Remove the MMX versions of these functions and modify the SSE implementations to avoid using MMX registers.
Unified · Split
| @@ -123,54 +123,43 @@ hvar_fn · 122 unmodified lines | ||
| 123 | 123 | ; - if (%2 & 8) fills 8 bytes into xmm$next |
| 124 | 124 | ; - if (%2 & 4) fills 4 bytes into xmm$next |
| 125 | 125 | ; - if (%2 & 3) fills 1, 2 or 4 bytes in eax |
| 126 | −; on mmx, - fills mm0-7 for consecutive sets of 8 pixels | |
| 127 | −; - if (%2 & 4) fills 4 bytes into mm$next | |
| 128 | −; - if (%2 & 3) fills 1, 2 or 4 bytes in eax | |
| 129 | 126 | ; writing data out is in the same way |
| 130 | 127 | %macro READ_NUM_BYTES 2 |
| 131 | 128 | %assign %%off 0 ; offset in source buffer |
| 132 | −%assign %%mmx_idx 0 ; mmx register index | |
| 133 | 129 | %assign %%xmm_idx 0 ; xmm register index |
| 134 | 130 | |
| 135 | 131 | %rep %2/mmsize |
| 136 | −%if mmsize == 16 | |
| 137 | 132 | movu xmm %+ %%xmm_idx, [srcq+%%off] |
| 138 | 133 | %assign %%xmm_idx %%xmm_idx+1 |
| 139 | −%else ; mmx | |
| 140 | − movu mm %+ %%mmx_idx, [srcq+%%off] | |
| 141 | −%assign %%mmx_idx %%mmx_idx+1 | |
| 142 | −%endif | |
| 143 | 134 | %assign %%off %%off+mmsize |
| 144 | 135 | %endrep ; %2/mmsize |
| 145 | 136 | |
| 146 | −%if mmsize == 16 | |
| 147 | 137 | %if (%2-%%off) >= 8 |
| 148 | 138 | %if %2 > 16 && (%2-%%off) > 8 |
| 149 | 139 | movu xmm %+ %%xmm_idx, [srcq+%2-16] |
| 150 | 140 | %assign %%xmm_idx %%xmm_idx+1 |
| 151 | 141 | %assign %%off %2 |
| 152 | 142 | %else |
| 153 | − movq mm %+ %%mmx_idx, [srcq+%%off] | |
| 154 | −%assign %%mmx_idx %%mmx_idx+1 | |
| 143 | + movq xmm %+ %%xmm_idx, [srcq+%%off] | |
| 144 | +%assign %%xmm_idx %%xmm_idx+1 | |
| 155 | 145 | %assign %%off %%off+8 |
| 156 | 146 | %endif |
| 157 | 147 | %endif ; (%2-%%off) >= 8 |
| 158 | −%endif | |
| 159 | 148 | |
| 160 | 149 | %if (%2-%%off) >= 4 |
| 161 | 150 | %if %2 > 8 && (%2-%%off) > 4 |
| 162 | − movq mm %+ %%mmx_idx, [srcq+%2-8] | |
| 151 | + movq xmm %+ %%xmm_idx, [srcq+%2-8] | |
| 163 | 152 | %assign %%off %2 |
| 164 | 153 | %else |
| 165 | − movd mm %+ %%mmx_idx, [srcq+%%off] | |
| 154 | + movd xmm %+ %%xmm_idx, [srcq+%%off] | |
| 166 | 155 | %assign %%off %%off+4 |
| 167 | 156 | %endif |
| 168 | −%assign %%mmx_idx %%mmx_idx+1 | |
| 157 | +%assign %%xmm_idx %%xmm_idx+1 | |
| 169 | 158 | %endif ; (%2-%%off) >= 4 |
| 170 | 159 | |
| 171 | 160 | %if (%2-%%off) >= 1 |
| 172 | 161 | %if %2 >= 4 |
| 173 | − movd mm %+ %%mmx_idx, [srcq+%2-4] | |
| 162 | + movd xmm %+ %%xmm_idx, [srcq+%2-4] | |
| 174 | 163 | %elif (%2-%%off) == 1 |
| 175 | 164 | mov valb, [srcq+%2-1] |
| 176 | 165 | %elif (%2-%%off) == 2 |
| @@ -185,48 +174,40 @@ hvar_fn · 8 unmodified lines | ||
| 185 | 174 | |
| 186 | 175 | %macro WRITE_NUM_BYTES 2 |
| 187 | 176 | %assign %%off 0 ; offset in destination buffer |
| 188 | −%assign %%mmx_idx 0 ; mmx register index | |
| 189 | 177 | %assign %%xmm_idx 0 ; xmm register index |
| 190 | 178 | |
| 191 | 179 | %rep %2/mmsize |
| 192 | −%if mmsize == 16 | |
| 193 | 180 | movu [dstq+%%off], xmm %+ %%xmm_idx |
