diff options
Diffstat (limited to '')
-rw-r--r-- | third_party/dav1d/src/x86/ipred_avx512.asm | 1432 |
1 files changed, 1432 insertions, 0 deletions
diff --git a/third_party/dav1d/src/x86/ipred_avx512.asm b/third_party/dav1d/src/x86/ipred_avx512.asm new file mode 100644 index 0000000000..38c86b54f5 --- /dev/null +++ b/third_party/dav1d/src/x86/ipred_avx512.asm @@ -0,0 +1,1432 @@ +; Copyright © 2020, VideoLAN and dav1d authors +; Copyright © 2020, Two Orioles, LLC +; All rights reserved. +; +; Redistribution and use in source and binary forms, with or without +; modification, are permitted provided that the following conditions are met: +; +; 1. Redistributions of source code must retain the above copyright notice, this +; list of conditions and the following disclaimer. +; +; 2. Redistributions in binary form must reproduce the above copyright notice, +; this list of conditions and the following disclaimer in the documentation +; and/or other materials provided with the distribution. +; +; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +; ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +; WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +; DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR +; ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES +; (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +; ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +; (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +%include "config.asm" +%include "ext/x86/x86inc.asm" + +%if ARCH_X86_64 + +SECTION_RODATA 64 + +%macro SMOOTH_WEIGHT_TABLE 1-* + %rep %0 + db %1-128, 127-%1 + %rotate 1 + %endrep +%endmacro + +smooth_weights: SMOOTH_WEIGHT_TABLE \ + 0, 0, 255, 128, 255, 149, 85, 64, \ + 255, 197, 146, 105, 73, 50, 37, 32, \ + 255, 225, 196, 170, 145, 123, 102, 84, \ + 68, 54, 43, 33, 26, 20, 17, 16, \ + 255, 240, 225, 210, 196, 182, 169, 157, \ + 145, 133, 122, 111, 101, 92, 83, 74, \ + 66, 59, 52, 45, 39, 34, 29, 25, \ + 21, 17, 14, 12, 10, 9, 8, 8, \ + 255, 248, 240, 233, 225, 218, 210, 203, \ + 196, 189, 182, 176, 169, 163, 156, 150, \ + 144, 138, 133, 127, 121, 116, 111, 106, \ + 101, 96, 91, 86, 82, 77, 73, 69, \ + 65, 61, 57, 54, 50, 47, 44, 41, \ + 38, 35, 32, 29, 27, 25, 22, 20, \ + 18, 16, 15, 13, 12, 10, 9, 8, \ + 7, 6, 6, 5, 5, 4, 4, 4 + +; dav1d_filter_intra_taps[], reordered for VNNI: p1 p2 p3 p4, p6 p5 p0 __ +filter_taps: db 10, 0, 0, 0, 2, 10, 0, 0, 1, 1, 10, 0, 1, 1, 2, 10 + db 6, 0, 0, 0, 2, 6, 0, 0, 2, 2, 6, 0, 1, 2, 2, 6 + db 0, 12, -6, 0, 0, 9, -5, 0, 0, 7, -3, 0, 0, 5, -3, 0 + db 12, 2, -4, 0, 9, 2, -3, 0, 7, 2, -3, 0, 5, 3, -3, 0 + db 16, 0, 0, 0, 0, 16, 0, 0, 0, 0, 16, 0, 0, 0, 0, 16 + db 16, 0, 0, 0, 0, 16, 0, 0, 0, 0, 16, 0, 0, 0, 0, 16 + db 0, 10,-10, 0, 0, 6, -6, 0, 0, 4, -4, 0, 0, 2, -2, 0 + db 10, 0,-10, 0, 6, 0, -6, 0, 4, 0, -4, 0, 2, 0, -2, 0 + db 8, 0, 0, 0, 0, 8, 0, 0, 0, 0, 8, 0, 0, 0, 0, 8 + db 4, 0, 0, 0, 0, 4, 0, 0, 0, 0, 4, 0, 0, 0, 0, 4 + db 0, 16, -8, 0, 0, 16, -8, 0, 0, 16, -8, 0, 0, 16, -8, 0 + db 16, 0, -4, 0, 16, 0, -4, 0, 16, 0, -4, 0, 16, 0, -4, 0 + db 8, 0, 0, 0, 3, 8, 0, 0, 2, 3, 8, 0, 1, 2, 3, 8 + db 4, 0, 0, 0, 3, 4, 0, 0, 2, 3, 4, 0, 2, 2, 3, 4 + db 0, 10, -2, 0, 0, 6, -1, 0, 0, 4, -1, 0, 0, 2, 0, 0 + db 10, 3, -1, 0, 6, 4, -1, 0, 4, 4, -1, 0, 3, 3, -1, 0 + db 14, 0, 0, 0, 0, 14, 0, 0, 0, 0, 14, 0, 0, 0, 0, 14 + db 12, 0, 0, 0, 1, 12, 0, 0, 0, 0, 12, 0, 0, 0, 1, 12 + db 0, 14,-12, 0, 0, 12,-10, 0, 0, 11, -9, 0, 0, 10, -8, 0 + db 14, 0,-10, 0, 12, 0, -9, 0, 11, 1, -8, 0, 9, 1, -7, 0 +filter_perm: db 0, 1, 2, 3, 24, 25, 26, 27, 4, 5, 6, 7, 28, 29, 30, 31 + db 15, 11, 7, 3, 15, 11, 7, 3, 15, 11, 7, 3, 15, 11, 7,131 + db 31, 27, 23, 19, 31, 27, 23, 19, 31, 27, 23, 19, 31, 27, 23,147 + db 47, 43, 39, 35, 47, 43, 39, 35, 47, 43, 39, 35, 47, 43, 39,163 +filter_end: dd 2, 3, 16, 17, -1, -1, 20, 21, 0, 6, 24, 30, 1, 7, 25, 31 +smooth_shuf: db 7, 7, 7, 7, 0, 1, 0, 1, 3, 3, 3, 3, 8, 9, 8, 9 + db 5, 5, 5, 5, 4, 5, 4, 5, 1, 1, 1, 1, 12, 13, 12, 13 + db 6, 6, 6, 6, 2, 3, 2, 3, 2, 2, 2, 2, 10, 11, 10, 11 + db 4, 4, 4, 4, 6, 7, 6, 7, 0, 0, 0, 0, 14, 15, 14, 15 +smooth_endA: db 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31 + db 33, 35, 37, 39, 41, 43, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63 + db 65, 67, 69, 71, 73, 75, 77, 79, 81, 83, 85, 87, 89, 91, 93, 95 + db 97, 99,101,103,105,107,109,111,113,115,117,119,121,123,125,127 +smooth_endB: db 1, 3, 5, 7, 9, 11, 13, 15, 65, 67, 69, 71, 73, 75, 77, 79 + db 17, 19, 21, 23, 25, 27, 29, 31, 81, 83, 85, 87, 89, 91, 93, 95 + db 33, 35, 37, 39, 41, 43, 45, 47, 97, 99,101,103,105,107,109,111 + db 49, 51, 53, 55, 57, 59, 61, 63,113,115,117,119,121,123,125,127 +ipred_h_shuf: db 7, 7, 7, 7, 6, 6, 6, 6, 5, 5, 5, 5, 4, 4, 4, 4 + db 3, 3, 3, 3, 2, 2, 2, 2, 1, 1, 1, 1, 0, 0, 0, 0 + +pb_127_m127: times 2 db 127, -127 +pb_128: times 4 db 128 +pw_128: times 2 dw 128 +pw_255: times 2 dw 255 + +%define pb_1 (ipred_h_shuf+24) +%define pb_2 (ipred_h_shuf+20) +%define pb_3 (ipred_h_shuf+16) +%define pd_8 (filter_taps+128) + +%macro JMP_TABLE 3-* + %xdefine %1_%2_table (%%table - 2*4) + %xdefine %%base mangle(private_prefix %+ _%1_%2) + %%table: + %rep %0 - 2 + dd %%base %+ .