diff --git a/vp9/encoder/vp9_quantize.h b/vp9/encoder/vp9_quantize.h index fd7a4bb4f..fb74cbd29 100644 --- a/vp9/encoder/vp9_quantize.h +++ b/vp9/encoder/vp9_quantize.h @@ -22,10 +22,6 @@ #define prototype_quantize_mb(sym) \ void (sym)(MACROBLOCK *x) -#if ARCH_X86 || ARCH_X86_64 -#include "x86/vp9_quantize_x86.h" -#endif - void vp9_regular_quantize_b_4x4_pair(MACROBLOCK *mb, int b_idx1, int b_idx2, int y_blocks); void vp9_regular_quantize_b_4x4(MACROBLOCK *mb, int b_idx, TX_TYPE tx_type, diff --git a/vp9/encoder/x86/vp9_quantize_mmx.asm b/vp9/encoder/x86/vp9_quantize_mmx.asm deleted file mode 100644 index 22e235610..000000000 --- a/vp9/encoder/x86/vp9_quantize_mmx.asm +++ /dev/null @@ -1,286 +0,0 @@ -; -; Copyright (c) 2010 The WebM project authors. All Rights Reserved. -; -; Use of this source code is governed by a BSD-style license -; that can be found in the LICENSE file in the root of the source -; tree. An additional intellectual property rights grant can be found -; in the file PATENTS. 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 vp9_fast_quantize_b_impl_mmx(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(vp9_fast_quantize_b_impl_mmx) PRIVATE -sym(vp9_fast_quantize_b_impl_mmx): - push rbp - mov rbp, rsp - SHADOW_ARGS_TO_STACK 8 - push rsi - push rdi - ; end prolog - - - mov rsi, arg(0) ;coeff_ptr - movq mm0, [rsi] - - mov rax, arg(1) ;zbin_ptr - movq mm1, [rax] - - movq mm3, mm0 - psraw mm0, 15 - - pxor mm3, mm0 - psubw mm3, mm0 ; abs - - movq mm2, mm3 - pcmpgtw mm1, mm2 - - pandn mm1, mm2 - movq mm3, mm1 - - mov rdx, arg(6) ;quant_ptr - movq mm1, [rdx] - - mov rcx, arg(5) ;round_ptr - movq mm2, [rcx] - - paddw mm3, mm2 - pmulhuw mm3, mm1 - - pxor mm3, mm0 - psubw mm3, mm0 ;gain the sign back - - mov rdi, arg(2) ;qcoeff_ptr - movq mm0, mm3 - - movq [rdi], mm3 - - mov rax, arg(3) ;dequant_ptr - movq mm2, [rax] - - pmullw mm3, mm2 - mov rax, arg(7) ;dqcoeff_ptr - - movq [rax], mm3 - - ; next 8 - movq mm4, [rsi+8] - - mov rax, arg(1) ;zbin_ptr - movq mm5, [rax+8] - - movq mm7, mm4 - psraw mm4, 15 - - pxor mm7, mm4 - psubw mm7, mm4 ; abs - - movq mm6, mm7 - pcmpgtw mm5, mm6 - - pandn mm5, mm6 - movq mm7, mm5 - - movq mm5, [rdx+8] - movq mm6, [rcx+8] - - paddw mm7, mm6 - pmulhuw mm7, mm5 - - pxor mm7, mm4 - psubw mm7, mm4;gain the sign back - - mov rdi, arg(2) ;qcoeff_ptr - - movq mm1, mm7 - movq [rdi+8], mm7 - - mov rax, arg(3) ;dequant_ptr - movq mm6, [rax+8] - - pmullw mm7, mm6 - mov rax, arg(7) ;dqcoeff_ptr - - movq [rax+8], mm7 - - - ; next 8 - movq mm4, [rsi+16] - - mov rax, arg(1) ;zbin_ptr - movq mm5, [rax+16] - - movq mm7, mm4 - psraw mm4, 15 - - pxor mm7, mm4 - psubw mm7, mm4 ; abs - - movq mm6, mm7 - pcmpgtw mm5, mm6 - - pandn mm5, mm6 - movq mm7, mm5 - - movq mm5, [rdx+16] - movq mm6, [rcx+16] - - paddw mm7, mm6 - pmulhuw mm7, mm5 - - pxor mm7, mm4 - psubw mm7, mm4;gain the sign back - - mov rdi, arg(2) ;qcoeff_ptr - - movq mm1, mm7 - movq [rdi+16], mm7 - - mov rax, arg(3) ;dequant_ptr - movq mm6, [rax+16] - - pmullw mm7, mm6 - mov rax, arg(7) ;dqcoeff_ptr - - movq [rax+16], mm7 - - - ; next 8 - movq mm4, [rsi+24] - - mov rax, arg(1) ;zbin_ptr - movq mm5, [rax+24] - - movq mm7, mm4 - psraw mm4, 15 - - pxor mm7, mm4 - psubw mm7, mm4 ; abs - - movq mm6, mm7 - pcmpgtw mm5, mm6 - - pandn mm5, mm6 - movq mm7, mm5 - - movq mm5, [rdx+24] - movq mm6, [rcx+24] - - paddw mm7, mm6 - pmulhuw mm7, mm5 - - pxor mm7, mm4 - psubw mm7, mm4;gain the sign back - - mov rdi, arg(2) ;qcoeff_ptr - - movq mm1, mm7 - movq [rdi+24], mm7 - - mov rax, arg(3) ;dequant_ptr - movq mm6, [rax+24] - - pmullw mm7, mm6 - mov rax, arg(7) ;dqcoeff_ptr - - movq [rax+24], mm7 - - - - mov rdi, arg(4) ;scan_mask - mov rsi, arg(2) ;qcoeff_ptr - - pxor mm5, mm5 - pxor mm7, mm7 - - movq mm0, [rsi] - movq mm1, [rsi+8] - - movq mm2, [rdi] - movq mm3, [rdi+8]; - - pcmpeqw mm0, mm7 - pcmpeqw mm1, mm7 - - pcmpeqw mm6, mm6 - pxor mm0, mm6 - - pxor mm1, mm6 - psrlw mm0, 15 - - psrlw mm1, 15 - pmaddwd mm0, mm2 - - pmaddwd mm1, mm3 - movq mm5, mm0 - - paddd mm5, mm1 - - movq mm0, [rsi+16] - movq mm1, [rsi+24] - - movq mm2, [rdi+16] - movq mm3, [rdi+24]; - - pcmpeqw mm0, mm7 - pcmpeqw mm1, mm7 - - pcmpeqw mm6, mm6 - pxor mm0, mm6 - - pxor mm1, mm6 - psrlw mm0, 15 - - psrlw mm1, 15 - pmaddwd mm0, mm2 - - pmaddwd mm1, mm3 - paddd mm5, mm0 - - paddd mm5, mm1 - movq mm0, mm5 - - psrlq mm5, 32 - paddd mm0, mm5 - - ; eob adjustment begins here - movq rcx, mm0 - and rcx, 0xffff - - xor rdx, rdx - sub rdx, rcx ; rdx=-rcx - - bsr rax, rcx - inc rax - - sar rdx, 31 - and rax, rdx - ; Substitute the sse assembly for the old mmx mixed assembly/C. The - ; following is kept as reference - ; movq rcx, mm0 - ; bsr rax, rcx - ; - ; mov eob, rax - ; mov eee, rcx - ; - ;if(eee==0) - ;{ - ; eob=-1; - ;} - ;else if(eee<0) - ;{ - ; eob=15; - ;} - ;d->eob = eob+1; - - ; begin epilog - pop rdi - pop rsi - UNSHADOW_ARGS - pop rbp - ret diff --git a/vp9/encoder/x86/vp9_quantize_sse2.asm b/vp9/encoder/x86/vp9_quantize_sse2.asm deleted file mode 100644 index 700e64b1f..000000000 --- a/vp9/encoder/x86/vp9_quantize_sse2.asm +++ /dev/null @@ -1,379 +0,0 @@ -; -; 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" - - -; void vp9_regular_quantize_b_sse2 | arg -; (BLOCK *b, | 0 -; BLOCKD *d) | 1 - -global sym(vp9_regular_quantize_b_sse2) PRIVATE -sym(vp9_regular_quantize_b_sse2): - push rbp - mov rbp, rsp - SAVE_XMM 7 - GET_GOT rbx - -%if ABI_IS_32BIT - push rdi - push rsi -%else - %if LIBVPX_YASM_WIN64 - push rdi - push rsi - %endif -%endif - - ALIGN_STACK 16, rax - %define zrun_zbin_boost 0 ; 8 - %define abs_minus_zbin 8 ; 32 - %define temp_qcoeff 40 ; 32 - %define qcoeff 72 ; 32 - %define stack_size 104 - sub rsp, stack_size - ; end prolog - -%if ABI_IS_32BIT - mov rdi, arg(0) ; BLOCK *b - mov rsi, arg(1) ; BLOCKD *d -%else - %if LIBVPX_YASM_WIN64 - mov rdi, rcx ; BLOCK *b - mov rsi, rdx ; BLOCKD *d - %else - ;mov rdi, rdi ; BLOCK *b - ;mov rsi, rsi ; BLOCKD *d - %endif -%endif - - mov rdx, [rdi + vp9_block_coeff] ; coeff_ptr - mov rcx, [rdi + vp9_block_zbin] ; zbin_ptr - movd xmm7, [rdi + vp9_block_zbin_extra] ; zbin_oq_value - - ; z - movdqa xmm0, [rdx] - movdqa xmm4, [rdx + 16] - mov rdx, [rdi + vp9_block_round] ; round_ptr - - pshuflw xmm7, xmm7, 0 - punpcklwd xmm7, xmm7 ; duplicated zbin_oq_value - - movdqa xmm1, xmm0 - movdqa xmm5, xmm4 - - ; sz - psraw xmm0, 15 - psraw xmm4, 15 - - ; (z ^ sz) - pxor xmm1, xmm0 - pxor xmm5, xmm4 - - ; x = abs(z) - psubw xmm1, xmm0 - psubw xmm5, xmm4 - - movdqa xmm2, [rcx] - movdqa xmm3, [rcx + 16] - mov rcx, [rdi + vp9_block_quant] ; quant_ptr - - ; *zbin_ptr + zbin_oq_value - paddw xmm2, xmm7 - paddw xmm3, xmm7 - - ; x - (*zbin_ptr + zbin_oq_value) - psubw xmm1, xmm2 - psubw xmm5, xmm3 - movdqa [rsp + abs_minus_zbin], xmm1 - movdqa [rsp + abs_minus_zbin + 16], xmm5 - - ; add (zbin_ptr + zbin_oq_value) back - paddw xmm1, xmm2 - paddw xmm5, xmm3 - - movdqa xmm2, [rdx] - movdqa xmm6, [rdx + 16] - - movdqa xmm3, [rcx] - movdqa xmm7, [rcx + 16] - - ; x + round - paddw xmm1, xmm2 - paddw xmm5, xmm6 - - ; y = x * quant_ptr >> 16 - pmulhw xmm3, xmm1 - pmulhw xmm7, xmm5 - - ; y += x - paddw xmm1, xmm3 - paddw xmm5, xmm7 - - movdqa [rsp + temp_qcoeff], xmm1 - movdqa [rsp + temp_qcoeff + 16], xmm5 - - pxor xmm6, xmm6 - ; zero qcoeff - movdqa [rsp + qcoeff], xmm6 - movdqa [rsp + qcoeff + 16], xmm6 - - mov rdx, [rdi + vp9_block_zrun_zbin_boost] ; zbin_boost_ptr - mov rax, [rdi + vp9_block_quant_shift] ; quant_shift_ptr - mov [rsp + zrun_zbin_boost], rdx - -%macro ZIGZAG_LOOP 1 - ; x - movsx ecx, WORD PTR[rsp + abs_minus_zbin + %1 * 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 - - movsx edi, WORD PTR[rsp + temp_qcoeff + %1 * 2] - - ; downshift by quant_shift[rc] - movsx cx, BYTE PTR[rax + %1] ; quant_shift_ptr[rc] - sar edi, cl ; also sets Z bit - je .rq_zigzag_loop_%1 ; !y - mov WORD PTR[rsp + qcoeff + %1 * 2], di ;qcoeff_ptr[rc] = temp_qcoeff[rc] - mov rdx, [rsp + zrun_zbin_boost] ; reset to b->zrun_zbin_boost -.rq_zigzag_loop_%1: -%endmacro -; in vp9_default_zig_zag1d order: see vp9/common/vp9_entropy.c -ZIGZAG_LOOP 0 -ZIGZAG_LOOP 1 -ZIGZAG_LOOP 4 -ZIGZAG_LOOP 8 -ZIGZAG_LOOP 5 -ZIGZAG_LOOP 2 -ZIGZAG_LOOP 3 -ZIGZAG_LOOP 6 -ZIGZAG_LOOP 9 -ZIGZAG_LOOP 12 -ZIGZAG_LOOP 13 -ZIGZAG_LOOP 10 -ZIGZAG_LOOP 7 -ZIGZAG_LOOP 11 -ZIGZAG_LOOP 14 -ZIGZAG_LOOP 15 - - movdqa xmm2, [rsp + qcoeff] - movdqa xmm3, [rsp + qcoeff + 16] - - mov rcx, [rsi + vp9_blockd_dequant] ; dequant_ptr - mov rdi, [rsi + vp9_blockd_dqcoeff] ; dqcoeff_ptr - - ; y ^ sz - pxor xmm2, xmm0 - pxor xmm3, xmm4 - ; x = (y ^ sz) - sz - psubw xmm2, xmm0 - psubw xmm3, xmm4 - - ; dequant - movdqa xmm0, [rcx] - movdqa xmm1, [rcx + 16] - - mov rcx, [rsi + vp9_blockd_qcoeff] ; qcoeff_ptr - - pmullw xmm0, xmm2 - pmullw xmm1, xmm3 - - movdqa [rcx], xmm2 ; store qcoeff - movdqa [rcx + 16], xmm3 - movdqa [rdi], xmm0 ; store dqcoeff - movdqa [rdi + 16], xmm1 - - ; select the last value (in zig_zag order) for EOB - pcmpeqw xmm2, xmm6 - pcmpeqw xmm3, xmm6 - ; ! - pcmpeqw xmm6, xmm6 - pxor xmm2, xmm6 - pxor xmm3, xmm6 - ; mask inv_zig_zag - pand xmm2, [GLOBAL(inv_zig_zag)] - pand xmm3, [GLOBAL(inv_zig_zag + 16)] - ; select the max value - pmaxsw xmm2, xmm3 - pshufd xmm3, xmm2, 00001110b - pmaxsw xmm2, xmm3 - pshuflw xmm3, xmm2, 00001110b - pmaxsw xmm2, xmm3 - pshuflw xmm3, xmm2, 00000001b - pmaxsw xmm2, xmm3 - movd eax, xmm2 - and eax, 0xff - mov [rsi + vp9_blockd_eob], eax - - ; begin epilog - add rsp, stack_size - pop rsp -%if ABI_IS_32BIT - pop rsi - pop rdi -%else - %if LIBVPX_YASM_WIN64 - pop rsi - pop rdi - %endif -%endif - RESTORE_GOT - RESTORE_XMM - pop rbp - ret - -; void vp9_fast_quantize_b_sse2 | arg -; (BLOCK *b, | 0 -; BLOCKD *d) | 1 - -global sym(vp9_fast_quantize_b_sse2) PRIVATE -sym(vp9_fast_quantize_b_sse2): - push rbp - mov rbp, rsp - GET_GOT rbx - -%if ABI_IS_32BIT - push rdi - push rsi -%else - %if LIBVPX_YASM_WIN64 - push rdi - push rsi - %else - ; these registers are used for passing arguments - %endif -%endif - - ; end prolog - -%if ABI_IS_32BIT - mov rdi, arg(0) ; BLOCK *b - mov rsi, arg(1) ; BLOCKD *d -%else - %if LIBVPX_YASM_WIN64 - 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 + vp9_block_coeff] - mov rcx, [rdi + vp9_block_round] - mov rdx, [rdi + vp9_block_quant_fast] - - ; z = coeff - movdqa xmm0, [rax] - movdqa xmm4, [rax + 16] - - ; dup z so we can save sz - movdqa xmm1, xmm0 - movdqa xmm5, xmm4 - - ; sz = z >> 15 - psraw xmm0, 15 - psraw xmm4, 15 - - ; x = abs(z) = (z ^ sz) - sz - pxor xmm1, xmm0 - pxor xmm5, xmm4 - psubw xmm1, xmm0 - psubw xmm5, xmm4 - - ; x += round - paddw xmm1, [rcx] - paddw xmm5, [rcx + 16] - - mov rax, [rsi + vp9_blockd_qcoeff] - mov rcx, [rsi + vp9_blockd_dequant] - mov rdi, [rsi + vp9_blockd_dqcoeff] - - ; y = x * quant >> 16 - pmulhw xmm1, [rdx] - pmulhw xmm5, [rdx + 16] - - ; x = (y ^ sz) - sz - pxor xmm1, xmm0 - pxor xmm5, xmm4 - psubw xmm1, xmm0 - psubw xmm5, xmm4 - - ; qcoeff = x - movdqa [rax], xmm1 - movdqa [rax + 16], xmm5 - - ; x * dequant - movdqa xmm2, xmm1 - movdqa xmm3, xmm5 - pmullw xmm2, [rcx] - pmullw xmm3, [rcx + 16] - - ; dqcoeff = x * dequant - movdqa [rdi], xmm2 - movdqa [rdi + 16], xmm3 - - pxor xmm4, xmm4 ;clear all bits - pcmpeqw xmm1, xmm4 - pcmpeqw xmm5, xmm4 - - pcmpeqw xmm4, xmm4 ;set all bits - pxor xmm1, xmm4 - pxor xmm5, xmm4 - - pand xmm1, [GLOBAL(inv_zig_zag)] - pand xmm5, [GLOBAL(inv_zig_zag + 16)] - - pmaxsw xmm1, xmm5 - - ; now down to 8 - pshufd xmm5, xmm1, 00001110b - - pmaxsw xmm1, xmm5 - - ; only 4 left - pshuflw xmm5, xmm1, 00001110b - - pmaxsw xmm1, xmm5 - - ; okay, just 2! - pshuflw xmm5, xmm1, 00000001b - - pmaxsw xmm1, xmm5 - - movd eax, xmm1 - and eax, 0xff - mov [rsi + vp9_blockd_eob], eax - - ; begin epilog -%if ABI_IS_32BIT - pop rsi - pop rdi -%else - %if LIBVPX_YASM_WIN64 - pop rsi - pop rdi - %endif -%endif - - RESTORE_GOT - pop rbp - ret - -SECTION_RODATA -align 16 -inv_zig_zag: - dw 0x0001, 0x0002, 0x0006, 0x0007 - dw 0x0003, 0x0005, 0x0008, 0x000d - dw 0x0004, 0x0009, 0x000c, 0x000e - dw 0x000a, 0x000b, 0x000f, 0x0010 diff --git a/vp9/encoder/x86/vp9_quantize_sse4.asm b/vp9/encoder/x86/vp9_quantize_sse4.asm deleted file mode 100644 index 4c14e5ffe..000000000 --- a/vp9/encoder/x86/vp9_quantize_sse4.asm +++ /dev/null @@ -1,253 +0,0 @@ -; -; 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" - - -; void vp9_regular_quantize_b_sse4 | arg -; (BLOCK *b, | 0 -; BLOCKD *d) | 1 - -global sym(vp9_regular_quantize_b_sse4) PRIVATE -sym(vp9_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 - %if LIBVPX_YASM_WIN64 - 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 - %if LIBVPX_YASM_WIN64 - 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 + vp9_block_coeff] - mov rcx, [rdi + vp9_block_zbin] - mov rdx, [rdi + vp9_block_round] - movd xmm7, [rdi + vp9_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 + vp9_block_quant_shift] - mov rcx, [rdi + vp9_block_quant] - mov rdx, [rdi + vp9_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 vp9_default_zig_zag1d order: see vp9/common/vp9_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 + vp9_blockd_dequant] - mov rdi, [rsi + vp9_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 + vp9_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 + vp9_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 - %if LIBVPX_YASM_WIN64 - pop rsi - pop rdi - RESTORE_XMM - %endif -%endif - - ret - -SECTION_RODATA -align 16 -; vp9/common/vp9_entropy.c: vp9_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/vp9/encoder/x86/vp9_quantize_ssse3.asm b/vp9/encoder/x86/vp9_quantize_ssse3.asm deleted file mode 100644 index 1fa052147..000000000 --- a/vp9/encoder/x86/vp9_quantize_ssse3.asm +++ /dev/null @@ -1,137 +0,0 @@ -; -; 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" - - -; void vp9_fast_quantize_b_ssse3 | arg -; (BLOCK *b, | 0 -; BLOCKD *d) | 1 -; - -global sym(vp9_fast_quantize_b_ssse3) PRIVATE -sym(vp9_fast_quantize_b_ssse3): - push rbp - mov rbp, rsp - GET_GOT rbx - -%if ABI_IS_32BIT - push rdi - push rsi -%else - %if LIBVPX_YASM_WIN64 - 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 - %if LIBVPX_YASM_WIN64 - 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 + vp9_block_coeff] - mov rcx, [rdi + vp9_block_round] - mov rdx, [rdi + vp9_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 - - ; sz = z >> 15 - psraw xmm0, 15 - psraw xmm4, 15 - - pabsw xmm1, xmm1 - pabsw xmm5, xmm5 - - paddw xmm1, xmm2 - paddw xmm5, xmm3 - - ; quant_fast - pmulhw xmm1, [rdx] - pmulhw xmm5, [rdx + 16] - - mov rax, [rsi + vp9_blockd_qcoeff] - mov rdi, [rsi + vp9_blockd_dequant] - mov rcx, [rsi + vp9_blockd_dqcoeff] - - pxor xmm1, xmm0 - pxor xmm5, xmm4 - psubw xmm1, xmm0 - psubw xmm5, xmm4 - - movdqa [rax], xmm1 - movdqa [rax + 16], xmm5 - - movdqa xmm2, [rdi] - movdqa xmm3, [rdi + 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 rdi, rdi - mov eax, -1 - xor dx, ax ;flip the bits for bsr - bsr eax, edx - - 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 + vp9_blockd_eob], eax - - ; begin epilog -%if ABI_IS_32BIT - pop rsi - pop rdi -%else - %if LIBVPX_YASM_WIN64 - pop rsi - pop rdi - %endif -%endif - - RESTORE_GOT - 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 diff --git a/vp9/encoder/x86/vp9_quantize_x86.h b/vp9/encoder/x86/vp9_quantize_x86.h deleted file mode 100644 index d1db17394..000000000 --- a/vp9/encoder/x86/vp9_quantize_x86.h +++ /dev/null @@ -1,48 +0,0 @@ -/* - * 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. - */ - -#ifndef VP9_ENCODER_X86_VP9_QUANTIZE_X86_H_ -#define VP9_ENCODER_X86_VP9_QUANTIZE_X86_H_ - - -/* Note: - * - * This platform is commonly built for runtime CPU detection. If you modify - * any of the function mappings present in this file, be sure to also update - * them in the function pointer initialization code - */ -#if HAVE_MMX - -#endif /* HAVE_MMX */ - - -#if HAVE_SSE2 -extern prototype_quantize_block(vp9_regular_quantize_b_sse2); -#if !CONFIG_RUNTIME_CPU_DETECT - -#undef vp9_quantize_quantb -#define vp9_quantize_quantb vp9_regular_quantize_b_sse2 -#endif /* !CONFIG_RUNTIME_CPU_DETECT */ - -#endif /* HAVE_SSE2 */ - - -#if HAVE_SSE4_1 -extern prototype_quantize_block(vp9_regular_quantize_b_sse4); - -#if !CONFIG_RUNTIME_CPU_DETECT - -#undef vp9_quantize_quantb -#define vp9_quantize_quantb vp9_regular_quantize_b_sse4 - -#endif /* !CONFIG_RUNTIME_CPU_DETECT */ - -#endif /* HAVE_SSE4_1 */ - -#endif /* QUANTIZE_X86_H */ diff --git a/vp9/vp9cx.mk b/vp9/vp9cx.mk index 63c6ed8b1..39f836fd4 100644 --- a/vp9/vp9cx.mk +++ b/vp9/vp9cx.mk @@ -17,15 +17,6 @@ VP9_CX_SRCS_REMOVE-no += $(VP9_COMMON_SRCS_REMOVE-no) VP9_CX_SRCS-yes += vp9_cx_iface.c -# encoder -#INCLUDES += algo/vpx_common/vpx_mem/include -#INCLUDES += common -#INCLUDES += common -#INCLUDES += common -#INCLUDES += algo/vpx_ref/cpu_id/include -#INCLUDES += common -#INCLUDES += encoder - VP9_CX_SRCS-yes += encoder/vp9_bitstream.c VP9_CX_SRCS-yes += encoder/vp9_boolhuff.c VP9_CX_SRCS-yes += encoder/vp9_dct.c @@ -81,7 +72,6 @@ VP9_CX_SRCS-yes += encoder/vp9_mbgraph.h VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_mcomp_x86.h -VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_quantize_x86.h VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_x86_csystemdependent.c VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_variance_mmx.c VP9_CX_SRCS-$(HAVE_MMX) += encoder/x86/vp9_variance_impl_mmx.asm @@ -94,17 +84,13 @@ VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_variance_impl_sse2.asm VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad_sse2.asm VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_sad4d_sse2.asm VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_fwalsh_sse2.asm -#VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_quantize_sse2.asm VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_subtract_sse2.asm VP9_CX_SRCS-$(HAVE_SSE2) += encoder/x86/vp9_temporal_filter_apply_sse2.asm VP9_CX_SRCS-$(HAVE_SSE3) += encoder/x86/vp9_sad_sse3.asm VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_sad_ssse3.asm VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_variance_ssse3.c VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_variance_impl_ssse3.asm -#VP9_CX_SRCS-$(HAVE_SSSE3) += encoder/x86/vp9_quantize_ssse3.asm VP9_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/vp9_sad_sse4.asm -#VP9_CX_SRCS-$(HAVE_SSE4_1) += encoder/x86/vp9_quantize_sse4.asm -VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_quantize_mmx.asm VP9_CX_SRCS-$(ARCH_X86)$(ARCH_X86_64) += encoder/x86/vp9_encodeopt.asm VP9_CX_SRCS-$(ARCH_X86_64) += encoder/x86/vp9_ssim_opt.asm