| 194 | 181 | %assign %%xmm_idx %%xmm_idx+1 |
| 195 | −%else ; mmx | |
| 196 | − movu [dstq+%%off], mm %+ %%mmx_idx | |
| 197 | −%assign %%mmx_idx %%mmx_idx+1 | |
| 198 | −%endif | |
| 199 | 182 | %assign %%off %%off+mmsize |
| 200 | 183 | %endrep ; %2/mmsize |
| 201 | 184 | |
| 202 | −%if mmsize == 16 | |
| 203 | 185 | %if (%2-%%off) >= 8 |
| 204 | 186 | %if %2 > 16 && (%2-%%off) > 8 |
| 205 | 187 | movu [dstq+%2-16], xmm %+ %%xmm_idx |
| 206 | 188 | %assign %%xmm_idx %%xmm_idx+1 |
| 207 | 189 | %assign %%off %2 |
| 208 | 190 | %else |
| 209 | − movq [dstq+%%off], mm %+ %%mmx_idx | |
| 210 | −%assign %%mmx_idx %%mmx_idx+1 | |
| 191 | + movq [dstq+%%off], xmm %+ %%xmm_idx | |
| 192 | +%assign %%xmm_idx %%xmm_idx+1 | |
| 211 | 193 | %assign %%off %%off+8 |
| 212 | 194 | %endif |
| 213 | 195 | %endif ; (%2-%%off) >= 8 |
| 214 | −%endif | |
| 215 | 196 | |
| 216 | 197 | %if (%2-%%off) >= 4 |
| 217 | 198 | %if %2 > 8 && (%2-%%off) > 4 |
| 218 | − movq [dstq+%2-8], mm %+ %%mmx_idx | |
| 199 | + movq [dstq+%2-8], xmm %+ %%xmm_idx | |
| 219 | 200 | %assign %%off %2 |
| 220 | 201 | %else |
| 221 | − movd [dstq+%%off], mm %+ %%mmx_idx | |
| 202 | + movd [dstq+%%off], xmm %+ %%xmm_idx | |
| 222 | 203 | %assign %%off %%off+4 |
| 223 | 204 | %endif |
| 224 | −%assign %%mmx_idx %%mmx_idx+1 | |
| 205 | +%assign %%xmm_idx %%xmm_idx+1 | |
| 225 | 206 | %endif ; (%2-%%off) >= 4 |
| 226 | 207 | |
| 227 | 208 | %if (%2-%%off) >= 1 |
| 228 | 209 | %if %2 >= 4 |
| 229 | − movd [dstq+%2-4], mm %+ %%mmx_idx | |
| 210 | + movd [dstq+%2-4], xmm %+ %%xmm_idx | |
| 230 | 211 | %elif (%2-%%off) == 1 |
| 231 | 212 | mov [dstq+%2-1], valb |
| 232 | 213 | %elif (%2-%%off) == 2 |
| @@ -318,11 +299,8 @@ cglobal emu_edge_vfix %+ %%n, 1, 5, 1, dst, src, start_y, end_y, bh · 85 unmodified lines | ||
| 318 | 299 | %endrep ; 1+%2-%1 |
| 319 | 300 | %endmacro ; VERTICAL_EXTEND |
| 320 | 301 | |
| 321 | −INIT_MMX mmx | |
| 322 | −VERTICAL_EXTEND 1, 15 | |
| 323 | − | |
| 324 | −INIT_XMM sse | |
| 325 | −VERTICAL_EXTEND 16, 22 | |
| 302 | +INIT_XMM sse2 | |
| 303 | +VERTICAL_EXTEND 1, 22 | |
| 326 | 304 | |
| 327 | 305 | ; left/right (horizontal) fast extend functions |
| 328 | 306 | ; these are essentially identical to the vertical extend ones above, |
| @@ -337,11 +315,7 @@ VERTICAL_EXTEND 16, 22 · 8 unmodified lines | ||
| 337 | 315 | imul vald, 0x01010101 |
| 338 | 316 | %if %1 >= 8 |
| 339 | 317 | movd m0, vald |
| 340 | −%if mmsize == 16 | |
| 341 | 318 | pshufd m0, m0, q0000 |
| 342 | −%else | |
| 343 | − punpckldq m0, m0 | |
| 344 | −%endif ; mmsize == 16 | |
| 345 | 319 | %endif ; %1 > 16 |
| 346 | 320 | %endif ; avx2 |
| 347 | 321 | %endmacro ; READ_V_PIXEL |
| @@ -356,7 +330,6 @@ VERTICAL_EXTEND 16, 22 · 8 unmodified lines | ||
| 356 | 330 | %assign %%off %%off+mmsize |
| 357 | 331 | %endrep ; %1/mmsize |
| 358 | 332 | |
| 359 | −%if mmsize == 16 | |
| 360 | 333 | %if %1-%%off >= 8 |
| 361 | 334 | %if %1 > 16 && %1-%%off > 8 |
| 362 | 335 | movu [%2+%1-16], m0 |