%3 - (%%table - 2*4) + %rotate 1 + %endrep +%endmacro + +%define ipred_dc_splat_8bpc_avx512icl_table (ipred_dc_8bpc_avx512icl_table + 10*4) + +JMP_TABLE ipred_h_8bpc, avx512icl, w4, w8, w16, w32, w64 +JMP_TABLE ipred_paeth_8bpc, avx512icl, w4, w8, w16, w32, w64 +JMP_TABLE ipred_smooth_8bpc, avx512icl, w4, w8, w16, w32, w64 +JMP_TABLE ipred_smooth_v_8bpc, avx512icl, w4, w8, w16, w32, w64 +JMP_TABLE ipred_smooth_h_8bpc, avx512icl, w4, w8, w16, w32, w64 +JMP_TABLE ipred_dc_8bpc, avx512icl, h4, h8, h16, h32, h64, w4, w8, w16, w32, w64, \ + s4-10*4, s8-10*4, s16-10*4, s32-10*4, s64-10*4 +JMP_TABLE ipred_dc_left_8bpc, avx512icl, h4, h8, h16, h32, h64 +JMP_TABLE pal_pred_8bpc, avx512icl, w4, w8, w16, w32, w64 + +SECTION .text + +INIT_ZMM avx512icl +cglobal ipred_dc_top_8bpc, 3, 7, 5, dst, stride, tl, w, h + lea r5, [ipred_dc_left_8bpc_avx512icl_table] + movd xm0, wm + tzcnt wd, wm + inc tlq + movifnidn hd, hm + movu ym1, [tlq] + movd xmm3, wd + movsxd r6, [r5+wq*4] + vpbroadcastd ym2, [r5-ipred_dc_left_8bpc_avx512icl_table+pb_1] + psrld xm0, 1 + vpdpbusd ym0, ym1, ym2 + add r6, r5 + add r5, ipred_dc_splat_8bpc_avx512icl_table-ipred_dc_left_8bpc_avx512icl_table + movsxd wq, [r5+wq*4] + add wq, r5 + jmp r6 + +cglobal ipred_dc_left_8bpc, 3, 7, 5, dst, stride, tl, w, h, stride3 + lea r5, [ipred_dc_left_8bpc_avx512icl_table] + mov hd, hm + tzcnt r6d, hd + sub tlq, hq + tzcnt wd, wm + movd xm0, hm + movu ym1, [tlq] + movd xmm3, r6d + movsxd r6, [r5+r6*4] + vpbroadcastd ym2, [r5-ipred_dc_left_8bpc_avx512icl_table+pb_1] + psrld xm0, 1 + vpdpbusd ym0, ym1, ym2 + add r6, r5 + add r5, ipred_dc_splat_8bpc_avx512icl_table-ipred_dc_left_8bpc_avx512icl_table + movsxd wq, [r5+wq*4] + add wq, r5 + jmp r6 +.h64: + movu ym1, [tlq+32] ; unaligned when jumping here from dc_top + vpdpbusd ym0, ym1, ym2 +.h32: + vextracti32x4 xm1, ym0, 1 + paddd xm0, xm1 +.h16: + punpckhqdq xm1, xm0, xm0 + paddd xm0, xm1 +.h8: + psrlq xm1, xm0, 32 + paddd xm0, xm1 +.h4: + vpsrlvd xm0, xmm3 + lea stride3q, [strideq*3] + vpbroadcastb m0, xm0 + jmp wq + +cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h, stride3 + movifnidn hd, hm + movifnidn wd, wm + tzcnt r6d, hd + lea r5d, [wq+hq] + movd xm0, r5d + tzcnt r5d, r5d + movd xmm4, r5d + lea r5, [ipred_dc_8bpc_avx512icl_table] + tzcnt wd, wd + movsxd r6, [r5+r6*4] + movsxd wq, [r5+wq*4+5*4] + vpbroadcastd ym3, [r5-ipred_dc_8bpc_avx512icl_table+pb_1] + psrld xm0, 1 + add r6, r5 + add wq, r5 + lea stride3q, [strideq*3] + jmp r6 +.h4: + movd xmm1, [tlq-4] + vpdpbusd xm0, xmm1, xm3 + jmp wq +.w4: + movd xmm1, [tlq+1] + vpdpbusd xm0, xmm1, xm3 + cmp hd, 4 + jg .w4_mul + psrlw xmm0, xm0, 3 + jmp .w4_end +.w4_mul: + punpckhqdq xmm1, xm0, xm0 + lea r2d, [hq*2] + mov r6d, 0x55563334 + paddd xmm1, xm0 + shrx r6d, r6d, r2d + psrlq xmm0, xmm1, 32 + paddd xmm0, xmm1 + movd xmm1, r6d + psrld xmm0, 2 + pmulhuw xmm0, xmm1 +.w4_end: + vpbroadcastb xm0, xmm0 +.s4: + movd [dstq+strideq*0], xm0 + movd [dstq+strideq*1], xm0 + movd [dstq+strideq*2], xm0 + movd [dstq+stride3q ], xm0 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .s4 + RET +.h8: + movq xmm1, [tlq-8] + vpdpbusd xm0, xmm1, xm3 + jmp wq +.w8: + movq xmm1, [tlq+1] + vextracti32x4 xm2, ym0, 1 + vpdpbusd xm0, xmm1, xm3 + paddd xmm2, xm2, xm0 + punpckhqdq xmm0, xmm2, xmm2 + paddd xmm0, xmm2 + psrlq xmm1, xmm0, 32 + paddd xmm0, xmm1 + vpsrlvd xmm0, xmm4 + cmp hd, 8 + je .w8_end + mov r6d, 0x5556 + mov r2d, 0x3334 + cmp hd, 32 + cmove r6d, r2d + movd xmm1, r6d + pmulhuw xmm0, xmm1 +.w8_end: + vpbroadcastb xm0, xmm0 +.s8: + movq [dstq+strideq*0], xm0 + movq [dstq+strideq*1], xm0 + movq [dstq+strideq*2], xm0 + movq [dstq+stride3q ], xm0 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .s8 + RET +.h16: + mova xmm1, [tlq-16] + vpdpbusd xm0, xmm1, xm3 + jmp wq +.w16: + movu xmm1, [tlq+1] + vextracti32x4 xm2, ym0, 1 + vpdpbusd xm0, xmm1, xm3 + paddd xmm2, xm2, xm0 + punpckhqdq xmm0, xmm2, xmm2 + paddd xmm0, xmm2 + psrlq xmm1, xmm0, 32 + paddd xmm0, xmm1 + vpsrlvd xmm0, xmm4 + cmp hd, 16 + je .w16_end + mov r6d, 0x5556 + mov r2d, 0x3334 + test hb, 8|32 + cmovz r6d, r2d + movd