGZAG_LOOP 0, 0, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 1, 1, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 4, 4, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 8, 0, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 5, 5, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 2, 2, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 3, 3, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 6, 6, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 9, 1, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 12, 4, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 13, 5, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 10, 2, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 7, 7, xmm2, xmm6, xmm4
+ZIGZAG_LOOP 11, 3, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 14, 6, xmm3, xmm7, xmm8
+ZIGZAG_LOOP 15, 7, xmm3, xmm7, xmm8
+
+ mov rcx, [rsi + vp8_blockd_dequant]
+ mov rdi, [rsi + vp8_blockd_dqcoeff]
+
+%if ABI_IS_32BIT
+ movdqa xmm4, [rsp + qcoeff]
+ movdqa xmm5, [rsp + qcoeff + 16]
+%else
+ %define xmm5 xmm8
+%endif
+
+ ; y ^ sz
+ pxor xmm4, xmm0
+ pxor xmm5, xmm1
+ ; x = (y ^ sz) - sz
+ psubw xmm4, xmm0
+ psubw xmm5, xmm1
+
+ ; dequant
+ movdqa xmm0, [rcx]
+ movdqa xmm1, [rcx + 16]
+
+ mov rcx, [rsi + vp8_blockd_qcoeff]
+
+ pmullw xmm0, xmm4
+ pmullw xmm1, xmm5
+
+ ; store qcoeff
+ movdqa [rcx], xmm4
+ movdqa [rcx + 16], xmm5
+
+ ; store dqcoeff
+ movdqa [rdi], xmm0
+ movdqa [rdi + 16], xmm1
+
+ ; select the last value (in zig_zag order) for EOB
+ pxor xmm6, xmm6
+ pcmpeqw xmm4, xmm6
+ pcmpeqw xmm5, xmm6
+
+ packsswb xmm4, xmm5
+ pshufb xmm4, [GLOBAL(zig_zag1d)]
+ pmovmskb edx, xmm4
+ xor rdi, rdi
+ mov eax, -1
+ xor dx, ax
+ bsr eax, edx
+ sub edi, edx
+ sar edi, 31
+ add eax, 1
+ and eax, edi
+
+ mov [rsi + vp8_blockd_eob], eax
+
+ ; begin epilog
+%if ABI_IS_32BIT
+ add rsp, stack_size
+ pop rsp
+
+ pop rsi
+ pop rdi
+ RESTORE_GOT
+ pop rbp
+%else
+ %undef xmm5
+ %ifidn __OUTPUT_FORMAT__,x64
+ pop rsi
+ pop rdi
+ RESTORE_XMM
+ %endif
+%endif
+
+ ret
+
+SECTION_RODATA
+align 16
+; vp8/common/entropy.c: vp8_default_zig_zag1d
+zig_zag1d:
+ db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15
diff --git a/vp8/encoder/x86/quantize_x86.h b/vp8/encoder/x86/quantize_x86.h
index f093580618ed7078e3cec45f0aebf6176cbfbbde..bbe475f8cc79065cc19566051bca5e2f67ee6be8 100644
--- a/vp8/encoder/x86/quantize_x86.h
+++ b/vp8/encoder/x86/quantize_x86.h
@@ -51,4 +51,17 @@ extern prototype_quantize_block(vp8_fast_quantize_b_ssse3);
#endif /* HAVE_SSSE3 */
+
+#if HAVE_SSE4_1
+extern prototype_quantize_block(vp8_regular_quantize_b_sse4);
+
+#if !CONFIG_RUNTIME_CPU_DETECT
+
+#undef vp8_quantize_quantb
+#define vp8_quantize_quantb vp8_regular_quantize_b_sse4
+
+#endif /* !CONFIG_RUNTIME_CPU_DETECT */
+
+#endif /* HAVE_SSE4_1 */
+
#endif /* QUANTIZE_X86_H */
diff --git a/vp8/encoder/x86/x86_csystemdependent.c b/vp8/encoder/x86/x86_csystemdependent.c
index 47a1788bc8bf64119d629812c3addf3411a23451..b01319fa4535e6dc3be39a162e593111dfe1da8e 100644
--- a/vp8/encoder/x86/x86_csystemdependent.c
+++ b/vp8/encoder/x86/x86_csystemdependent.c
@@ -313,6 +313,8 @@ void vp8_arch_x86_encoder_init(VP8_COMP *cpi)
cpi->rtcd.variance.sad8x8x8 = vp8_sad8x8x8_sse4;
cpi->rtcd.variance.sad4x4x8 = vp8_sad4x4x8_sse4;
cpi->rtcd.search.full_search = vp8_full_search_sadx8;
+
+ cpi->rtcd.quantize.quantb = vp8_regular_quantize_b_sse4;
}
#endif
diff --git a/vp8/vp8cx.mk b/vp8/vp8cx.mk
index a7b68dba7578d5d941b3b52f322f5e455a3308d9..c1783716479b9ec095599342dc7ba3e6a30a388f 100644
--- a/vp8/vp8cx.mk
+++ b/vp8/vp8cx.mk
@@ -117,6 +117,7 @@ VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/variance_ssse3.c
VP8_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/variance_impl_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-$(HAVE_SSE4_1) += encoder/x86/quantize_sse4.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/quantize_mmx.asm
VP8_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/encodeopt.asm
VP8_CX_SRCS-$(ARCH_X86_64) += encoder/x86/ssim_opt.asm