| @@ -366,7 +339,6 @@ VERTICAL_EXTEND 16, 22 · 3 unmodified lines | ||
| 366 | 339 | %assign %%off %%off+8 |
| 367 | 340 | %endif |
| 368 | 341 | %endif ; %1-%%off >= 8 |
| 369 | −%endif ; mmsize == 16 | |
| 370 | 342 | |
| 371 | 343 | %if %1-%%off >= 4 |
| 372 | 344 | %if %1 > 8 && %1-%%off > 4 |
| @@ -415,11 +387,8 @@ cglobal emu_edge_hfix %+ %%n, 4, 5, 1, dst, dst_stride, start_x, bh, val · 42 unmodified lines | ||
| 415 | 387 | %endrep ; 1+(%2-%1)/2 |
| 416 | 388 | %endmacro ; H_EXTEND |
| 417 | 389 | |
| 418 | −INIT_MMX mmx | |
| 419 | −H_EXTEND 2, 14 | |
| 420 | − | |
| 421 | 390 | INIT_XMM sse2 |
| 422 | −H_EXTEND 16, 22 | |
| 391 | +H_EXTEND 2, 22 | |
| 423 | 392 | |
| 424 | 393 | %if HAVE_AVX2_EXTERNAL |
| 425 | 394 | INIT_XMM avx2 |
| @@ -37,37 +37,37 @@ typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride, · 36 unmodified lines | ||
| 37 | 37 | x86_reg start_y, x86_reg end_y, x86_reg bh, |
| 38 | 38 | x86_reg w); |
| 39 | 39 | |
| 40 | −extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx; | |
| 41 | −extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx; | |
| 42 | −extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx; | |
| 43 | −extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx; | |
| 44 | −extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx; | |
| 45 | −extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx; | |
| 46 | −extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx; | |
| 47 | −extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx; | |
| 48 | −extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx; | |
| 49 | −extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx; | |
| 50 | −extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx; | |
| 51 | −extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx; | |
| 52 | −extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx; | |
| 53 | −extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx; | |
| 54 | −extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx; | |
| 55 | −extern emu_edge_vfix_func ff_emu_edge_vfix16_sse; | |
| 56 | −extern emu_edge_vfix_func ff_emu_edge_vfix17_sse; | |
| 57 | −extern emu_edge_vfix_func ff_emu_edge_vfix18_sse; | |
| 58 | −extern emu_edge_vfix_func ff_emu_edge_vfix19_sse; | |
| 59 | −extern emu_edge_vfix_func ff_emu_edge_vfix20_sse; | |
| 60 | −extern emu_edge_vfix_func ff_emu_edge_vfix21_sse; | |
| 61 | −extern emu_edge_vfix_func ff_emu_edge_vfix22_sse; | |
| 62 | −static emu_edge_vfix_func * const vfixtbl_sse[22] = { | |
| 63 | − ff_emu_edge_vfix1_mmx, ff_emu_edge_vfix2_mmx, ff_emu_edge_vfix3_mmx, | |
| 64 | − ff_emu_edge_vfix4_mmx, ff_emu_edge_vfix5_mmx, ff_emu_edge_vfix6_mmx, | |
| 65 | − ff_emu_edge_vfix7_mmx, ff_emu_edge_vfix8_mmx, ff_emu_edge_vfix9_mmx, | |
| 66 | − ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx, | |
| 67 | − ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx, | |
| 68 | − ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse, | |
| 69 | − ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse, | |