xmm1, r6d + pmulhuw xmm0, xmm1 +.w16_end: + vpbroadcastb xm0, xmm0 +.s16: + mova [dstq+strideq*0], xm0 + mova [dstq+strideq*1], xm0 + mova [dstq+strideq*2], xm0 + mova [dstq+stride3q ], xm0 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .s16 + RET +.h32: + mova ym1, [tlq-32] + vpdpbusd ym0, ym1, ym3 + jmp wq +.w32: + movu ym1, [tlq+1] + vpdpbusd ym0, ym1, ym3 + vextracti32x4 xm1, ym0, 1 + paddd xmm1, xm1, xm0 + punpckhqdq xmm0, xmm1, xmm1 + paddd xmm0, xmm1 + psrlq xmm1, xmm0, 32 + paddd xmm0, xmm1 + vpsrlvd xmm0, xmm4 + cmp hd, 32 + je .w32_end + lea r2d, [hq*2] + mov r6d, 0x33345556 + shrx r6d, r6d, r2d + movd xmm1, r6d + pmulhuw xmm0, xmm1 +.w32_end: + vpbroadcastb ym0, xmm0 +.s32: + mova [dstq+strideq*0], ym0 + mova [dstq+strideq*1], ym0 + mova [dstq+strideq*2], ym0 + mova [dstq+stride3q ], ym0 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .s32 + RET +.h64: + mova ym1, [tlq-64] + mova ym2, [tlq-32] + vpdpbusd ym0, ym1, ym3 + vpdpbusd ym0, ym2, ym3 + jmp wq +.w64: + movu ym1, [tlq+ 1] + movu ym2, [tlq+33] + vpdpbusd ym0, ym1, ym3 + vpdpbusd ym0, ym2, ym3 + vextracti32x4 xm1, ym0, 1 + paddd xmm1, xm1, xm0 + punpckhqdq xmm0, xmm1, xmm1 + paddd xmm0, xmm1 + psrlq xmm1, xmm0, 32 + paddd xmm0, xmm1 + vpsrlvd xmm0, xmm4 + cmp hd, 64 + je .w64_end + mov r6d, 0x33345556 + shrx r6d, r6d, hd + movd xmm1, r6d + pmulhuw xmm0, xmm1 +.w64_end: + vpbroadcastb m0, xmm0 +.s64: + mova [dstq+strideq*0], m0 + mova [dstq+strideq*1], m0 + mova [dstq+strideq*2], m0 + mova [dstq+stride3q ], m0 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .s64 + RET + +cglobal ipred_dc_128_8bpc, 2, 7, 5, dst, stride, tl, w, h, stride3 + lea r5, [ipred_dc_splat_8bpc_avx512icl_table] + tzcnt wd, wm + movifnidn hd, hm + movsxd wq, [r5+wq*4] + vpbroadcastd m0, [r5-ipred_dc_splat_8bpc_avx512icl_table+pb_128] + add wq, r5 + lea stride3q, [strideq*3] + jmp wq + +cglobal ipred_v_8bpc, 3, 7, 5, dst, stride, tl, w, h, stride3 + lea r5, [ipred_dc_splat_8bpc_avx512icl_table] + tzcnt wd, wm + movu m0, [tlq+1] + movifnidn hd, hm + movsxd wq, [r5+wq*4] + add wq, r5 + lea stride3q, [strideq*3] + jmp wq + +cglobal ipred_h_8bpc, 3, 7, 8, dst, stride, tl, w, h, stride3 +%define base r6-ipred_h_8bpc_avx512icl_table + lea r6, [ipred_h_8bpc_avx512icl_table] + tzcnt wd, wm + mov hd, hm + movsxd wq, [r6+wq*4] + lea stride3q, [strideq*3] + sub tlq, hq + add wq, r6 + jmp wq +.w4: + mova xmm1, [base+ipred_h_shuf+16] +.w4_loop: + movd xmm0, [tlq+hq-4] + pshufb xmm0, xmm1 + movd [dstq+strideq*0], xmm0 + pextrd [dstq+strideq*1], xmm0, 1 + pextrd [dstq+strideq*2], xmm0, 2 + pextrd [dstq+stride3q ], xmm0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w4_loop + RET +.w8: + movsldup xmm2, [base+ipred_h_shuf+16] + movshdup xmm3, [base+ipred_h_shuf+16] +.w8_loop: + movd xmm1, [tlq+hq-4] + pshufb xmm0, xmm1, xmm2 + pshufb xmm1, xmm3 + movq [dstq+strideq*0], xmm0 + movq [dstq+strideq*1], xmm1 + movhps [dstq+strideq*2], xmm0 + movhps [dstq+stride3q ], xmm1 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w8_loop + RET +.w16: + movsldup m1, [base+smooth_shuf] +.w16_loop: + vpbroadcastd m0, [tlq+hq-4] + pshufb m0, m1 + mova [dstq+strideq*0], xm0 + vextracti32x4 [dstq+strideq*1], m0, 2 + vextracti32x4 [dstq+strideq*2], ym0, 1 + vextracti32x4 [dstq+stride3q ], m0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w16 + RET +.w32: + vpbroadcastd ym3, [base+pb_1] + vpord m2, m3, [base+pb_2] {1to16} +.w32_loop: + vpbroadcastd m1, [tlq+hq-4] + pshufb m0, m1, m2 + pshufb m1, m3 + mova [dstq+strideq*0], ym0 + vextracti32x8 [dstq+strideq*1], m0, 1 + mova [dstq+strideq*2], ym1 + vextracti32x8 [dstq+stride3q ], m1, 1 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w32_loop + RET +.w64: + vpbroadcastd m4, [base+pb_3] + vpbroadcastd m5, [base+pb_2] + vpbroadcastd m6, [base+pb_1] + pxor m7, m7 +.w64_loop: + vpbroadcastd m3, [tlq+hq-4] + pshufb m0, m3, m4 + pshufb m1, m3, m5 + pshufb m2, m3, m6 + pshufb m3, m7 + mova [dstq+strideq*0], m0 + mova [dstq+strideq*1], m1 + mova [dstq+strideq*2], m2 + mova [dstq+stride3q ], m3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w64_loop + RET + +%macro PAETH 0 + psubusb m1, m5, m4 + psubusb m0, m4, m5 + por m1, m0 ; tdiff + pavgb m2, m6, m4 + vpcmpub k1, m1, m7, 1 ; tdiff < ldiff + vpblendmb m0{k1}, m4, m6 + vpternlogd m4, m6, m8, 0x28 ; (m4 ^ m6) & m8 + psubusb m3, m5, m2 + psubb m2, m4 + psubusb m2, m5 + por m2, m3 + pminub m1, m7 + paddusb m2, m2 + por m2, m4 ; min(tldiff, 255) + vpcmpub k1, m2, m1, 1 ; tldiff < ldiff && tldiff < tdiff + vmovdqu8 m0{k1}, m5 +%endmacro + +cglobal ipred_paeth_8bpc, 3, 7, 10, dst, stride, tl, w, h, top, stride3 + lea