| 70 | − ff_emu_edge_vfix22_sse | |
| 40 | +extern emu_edge_vfix_func ff_emu_edge_vfix1_sse2; | |
| 41 | +extern emu_edge_vfix_func ff_emu_edge_vfix2_sse2; | |
| 42 | +extern emu_edge_vfix_func ff_emu_edge_vfix3_sse2; | |
| 43 | +extern emu_edge_vfix_func ff_emu_edge_vfix4_sse2; | |
| 44 | +extern emu_edge_vfix_func ff_emu_edge_vfix5_sse2; | |
| 45 | +extern emu_edge_vfix_func ff_emu_edge_vfix6_sse2; | |
| 46 | +extern emu_edge_vfix_func ff_emu_edge_vfix7_sse2; | |
| 47 | +extern emu_edge_vfix_func ff_emu_edge_vfix8_sse2; | |
| 48 | +extern emu_edge_vfix_func ff_emu_edge_vfix9_sse2; | |
| 49 | +extern emu_edge_vfix_func ff_emu_edge_vfix10_sse2; | |
| 50 | +extern emu_edge_vfix_func ff_emu_edge_vfix11_sse2; | |
| 51 | +extern emu_edge_vfix_func ff_emu_edge_vfix12_sse2; | |
| 52 | +extern emu_edge_vfix_func ff_emu_edge_vfix13_sse2; | |
| 53 | +extern emu_edge_vfix_func ff_emu_edge_vfix14_sse2; | |
| 54 | +extern emu_edge_vfix_func ff_emu_edge_vfix15_sse2; | |
| 55 | +extern emu_edge_vfix_func ff_emu_edge_vfix16_sse2; | |
| 56 | +extern emu_edge_vfix_func ff_emu_edge_vfix17_sse2; | |
| 57 | +extern emu_edge_vfix_func ff_emu_edge_vfix18_sse2; | |
| 58 | +extern emu_edge_vfix_func ff_emu_edge_vfix19_sse2; | |
| 59 | +extern emu_edge_vfix_func ff_emu_edge_vfix20_sse2; | |
| 60 | +extern emu_edge_vfix_func ff_emu_edge_vfix21_sse2; | |
| 61 | +extern emu_edge_vfix_func ff_emu_edge_vfix22_sse2; | |
| 62 | +static emu_edge_vfix_func * const vfixtbl_sse2[22] = { | |
| 63 | + ff_emu_edge_vfix1_sse2, ff_emu_edge_vfix2_sse2, ff_emu_edge_vfix3_sse2, | |
| 64 | + ff_emu_edge_vfix4_sse2, ff_emu_edge_vfix5_sse2, ff_emu_edge_vfix6_sse2, | |
| 65 | + ff_emu_edge_vfix7_sse2, ff_emu_edge_vfix8_sse2, ff_emu_edge_vfix9_sse2, | |
| 66 | + ff_emu_edge_vfix10_sse2, ff_emu_edge_vfix11_sse2, ff_emu_edge_vfix12_sse2, | |
| 67 | + ff_emu_edge_vfix13_sse2, ff_emu_edge_vfix14_sse2, ff_emu_edge_vfix15_sse2, | |
| 68 | + ff_emu_edge_vfix16_sse2, ff_emu_edge_vfix17_sse2, ff_emu_edge_vfix18_sse2, | |
| 69 | + ff_emu_edge_vfix19_sse2, ff_emu_edge_vfix20_sse2, ff_emu_edge_vfix21_sse2, | |
| 70 | + ff_emu_edge_vfix22_sse2 | |
| 71 | 71 | }; |
| 72 | 72 | extern emu_edge_vvar_func ff_emu_edge_vvar_sse; |
| 73 | 73 | |
| @@ -76,21 +76,21 @@ typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride, · 2 unmodified lines | ||
| 76 | 76 | typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride, |
| 77 | 77 | x86_reg start_x, x86_reg n_words, x86_reg bh); |
| 78 | 78 | |
| 79 | −extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx; | |
| 80 | −extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx; | |
| 81 | −extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx; | |
| 82 | −extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx; | |
| 83 | −extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx; | |
| 84 | −extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx; | |
| 85 | −extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx; | |
| 79 | +extern emu_edge_hfix_func ff_emu_edge_hfix2_sse2; | |