r6, [ipred_paeth_8bpc_avx512icl_table] + tzcnt wd, wm + vpbroadcastb m5, [tlq] ; topleft + mov hd, hm + movsxd wq, [r6+wq*4] + vpbroadcastd m8, [r6-ipred_paeth_8bpc_avx512icl_table+pb_1] + lea topq, [tlq+1] + sub tlq, hq + add wq, r6 + lea stride3q, [strideq*3] + jmp wq +INIT_YMM avx512icl +.w4: + vpbroadcastd m6, [topq] + mova m9, [ipred_h_shuf] + psubusb m7, m5, m6 + psubusb m0, m6, m5 + por m7, m0 ; ldiff +.w4_loop: + vpbroadcastq m4, [tlq+hq-8] + pshufb m4, m9 ; left + PAETH + movd [dstq+strideq*0], xm0 + pextrd [dstq+strideq*1], xm0, 1 + pextrd [dstq+strideq*2], xm0, 2 + pextrd [dstq+stride3q ], xm0, 3 + sub hd, 8 + jl .w4_ret + vextracti32x4 xm0, m0, 1 + lea dstq, [dstq+strideq*4] + movd [dstq+strideq*0], xm0 + pextrd [dstq+strideq*1], xm0, 1 + pextrd [dstq+strideq*2], xm0, 2 + pextrd [dstq+stride3q ], xm0, 3 + lea dstq, [dstq+strideq*4] + jg .w4_loop +.w4_ret: + RET +INIT_ZMM avx512icl +.w8: + vpbroadcastq m6, [topq] + movsldup m9, [smooth_shuf] + psubusb m7, m5, m6 + psubusb m0, m6, m5 + por m7, m0 +.w8_loop: + vpbroadcastq m4, [tlq+hq-8] + pshufb m4, m9 + PAETH + vextracti32x4 xm1, m0, 2 + vextracti32x4 xm2, ym0, 1 + vextracti32x4 xm3, m0, 3 + movq [dstq+strideq*0], xm0 + movq [dstq+strideq*1], xm1 + movq [dstq+strideq*2], xm2 + movq [dstq+stride3q ], xm3 + sub hd, 8 + jl .w8_ret + lea dstq, [dstq+strideq*4] + movhps [dstq+strideq*0], xm0 + movhps [dstq+strideq*1], xm1 + movhps [dstq+strideq*2], xm2 + movhps [dstq+stride3q ], xm3 + lea dstq, [dstq+strideq*4] + jg .w8_loop +.w8_ret: + RET +.w16: + vbroadcasti32x4 m6, [topq] + movsldup m9, [smooth_shuf] + psubusb m7, m5, m6 + psubusb m0, m6, m5 + por m7, m0 +.w16_loop: + vpbroadcastd m4, [tlq+hq-4] + pshufb m4, m9 + PAETH + mova [dstq+strideq*0], xm0 + vextracti32x4 [dstq+strideq*1], m0, 2 + vextracti32x4 [dstq+strideq*2], ym0, 1 + vextracti32x4 [dstq+stride3q ], m0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w16_loop + RET +.w32: + vbroadcasti32x8 m6, [topq] + mova ym9, ym8 + psubusb m7, m5, m6 + psubusb m0, m6, m5 + por m7, m0 +.w32_loop: + vpbroadcastd m4, [tlq+hq-2] + pshufb m4, m9 + PAETH + mova [dstq+strideq*0], ym0 + vextracti32x8 [dstq+strideq*1], m0, 1 + lea dstq, [dstq+strideq*2] + sub hd, 2 + jg .w32_loop + RET +.w64: + movu m6, [topq] + psubusb m7, m5, m6 + psubusb m0, m6, m5 + por m7, m0 +.w64_loop: + vpbroadcastb m4, [tlq+hq-1] + PAETH + mova [dstq], m0 + add dstq, strideq + dec hd + jg .w64_loop + RET + +cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl, w, h, weights, stride3 +%define base r6-ipred_smooth_v_8bpc_avx512icl_table + lea r6, [ipred_smooth_v_8bpc_avx512icl_table] + tzcnt wd, wm + mov hd, hm + movsxd wq, [r6+wq*4] + vpbroadcastd m0, [base+pb_127_m127] + vpbroadcastd m1, [base+pw_128] + lea weightsq, [base+smooth_weights+hq*4] + neg hq + vpbroadcastb m4, [tlq+hq] ; bottom + add wq, r6 + lea stride3q, [strideq*3] + jmp wq +.w4: + vpbroadcastd m2, [tlq+1] + movshdup m5, [smooth_shuf] + mova ym6, [smooth_endA] + punpcklbw m2, m4 ; top, bottom + pmaddubsw m3, m2, m0 + paddw m1, m2 ; 1 * top + 256 * bottom + 128, overflow is ok + paddw m3, m1 ; 128 * top + 129 * bottom + 128 +.w4_loop: + vbroadcasti32x4 m0, [weightsq+hq*2] + pshufb m0, m5 + pmaddubsw m0, m2, m0 + paddw m0, m3 + vpermb m0, m6, m0 + vextracti32x4 xm1, ym0, 1 + movd [dstq+strideq*0], xm0 + movd [dstq+strideq*1], xm1 + pextrd [dstq+strideq*2], xm0, 2 + pextrd [dstq+stride3q ], xm1, 2 + add hq, 8 + jg .ret + lea dstq, [dstq+strideq*4] + pextrd [dstq+strideq*0], xm0, 1 + pextrd [dstq+strideq*1], xm1, 1 + pextrd [dstq+strideq*2], xm0, 3 + pextrd [dstq+stride3q ], xm1, 3 + lea dstq, [dstq+strideq*4] + jl .w4_loop +.ret: + RET +.w8: + vpbroadcastq m2, [tlq+1] + movshdup m5, [smooth_shuf] + mova ym6, [smooth_endA] + punpcklbw m2, m4 + pmaddubsw m3, m2, m0 + paddw m1, m2 + paddw m3, m1 +.w8_loop: + vpbroadcastq m0, [weightsq+hq*2] + pshufb m0, m5 + pmaddubsw m0, m2, m0 + paddw m0, m3 + vpermb m0, m6, m0 + vextracti32x4 xm1, ym0, 1 + movq [dstq+strideq*0], xm0 + movq [dstq+strideq*1], xm1 + movhps [dstq+strideq*2], xm0 + movhps [dstq+stride3q ], xm1 + lea dstq, [dstq+strideq*4] + add hq, 4 + jl .w8_loop + RET +.w16: + vbroadcasti32x4 m3, [tlq+1] + movshdup m6, [smooth_shuf] + mova m7, [smooth_endB] + punpcklbw m2, m3, m4 + punpckhbw m3, m4 + pmaddubsw m4, m2, m0 + pmaddubsw m5, m3, m0 + paddw m0, m1, m2 + paddw m1, m3 + paddw m4, m0 + paddw m5, m1 +.w16_loop: + vpbroadcastq m1, [weightsq+hq*2] + pshufb m1, m6 + pmaddubsw m0, m2, m1 + pmaddubsw m1, m3, m1 + paddw m0, m4 + paddw m1, m5 + vpermt2b m0, m7, m1 + mova [dstq+strideq*0], xm0 + vextracti32x4 [dstq+strideq*1], m0, 2 + vextracti32x4 [dstq+strideq*2], ym0, 1 + vextracti32x4 [dstq+stride3q ], m0, 3 + lea dstq, [dstq+strideq*4] + add hq, 4 + jl .w16_loop + RET +.w32: + vbroadcasti32x8 m3, [tlq+1] + movshdup m6, [smooth_shuf] + mova m7, [smooth_endB] + punpcklbw m2, m3, m4 + punpckhbw m3, m4 + pmaddubsw m4, m2, m0 + pmaddubsw m5, m3, m0 + paddw m0, m1, m2 + paddw m1, m3 + paddw m4, m0 + paddw m5, m1 +.w32_loop: + vpbroadcastd m1, [weightsq+hq*2] + pshufb m1, m6 + pmaddubsw m0, m2, m1 + pmaddubsw m1, m3, m1 + paddw m0, m4 + paddw m1, m5 + vpermt2b m0, m7, m1 + mova [dstq+strideq*0], ym0 + vextracti32x8 [dstq+strideq*1], m0, 1 + lea dstq, [dstq+strideq*2] + add hq, 2 + jl .w32_loop + RET +.w64: + movu m3, [tlq+1] + mova m6, [smooth_endB] + punpcklbw m2, m3, m4 + punpckhbw m3, m4 + pmaddubsw m4, m2, m0 + pmaddubsw m5, m3, m0 + paddw m0, m1, m2 + paddw m1, m3 + paddw m4, m0 + paddw m5, m1 +.w64_loop: + vpbroadcastw m1, [weightsq+hq*2] + pmaddubsw m0, m2, m1 + pmaddubsw m1, m3, m1 + paddw m0, m4 + paddw m1, m5 + vpermt2b m0, m6, m1 + mova [dstq], m0 + add dstq, strideq + inc hq + jl .w64_loop + RET + +cglobal ipred_smooth_h_8bpc, 4, 7, 11, dst, stride, tl, w, h, stride3 +%define base r5-ipred_smooth_h_8bpc_avx512icl_table + lea r5, [ipred_smooth_h_8bpc_avx512icl_table] + mov r6d, wd + tzcnt wd, wd + vpbroadcastb m4, [tlq+r6] ; right + mov hd, hm + movsxd wq, [r5+wq*4] + vpbroadcastd m5, [base+pb_127_m127] + vpbroadcastd m6, [base+pw_128] + sub tlq, hq + add wq, r5 + vpmovb2m k1, m6 + lea stride3q, [strideq*3] + jmp wq +.w4: + movsldup m3, [smooth_shuf] + vpbroadcastq m7, [smooth_weights+4*2] + mova ym8, [smooth_endA] +.w4_loop: + vpbroadcastq m0, [tlq+hq-8] + mova m2, m4 + vpshufb m2{k1}, m0, m3 ; left, right + pmaddubsw m0, m2, m5 + pmaddubsw m1, m2, m7 + paddw m2, m6 + paddw m0, m2 + paddw m0, m1 + vpermb m0, m8, m0 + vextracti32x4 xm1, ym0, 1 + movd [dstq+strideq*0], xm0 + movd [dstq+strideq*1], xm1 + pextrd [dstq+strideq*2], xm0, 2 + pextrd [dstq+stride3q ], xm1, 2 + sub hd, 8 + jl .ret + lea dstq, [dstq+strideq*4] + pextrd [dstq+strideq*0], xm0, 1 + pextrd [dstq+strideq*1], xm1, 1 + pextrd [dstq+strideq*2], xm0, 3 + pextrd [dstq+stride3q ], xm1, 3 + lea dstq, [dstq+strideq*4] + jg .w4_loop +.ret: + RET +.w8: + movsldup m3, [smooth_shuf] + vbroadcasti32x4 m7, [smooth_weights+8*2] + mova ym8, [smooth_endA] +.w8_loop: + vpbroadcastd m0, [tlq+hq-4] + mova m2, m4 + vpshufb m2{k1}, m0, m3 + pmaddubsw m0, m2, m5 + pmaddubsw m1, m2, m7 + paddw m2, m6 + paddw m0, m2 + paddw m0, m1 + vpermb m0, m8, m0 + vextracti32x4 xm1, ym0, 1 + movq [dstq+strideq*0], xm0 + movq [dstq+strideq*1], xm1 + movhps [dstq+strideq*2], xm0 + movhps [dstq+stride3q ], xm1 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w8_loop + RET +.w16: + movsldup m7, [smooth_shuf] + vbroadcasti32x4 m8, [smooth_weights+16*2] + vbroadcasti32x4 m9, [smooth_weights+16*3] + mova m10, [smooth_endB] +.w16_loop: + vpbroadcastd m0, [tlq+hq-4] + mova m3, m4 + vpshufb m3{k1}, m0, m7 + pmaddubsw m2, m3, m5 + pmaddubsw m0, m3, m8 + pmaddubsw m1, m3, m9 + paddw m3, m6 + paddw m2, m3 + paddw m0, m2 + paddw m1, m2 + vpermt2b m0, m10, m1 + mova [dstq+strideq*0], xm0 + vextracti32x4 [dstq+strideq*1], m0, 2 + vextracti32x4 [dstq+strideq*2], ym0, 1 + vextracti32x4 [dstq+stride3q ], m0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w16_loop + RET +.w32: + mova m10, [smooth_endA] + vpbroadcastd ym7, [pb_1] + vbroadcasti32x8 m8, [smooth_weights+32*2] + vbroadcasti32x8 m9, [smooth_weights+32*3] + vshufi32x4 m10, m10, q3120 +.w32_loop: + vpbroadcastd m0, [tlq+hq-2] + mova m3, m4 + vpshufb m3{k1}, m0, m7 + pmaddubsw m2, m3, m5 + pmaddubsw m0, m3, m8 + pmaddubsw m1, m3, m9 + paddw m3, m6 + paddw m2, m3 + paddw m0, m2 + paddw m1, m2 + vpermt2b m0, m10, m1 + mova [dstq+strideq*0], ym0 + vextracti32x8 [dstq+strideq*1], m0, 1 + lea dstq, [dstq+strideq*2] + sub hd, 2 + jg .w32_loop + RET +.w64: + mova m7, [smooth_weights+64*2] + mova m8, [smooth_weights+64*3] + mova m9, [smooth_endA] +.w64_loop: + mova m3, m4 + vpbroadcastb m3{k1}, [tlq+hq-1] + pmaddubsw m2, m3, m5 + pmaddubsw m0, m3, m7 + pmaddubsw m1, m3, m8 + paddw m3, m6 + paddw m2, m3 + paddw m0, m2 + paddw m1, m2 + vpermt2b m0, m9, m1 + mova [dstq], m0 + add dstq, strideq + dec hd + jg .w64_loop + RET + +cglobal ipred_smooth_8bpc, 4, 7, 16, dst, stride, tl, w, h, v_weights, stride3 +%define base r5-ipred_smooth_8bpc_avx512icl_table + lea r5, [ipred_smooth_8bpc_avx512icl_table] + mov r6d, wd + tzcnt wd, wd + mov hd, hm + vpbroadcastb m6, [tlq+r6] ; right + sub tlq, hq + movsxd wq, [r5+wq*4] + vpbroadcastd m7, [base+pb_127_m127] + vpbroadcastb m0, [tlq] ; bottom + vpbroadcastd m1, [base+pw_255] + add