| 80 | +extern emu_edge_hfix_func ff_emu_edge_hfix4_sse2; | |
| 81 | +extern emu_edge_hfix_func ff_emu_edge_hfix6_sse2; | |
| 82 | +extern emu_edge_hfix_func ff_emu_edge_hfix8_sse2; | |
| 83 | +extern emu_edge_hfix_func ff_emu_edge_hfix10_sse2; | |
| 84 | +extern emu_edge_hfix_func ff_emu_edge_hfix12_sse2; | |
| 85 | +extern emu_edge_hfix_func ff_emu_edge_hfix14_sse2; | |
| 86 | 86 | extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2; |
| 87 | 87 | extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2; |
| 88 | 88 | extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2; |
| 89 | 89 | extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2; |
| 90 | 90 | static emu_edge_hfix_func * const hfixtbl_sse2[11] = { |
| 91 | − ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, | |
| 92 | − ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx, | |
| 93 | − ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2, | |
| 91 | + ff_emu_edge_hfix2_sse2, ff_emu_edge_hfix4_sse2, ff_emu_edge_hfix6_sse2, | |
| 92 | + ff_emu_edge_hfix8_sse2, ff_emu_edge_hfix10_sse2, ff_emu_edge_hfix12_sse2, | |
| 93 | + ff_emu_edge_hfix14_sse2, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2, | |
| 94 | 94 | ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2 |
| 95 | 95 | }; |
| 96 | 96 | extern emu_edge_hvar_func ff_emu_edge_hvar_sse2; |
| @@ -104,7 +104,7 @@ extern emu_edge_hfix_func ff_emu_edge_hfix18_avx2; · 7 unmodified lines | ||
| 104 | 104 | extern emu_edge_hfix_func ff_emu_edge_hfix20_avx2; |
| 105 | 105 | extern emu_edge_hfix_func ff_emu_edge_hfix22_avx2; |
| 106 | 106 | static emu_edge_hfix_func * const hfixtbl_avx2[11] = { |
| 107 | − ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx, | |
| 107 | + ff_emu_edge_hfix2_sse2, ff_emu_edge_hfix4_sse2, ff_emu_edge_hfix6_sse2, | |
| 108 | 108 | ff_emu_edge_hfix8_avx2, ff_emu_edge_hfix10_avx2, ff_emu_edge_hfix12_avx2, |
| 109 | 109 | ff_emu_edge_hfix14_avx2, ff_emu_edge_hfix16_avx2, ff_emu_edge_hfix18_avx2, |
| 110 | 110 | ff_emu_edge_hfix20_avx2, ff_emu_edge_hfix22_avx2 |
| @@ -196,7 +196,7 @@ static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src, · 85 unmodified lines | ||
| 196 | 196 | int h) |
| 197 | 197 | { |
| 198 | 198 | emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, |
| 199 | − src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse, | |
| 199 | + src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse, | |
| 200 | 200 | hfixtbl_sse2, &ff_emu_edge_hvar_sse2); |
| 201 | 201 | } |
| 202 | 202 | |
| @@ -209,7 +209,7 @@ static av_noinline void emulated_edge_mc_avx2(uint8_t *buf, const uint8_t *src, · 6 unmodified lines | ||
| 209 | 209 | int h) |
| 210 | 210 | { |
| 211 | 211 | emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h, |
| 212 | − src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse, | |
| 212 | + src_x, src_y, w, h, vfixtbl_sse2, &ff_emu_edge_vvar_sse, | |
| 213 | 213 | hfixtbl_avx2, &ff_emu_edge_hvar_avx2); |
| 214 | 214 | } |
| 215 | 215 | #endif /* HAVE_AVX2_EXTERNAL */ |