wq, r5 + lea v_weightsq, [base+smooth_weights+hq*2] + vpmovb2m k1, m1 + lea stride3q, [strideq*3] + jmp wq +.w4: + vpbroadcastd m8, [tlq+hq+1] + movsldup m4, [smooth_shuf] + movshdup m5, [smooth_shuf] + vpbroadcastq m9, [smooth_weights+4*2] + mova ym11, [smooth_endA] + + punpcklbw m8, m0 ; top, bottom + pmaddubsw m10, m8, m7 + paddw m1, m8 ; 1 * top + 256 * bottom + 255 + paddw m10, m1 ; 128 * top + 129 * bottom + 255 +.w4_loop: + vpbroadcastq m1, [tlq+hq-8] + vbroadcasti32x4 m0, [v_weightsq] + add v_weightsq, 16 + mova m2, m6 + vpshufb m2{k1}, m1, m4 ; left, right + pmaddubsw m1, m2, m7 ; 127 * left - 127 * right + pshufb m0, m5 + pmaddubsw m0, m8, m0 + paddw m1, m2 ; 128 * left + 129 * right + pmaddubsw m2, m9 + paddw m0, m10 + paddw m1, m2 + pavgw m0, m1 + vpermb m0, m11, m0 + vextracti32x4 xm1, ym0, 1 + movd [dstq+strideq*0], xm0 + movd [dstq+strideq*1], xm1 + pextrd [dstq+strideq*2], xm0, 2 + pextrd [dstq+stride3q ], xm1, 2 + sub hd, 8 + jl .ret + lea dstq, [dstq+strideq*4] + pextrd [dstq+strideq*0], xm0, 1 + pextrd [dstq+strideq*1], xm1, 1 + pextrd [dstq+strideq*2], xm0, 3 + pextrd [dstq+stride3q ], xm1, 3 + lea dstq, [dstq+strideq*4] + jg .w4_loop +.ret: + RET +.w8: + vpbroadcastq m8, [tlq+hq+1] + movsldup m4, [smooth_shuf] + movshdup m5, [smooth_shuf] + vbroadcasti32x4 m9, [smooth_weights+8*2] + mova ym11, [smooth_endA] + punpcklbw m8, m0 + pmaddubsw m10, m8, m7 + paddw m1, m8 + paddw m10, m1 +.w8_loop: + vpbroadcastd m1, [tlq+hq-4] + vpbroadcastq m0, [v_weightsq] + add v_weightsq, 8 + mova m2, m6 + vpshufb m2{k1}, m1, m4 + pmaddubsw m1, m2, m7 + pshufb m0, m5 + pmaddubsw m0, m8, m0 + paddw m1, m2 + pmaddubsw m2, m9 + paddw m0, m10 + paddw m1, m2 + pavgw m0, m1 + vpermb m0, m11, m0 + vextracti32x4 xm1, ym0, 1 + movq [dstq+strideq*0], xm0 + movq [dstq+strideq*1], xm1 + movhps [dstq+strideq*2], xm0 + movhps [dstq+stride3q ], xm1 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w8_loop + RET +.w16: + vbroadcasti32x4 m9, [tlq+hq+1] + movsldup m5, [smooth_shuf] + movshdup m10, [smooth_shuf] + vbroadcasti32x4 m11, [smooth_weights+16*2] + vbroadcasti32x4 m12, [smooth_weights+16*3] + mova m15, [smooth_endB] + punpcklbw m8, m9, m0 + punpckhbw m9, m0 + pmaddubsw m13, m8, m7 + pmaddubsw m14, m9, m7 + paddw m0, m1, m8 + paddw m1, m9 + paddw m13, m0 + paddw m14, m1 +.w16_loop: + vpbroadcastd m0, [tlq+hq-4] + vpbroadcastq m1, [v_weightsq] + add v_weightsq, 8 + mova m4, m6 + vpshufb m4{k1}, m0, m5 + pmaddubsw m2, m4, m7 + pshufb m1, m10 + pmaddubsw m0, m8, m1 + pmaddubsw m1, m9, m1 + paddw m2, m4 + pmaddubsw m3, m4, m11 + pmaddubsw m4, m12 + paddw m0, m13 + paddw m1, m14 + paddw m3, m2 + paddw m4, m2 + pavgw m0, m3 + pavgw m1, m4 + vpermt2b m0, m15, m1 + mova [dstq+strideq*0], xm0 + vextracti32x4 [dstq+strideq*1], m0, 2 + vextracti32x4 [dstq+strideq*2], ym0, 1 + vextracti32x4 [dstq+stride3q ], m0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w16_loop + RET +.w32: + vbroadcasti32x8 m9, [tlq+hq+1] + movshdup m10, [smooth_shuf] + mova m12, [smooth_weights+32*2] + vpbroadcastd ym5, [pb_1] + mova m15, [smooth_endB] + punpcklbw m8, m9, m0 + punpckhbw m9, m0 + pmaddubsw m13, m8, m7 + pmaddubsw m14, m9, m7 + vshufi32x4 m11, m12, m12, q2020 + vshufi32x4 m12, m12, q3131 + paddw m0, m1, m8 + paddw m1, m9 + paddw m13, m0 + paddw m14, m1 +.w32_loop: + vpbroadcastd m0, [tlq+hq-2] + vpbroadcastd m1, [v_weightsq] + add v_weightsq, 4 + mova m4, m6 + vpshufb m4{k1}, m0, m5 + pmaddubsw m2, m4, m7 + pshufb m1, m10 + pmaddubsw m0, m8, m1 + pmaddubsw m1, m9, m1 + paddw m2, m4 + pmaddubsw m3, m4, m11 + pmaddubsw m4, m12 + paddw m0, m13 + paddw m1, m14 + paddw m3, m2 + paddw m4, m2 + pavgw m0, m3 + pavgw m1, m4 + vpermt2b m0, m15, m1 + mova [dstq+strideq*0], ym0 + vextracti32x8 [dstq+strideq*1], m0, 1 + lea dstq, [dstq+strideq*2] + sub hd, 2 + jg .w32_loop + RET +.w64: + movu m9, [tlq+hq+1] + mova m11, [smooth_weights+64*2] + mova m2, [smooth_weights+64*3] + mova m14, [smooth_endB] + punpcklbw m8, m9, m0 + punpckhbw m9, m0 + pmaddubsw m12, m8, m7 + pmaddubsw m13, m9, m7 + vshufi32x4 m10, m11, m2, q2020 + vshufi32x4 m11, m2, q3131 + paddw m0, m1, m8 + paddw m1, m9 + paddw m12, m0 + paddw m13, m1 +.w64_loop: + mova m4, m6 + vpbroadcastb m4{k1}, [tlq+hq-1] + vpbroadcastw m1, [v_weightsq] + add v_weightsq, 2 + pmaddubsw m2, m4, m7 + pmaddubsw m0, m8, m1 + pmaddubsw m1, m9, m1 + paddw m2, m4 + pmaddubsw m3, m4, m10 + pmaddubsw m4, m11 + paddw m0, m12 + paddw m1, m13 + paddw m3, m2 + paddw m4, m2 + pavgw m0, m3 + pavgw m1, m4 + vpermt2b m0, m14, m1 + mova [dstq], m0 + add dstq, strideq + dec hd + jg .w64_loop + RET + +cglobal pal_pred_8bpc, 4, 7, 5, dst, stride, pal, idx, w, h, stride3 + lea r6, [pal_pred_8bpc_avx512icl_table] + tzcnt wd, wm + vbroadcasti32x4 m4, [palq] + movifnidn hd, hm + movsxd wq, [r6+wq*4] + packuswb m4, m4 + add wq, r6 + lea stride3q, [strideq*3] + jmp wq +.w4: + pshufb xmm0, xm4, [idxq] + add idxq, 16 + movd [dstq+strideq*0], xmm0 + pextrd [dstq+strideq*1], xmm0, 1 + pextrd [dstq+strideq*2], xmm0, 2 + pextrd [dstq+stride3q ], xmm0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w4 + RET +.w8: + pshufb xmm0, xm4, [idxq+16*0] + pshufb xmm1, xm4, [idxq+16*1] + add idxq, 16*2 + movq [dstq+strideq*0], xmm0 + movhps [dstq+strideq*1], xmm0 + movq [dstq+strideq*2], xmm1 + movhps [dstq+stride3q ], xmm1 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w8 + RET +.w16: + pshufb m0, m4, [idxq] + add idxq, 64 + mova [dstq+strideq*0], xm0 + vextracti32x4 [dstq+strideq*1], ym0, 1 + vextracti32x4 [dstq+strideq*2], m0, 2 + vextracti32x4 [dstq+stride3q ], m0, 3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w16 + RET +.w32: + pshufb m0, m4, [idxq+64*0] + pshufb m1, m4, [idxq+64*1] + add idxq, 64*2 + mova [dstq+strideq*0], ym0 + vextracti32x8 [dstq+strideq*1], m0, 1 + mova [dstq+strideq*2], ym1 + vextracti32x8 [dstq+stride3q ], m1, 1 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w32 + RET +.w64: + pshufb m0, m4, [idxq+64*0] + pshufb m1, m4, [idxq+64*1] + pshufb m2, m4, [idxq+64*2] + pshufb m3, m4, [idxq+64*3] + add idxq, 64*4 + mova [dstq+strideq*0], m0 + mova [dstq+strideq*1], m1 + mova [dstq+strideq*2], m2 + mova [dstq+stride3q ], m3 + lea dstq, [dstq+strideq*4] + sub hd, 4 + jg .w64 + RET + +; The ipred_filter code processes 4x2 blocks in the following order +; which increases parallelism compared to doing things row by row. +; Some redundant blocks are calculated for w > 4. +; w4 w8 w16 w32 +; 1 1 2 1 2 3 4 1 2 3 4 9 a b c +; 2 2 3 2 3 4 5 2 3 4 5 a b c d +; 3 3 4 3 4 5 6 3 4 5 6 b c d e +; 4 4 5 4 5 6 7 4 5 6 7 c d e f +; 5 5 6 5 6 7 8 5 6 7 8 d e f g +; 6 6 7 6 7 8 9 6 7 8 9 e f g h +; 7 7 8 7 8 9 a 7 8 9 a f g h i +; ___ 8 ___ 8 9 ___ 8 9 a b ___ 8 9 a b g h i j ___ +; 9 9 a b h i j +; a b i j +; b j + +cglobal ipred_filter_8bpc, 4, 7, 14, dst, stride, tl, w, h, flt +%define base r6-filter_taps + lea r6, [filter_taps] +%ifidn fltd, fltm + movzx fltd, fltb +%else + movzx fltd, byte fltm +%endif + vpbroadcastd xmm2, [tlq+1] ; t0 t0 t0 t0 + movifnidn hd, hm + shl fltd, 6 + vpbroadcastd m6, [base+pd_8] + vpbroadcastd xmm3, [tlq-2] ; l1 l0 tl __ + vbroadcasti32x4 m7, [r6+fltq+16*0] ; p1 p2 p3 p4 + vbroadcasti32x4 m8, [r6+fltq+16*1] + vbroadcasti32x4 m9, [r6+fltq+16*2] ; p6 p5 p0 __ + vbroadcasti32x4 m10, [r6+fltq+16*3] + mova xmm0, xm6 + vpdpbusd xmm0, xmm2, xm7 + mova xmm1, xm6 + vpdpbusd xmm1, xmm2, xm8 + vpdpbusd xmm0, xmm3, xm9 + vpdpbusd xmm1, xmm3, xm10 + packssdw xmm0, xmm1 + cmp wd, 8 + jb .w4 + vpbroadcastd ym2, [tlq+5] + mova m11, [base+filter_perm] + mov r5, 0xffffffffffff000f + psrldq xmm2, 1 ; __ t0 + kmovq k1, r5 ; 0x000f + psraw xm5, xmm0, 4 + packuswb xmm2, xm5 ; __ t0 a0 b0 + pshufd ym2{k1}, ymm2, q3333 ; b0 b0 b0 b0 t1 t1 t1 t1 + je .w8 + kxnorb k3, k3, k3 ; 0x00ff + vpbroadcastd xm3, [tlq-4] + kandnq k2, k3, k1 ; 0xffffffffffff0000 + vpermb ym3{k2}, ym11, ymm2 ; l3 l2 l1 __ b3 a3 t3 __ + mova ym0, ym6 + vpdpbusd ym0, ym2, ym7 + mova ym1, ym6 + vpdpbusd ym1, ym2, ym8 + pshufb ym5{k2}, ym2, ym11 ; a0 b0 __ t0 + vpbroadcastd m2, [tlq+9] + vpdpbusd ym0, ym3, ym9 + vpdpbusd ym1, ym3, ym10 + vpbroadcastd xm3, [tlq-6] ; l5 l4 l3 __ + kunpckbw k4, k1, k3 ; 0x0fff + packssdw ym0, ym1 + psraw ym0, 4 ; a0 d0 a1 b1 + packuswb ym5, ym0 ; a0 b0 c0 d0 __ t1 a1 b1 + pshufd m2{k3}, m5, q3333 ; d0 d0 d0 d0 b1 b1 b1 b1 t2 t2 t2 t2 + vpermb m3{k2}, m11, m5 ; l5 l4 l3 __ d3 c3 b3 __ b7 a7 t7 __ + mova m4, m6 + vpdpbusd m4, m2, m7 + mova m1, m6 + vpdpbusd m1, m2, m8 + psrldq m0, m2, 1 ; __ d0 __ b0 __ t0 + vpbroadcastd m2, [tlq+13] + vpdpbusd m4, m3, m9 + vpdpbusd m1, m3, m10 + mova m12, [base+filter_end] + lea r5d, [hq-6] + mov r6, dstq + cmovp hd, r5d ; w == 16 ? h : h - 6 + packssdw m4, m1 + psraw m4, 4 ; e0 f0 c1 d1 a2 b2 + packuswb m0, m4 ; __ d0 e0 f0 __ b1 c1 d1 __ t2 a2 b2 + pshufd m2{k4}, m0, q3333 ; f0 f0 f0 f0 d1 d1 d1 d1 b2 b2 b2 b2 t3 t3 t3 t3 +.w16_loop: + vpbroadcastd xm3, [tlq-8] + vpermb m3{k2}, m11, m0 ; l7 l6 l5 __ f3 e3 d3 __ d7 c7 b7 __ bb ab tb __ + mova m1, m6 + vpdpbusd m1, m2, m7 + mova m0, m6 + vpdpbusd m0, m2, m8 + sub tlq, 2 + vpdpbusd m1, m3, m9 + vpdpbusd m0, m3, m10 + packssdw m1, m0 + mova m0, m4 + psraw m4, m1, 4 ; g0 h0 e1 f1 c2 d2 a3 b3 + packuswb m0, m4 ; e0 f0 g0 h0 c1 d1 e1 f1 a2 b2 c2 d2 __ __ a3 b3 + pshufd m2, m0, q3333 ; h0 h0 h0 h0 f1 f1 f1 f1 d2 d2 d2 d2 b3 b3 b3 b3 + vpermt2d m5, m12, m0 ; c0 d0 e0 f0 __ __ c1 d1 a0 a1 a2 a3 b0 b1 b2 b3 + vextracti32x4 [dstq+strideq*0], m5, 2 + vextracti32x4 [dstq+strideq*1], m5, 3 + lea dstq, [dstq+strideq*2] + sub hd, 2 + jg .w16_loop + cmp wd, 16 + je .ret + mova xm13, [filter_perm+16] + mova xmm3, [r6+strideq*0] + punpckhdq xmm3, [r6+strideq*1] + vpbroadcastd m2{k1}, [tlq+r5+17] ; t4 t4 t4 t4 f1 f1 f1 f1 d2 d2 d2 d2 b3 b3 b3 b3 + pinsrb xm3, xmm3, [tlq+r5+16], 7 + pshufb xm3, xm13 + vpermb m3{k2}, m11, m0 ; bf af tf __ h3 g3 f3 __ f7 e7 d7 __ db cb bb __ + mova m0, m6 + vpdpbusd m0, m2, m7 + mova m1, m6 + vpdpbusd m1, m2, m8 + kunpckbw k5, k3, k1 ; 0xff0f + lea r3, [strideq*3] + vpdpbusd m0, m3, m9 + vpdpbusd m1, m3, m10 + packssdw m0, m1 + psraw m0, 4 ; a4 b4 g1 h1 e2 f2 c3 d3 + packuswb m4, m0 ; g0 h0 a4 b4 e1 f1 g1 h1 c2 d2 e2 f2 __ __ c3 d3 + vpblendmb m1{k3}, m4, m2 ; __ t4 a4 b4 e1 f1 g1 h1 c2 d2 e2 f2 __ __ c3 d3 + vpbroadcastd ym2, [tlq+r5+21] + pshufd m2{k5}, m4, q3333 ; b4 b4 b4 b4 t5 t5 t5 t5 f2 f2 f2 f2 d3 d3 d3 d3 + vpermt2d m5, m12, m4 ; e0 f0 g0 h0 __ __ e1 f1 c0 c1 c2 c3 d0 d1 d2 d3 + vextracti32x4 [dstq+strideq*0], m5, 2 + vextracti32x4 [dstq+strideq*1], m5, 3 + punpckhqdq xmm3, [r6+r3] + pinsrb xmm3, [r6+strideq*2+15], 11 + pshufb xm3, xmm3, xm13 + vpermb m3{k2}, m11, m1 ; df cf bf __ bj aj tj __ h7 g7 f7 __ fb eb db __ + mova m4, m6 + vpdpbusd m4, m2, m7 + mova m1, m6 + vpdpbusd m1, m2, m8 + kxnord k3, k3, k4 ; 0xfffff0ff + lea r4, [strideq*5] + vpdpbusd m4, m3, m9 + vpdpbusd m1, m3, m10 + packssdw m4, m1 + psraw m4, 4 ; c4 d4 a5 b5 g2 h2 e3 f3 + packuswb m0, m4 ; a4 b4 c4 d4 g1 h1 a5 b5 e2 f2 g2 h2 __ __ e3 f3 + vpblendmw m1{k3}, m2, m0 ; a4 b4 c4 d4 __ t5 a5 b5 e2 f2 g2 h2 __ __ e3 f3 + vpbroadcastd m2, [tlq+r5+25] + pshufd m2{k3}, m0, q3333 ; d4 d4 d4 d4 b5 b5 b5 b5 t6 t6 t6 t6 f3 f3 f3 f3 + vpermt2d m5, m12, m0 ; g0 h0 a4 b4 __ __ g1 h1 e0 e1 e2 e3 f0 f1 f2 f3 + vextracti32x4 [dstq+strideq*2], m5, 2 + vextracti32x4 [dstq+r3 ], m5, 3 + punpckhqdq xmm3, [r6+r4] + pinsrb xmm3, [r6+strideq*4+15], 11 + pshufb xm3, xmm3, xm13 + vpermb m3{k2}, m11, m1 ; ff ef df __ dj cj bj __ bn an tn __ hb hb fb __ + mova m0, m6 + vpdpbusd m0, m2, m7 + mova m1, m6 + vpdpbusd m1, m2, m8 + kunpckwd k1, k1, k2 ; 0x000f0000 + vpdpbusd m0, m3, m9 + vpdpbusd m1, m3, m10 + packssdw m0, m1 + psraw m0, 4 ; e4 f4 c5 d5 a6 b6 g3 h3 + packuswb m4, m0 ; c4 d4 e4 f4 a5 b5 c5 d5 g2 h2 a6 b6 __ __ g3 h3 + vpblendmw m1{k1}, m4, m2 ; c4 d4 e4 f4 a5 b5 c5 d5 __ t6 a6 b6 __ __ g3 h3 + vpbroadcastd m2, [tlq+r5+29] + pshufd m2{k4}, m4, q3333 ; f4 f4 f4 f4 d5 d5 d5 d5 b6 b6 b6 b6 t7 t7 t7 t7 + vpermt2d m5, m12, m4 ; a4 b4 c4 d4 __ __ a5 b5 g0 g1 g2 g3 h0 h1 h2 h3 + vextracti32x4 [dstq+strideq*4], m5, 2 + vextracti32x4 [dstq+r4 ], m5, 3 + lea r0, [strideq+r3*2] +.w32_loop: + punpckhqdq xmm3, [r6+r0] + pinsrb xmm3, [r6+r3*2+15], 11 + pshufb xm3, xmm3, xm13 + vpermb m3{k2}, m11, m1 ; hf gf ff __ fj ej dj __ dn cn bn __ br ar tr __ +.w32_loop_tail: + mova m4, m6 + vpdpbusd m4, m2, m7 + mova m1, m6 + vpdpbusd m1, m2, m8 + vpdpbusd m4, m3, m9 + vpdpbusd m1, m3, m10 + packssdw m4, m1 + mova m1, m0 + psraw m0, m4, 4 ; g4 h4 e5 f5 c6 d6 a7 b7 + packuswb m1, m0 ; e4 f4 g4 h4 c5 d5 e5 f5 a6 b6 c6 d6 __ __ a7 b7 + pshufd m2, m1, q3333 ; h4 h4 h4 h4 f5 f5 f5 f5 d6 d6 d6 d6 b7 b7 b7 b7 + vpermt2d m5, m12, m1 ; c4 d4 e4 f4 __ __ c5 d5 a4 a5 a6 a7 b4 b5 b6 b7 + vextracti32x4 [r6+strideq*0+16], m5, 2 + vextracti32x4 [r6+strideq*1+16], m5, 3 + lea r6, [r6+strideq*2] + sub r5d, 2 + jg .w32_loop + vpermb m3, m11, m1 + cmp r5d, -6 + jg .w32_loop_tail +.ret: + RET +.w8: + vpermb ym3, ym11, ymm2 +.w8_loop: + vpbroadcastd ym3{k1}, [tlq-4] ; l3 l2 l1 __ b3 a3 t3 __ + mova ym0, ym6 + vpdpbusd ym0, ym2, ym7 + mova ym1, ym6 + vpdpbusd ym1, ym2, ym8 + sub tlq, 2 + vpdpbusd ym0, ym3, ym9 + vpdpbusd ym1, ym3, ym10 + mova ym3, ym5 + packssdw ym0, ym1 + psraw ym5, ym0, 4 ; c0 d0 a1 b1 + packuswb ym3, ym5 ; a0 b0 c0 d0 __ __ a1 b1 + pshufd ym2, ym3, q3333 ; d0 d0 d0 d0 b1 b1 b1 b1 + vpermb ym3, ym11, ym3 ; a0 a1 b0 b1 + movq [dstq+strideq*0], xm3 + movhps [dstq+strideq*1], xm3 + lea dstq, [dstq+strideq*2] + sub hd, 2 + jg .w8_loop + RET +.w4_loop: + vpbroadcastd xmm3, [tlq-4] ; l3 l2 l1 __ + mova xmm0, xm6 + vpdpbusd xmm0, xmm2, xm7 + mova xmm1, xm6 + vpdpbusd xmm1, xmm2, xm8 + sub tlq, 2 + vpdpbusd xmm0, xmm3, xm9 + vpdpbusd xmm1, xmm3, xm10 + packssdw xmm0, xmm1 +.w4: + psraw xmm0, 4 ; a0 b0 + packuswb xmm0, xmm0 + movd [dstq+strideq*0], xmm0 + pshufd xmm2, xmm0, q1111 ; b0 b0 b0 b0 + movd [dstq+strideq*1], xmm2 + lea dstq, [dstq+strideq*2] + sub hd, 2 + jg .w4_loop + RET + +%endif ; ARCH_X86_64 |