1; 2; jchuff-sse2.asm - Huffman entropy encoding (SSE2) 3; 4; Copyright (C) 2009-2011, 2014-2017, D. R. Commander. 5; Copyright (C) 2015, Matthieu Darbois. 6; 7; Based on the x86 SIMD extension for IJG JPEG library 8; Copyright (C) 1999-2006, MIYASAKA Masaru. 9; For conditions of distribution and use, see copyright notice in jsimdext.inc 10; 11; This file should be assembled with NASM (Netwide Assembler), 12; can *not* be assembled with Microsoft's MASM or any compatible 13; assembler (including Borland's Turbo Assembler). 14; NASM is available from http://nasm.sourceforge.net/ or 15; http://sourceforge.net/project/showfiles.php?group_id=6208 16; 17; This file contains an SSE2 implementation for Huffman coding of one block. 18; The following code is based directly on jchuff.c; see jchuff.c for more 19; details. 20 21%include "jsimdext.inc" 22 23; -------------------------------------------------------------------------- 24 SECTION SEG_CONST 25 26 alignz 32 27 GLOBAL_DATA(jconst_huff_encode_one_block) 28 29EXTN(jconst_huff_encode_one_block): 30 31%include "jpeg_nbits_table.inc" 32 33 alignz 32 34 35; -------------------------------------------------------------------------- 36 SECTION SEG_TEXT 37 BITS 32 38 39; These macros perform the same task as the emit_bits() function in the 40; original libjpeg code. In addition to reducing overhead by explicitly 41; inlining the code, additional performance is achieved by taking into 42; account the size of the bit buffer and waiting until it is almost full 43; before emptying it. This mostly benefits 64-bit platforms, since 6 44; bytes can be stored in a 64-bit bit buffer before it has to be emptied. 45 46%macro EMIT_BYTE 0 47 sub put_bits, 8 ; put_bits -= 8; 48 mov edx, put_buffer 49 mov ecx, put_bits 50 shr edx, cl ; c = (JOCTET)GETJOCTET(put_buffer >> put_bits); 51 mov byte [eax], dl ; *buffer++ = c; 52 add eax, 1 53 cmp dl, 0xFF ; need to stuff a zero byte? 54 jne %%.EMIT_BYTE_END 55 mov byte [eax], 0 ; *buffer++ = 0; 56 add eax, 1 57%%.EMIT_BYTE_END: 58%endmacro 59 60%macro PUT_BITS 1 61 add put_bits, ecx ; put_bits += size; 62 shl put_buffer, cl ; put_buffer = (put_buffer << size); 63 or put_buffer, %1 64%endmacro 65 66%macro CHECKBUF15 0 67 cmp put_bits, 16 ; if (put_bits > 31) { 68 jl %%.CHECKBUF15_END 69 mov eax, POINTER [esp+buffer] 70 EMIT_BYTE 71 EMIT_BYTE 72 mov POINTER [esp+buffer], eax 73%%.CHECKBUF15_END: 74%endmacro 75 76%macro EMIT_BITS 1 77 PUT_BITS %1 78 CHECKBUF15 79%endmacro 80 81%macro kloop_prepare 37 ;(ko, jno0, ..., jno31, xmm0, xmm1, xmm2, xmm3) 82 pxor xmm4, xmm4 ; __m128i neg = _mm_setzero_si128(); 83 pxor xmm5, xmm5 ; __m128i neg = _mm_setzero_si128(); 84 pxor xmm6, xmm6 ; __m128i neg = _mm_setzero_si128(); 85 pxor xmm7, xmm7 ; __m128i neg = _mm_setzero_si128(); 86 pinsrw %34, word [esi + %2 * SIZEOF_WORD], 0 ; xmm_shadow[0] = block[jno0]; 87 pinsrw %35, word [esi + %10 * SIZEOF_WORD], 0 ; xmm_shadow[8] = block[jno8]; 88 pinsrw %36, word [esi + %18 * SIZEOF_WORD], 0 ; xmm_shadow[16] = block[jno16]; 89 pinsrw %37, word [esi + %26 * SIZEOF_WORD], 0 ; xmm_shadow[24] = block[jno24]; 90 pinsrw %34, word [esi + %3 * SIZEOF_WORD], 1 ; xmm_shadow[1] = block[jno1]; 91 pinsrw %35, word [esi + %11 * SIZEOF_WORD], 1 ; xmm_shadow[9] = block[jno9]; 92 pinsrw %36, word [esi + %19 * SIZEOF_WORD], 1 ; xmm_shadow[17] = block[jno17]; 93 pinsrw %37, word [esi + %27 * SIZEOF_WORD], 1 ; xmm_shadow[25] = block[jno25]; 94 pinsrw %34, word [esi + %4 * SIZEOF_WORD], 2 ; xmm_shadow[2] = block[jno2]; 95 pinsrw %35, word [esi + %12 * SIZEOF_WORD], 2 ; xmm_shadow[10] = block[jno10]; 96 pinsrw %36, word [esi + %20 * SIZEOF_WORD], 2 ; xmm_shadow[18] = block[jno18]; 97 pinsrw %37, word [esi + %28 * SIZEOF_WORD], 2 ; xmm_shadow[26] = block[jno26]; 98 pinsrw %34, word [esi + %5 * SIZEOF_WORD], 3 ; xmm_shadow[3] = block[jno3]; 99 pinsrw %35, word [esi + %13 * SIZEOF_WORD], 3 ; xmm_shadow[11] = block[jno11]; 100 pinsrw %36, word [esi + %21 * SIZEOF_WORD], 3 ; xmm_shadow[19] = block[jno19]; 101 pinsrw %37, word [esi + %29 * SIZEOF_WORD], 3 ; xmm_shadow[27] = block[jno27]; 102 pinsrw %34, word [esi + %6 * SIZEOF_WORD], 4 ; xmm_shadow[4] = block[jno4]; 103 pinsrw %35, word [esi + %14 * SIZEOF_WORD], 4 ; xmm_shadow[12] = block[jno12]; 104 pinsrw %36, word [esi + %22 * SIZEOF_WORD], 4 ; xmm_shadow[20] = block[jno20]; 105 pinsrw %37, word [esi + %30 * SIZEOF_WORD], 4 ; xmm_shadow[28] = block[jno28]; 106 pinsrw %34, word [esi + %7 * SIZEOF_WORD], 5 ; xmm_shadow[5] = block[jno5]; 107 pinsrw %35, word [esi + %15 * SIZEOF_WORD], 5 ; xmm_shadow[13] = block[jno13]; 108 pinsrw %36, word [esi + %23 * SIZEOF_WORD], 5 ; xmm_shadow[21] = block[jno21]; 109 pinsrw %37, word [esi + %31 * SIZEOF_WORD], 5 ; xmm_shadow[29] = block[jno29]; 110 pinsrw %34, word [esi + %8 * SIZEOF_WORD], 6 ; xmm_shadow[6] = block[jno6]; 111 pinsrw %35, word [esi + %16 * SIZEOF_WORD], 6 ; xmm_shadow[14] = block[jno14]; 112 pinsrw %36, word [esi + %24 * SIZEOF_WORD], 6 ; xmm_shadow[22] = block[jno22]; 113 pinsrw %37, word [esi + %32 * SIZEOF_WORD], 6 ; xmm_shadow[30] = block[jno30]; 114 pinsrw %34, word [esi + %9 * SIZEOF_WORD], 7 ; xmm_shadow[7] = block[jno7]; 115 pinsrw %35, word [esi + %17 * SIZEOF_WORD], 7 ; xmm_shadow[15] = block[jno15]; 116 pinsrw %36, word [esi + %25 * SIZEOF_WORD], 7 ; xmm_shadow[23] = block[jno23]; 117%if %1 != 32 118 pinsrw %37, word [esi + %33 * SIZEOF_WORD], 7 ; xmm_shadow[31] = block[jno31]; 119%else 120 pinsrw %37, ecx, 7 ; xmm_shadow[31] = block[jno31]; 121%endif 122 pcmpgtw xmm4, %34 ; neg = _mm_cmpgt_epi16(neg, x1); 123 pcmpgtw xmm5, %35 ; neg = _mm_cmpgt_epi16(neg, x1); 124 pcmpgtw xmm6, %36 ; neg = _mm_cmpgt_epi16(neg, x1); 125 pcmpgtw xmm7, %37 ; neg = _mm_cmpgt_epi16(neg, x1); 126 paddw %34, xmm4 ; x1 = _mm_add_epi16(x1, neg); 127 paddw %35, xmm5 ; x1 = _mm_add_epi16(x1, neg); 128 paddw %36, xmm6 ; x1 = _mm_add_epi16(x1, neg); 129 paddw %37, xmm7 ; x1 = _mm_add_epi16(x1, neg); 130 pxor %34, xmm4 ; x1 = _mm_xor_si128(x1, neg); 131 pxor %35, xmm5 ; x1 = _mm_xor_si128(x1, neg); 132 pxor %36, xmm6 ; x1 = _mm_xor_si128(x1, neg); 133 pxor %37, xmm7 ; x1 = _mm_xor_si128(x1, neg); 134 pxor xmm4, %34 ; neg = _mm_xor_si128(neg, x1); 135 pxor xmm5, %35 ; neg = _mm_xor_si128(neg, x1); 136 pxor xmm6, %36 ; neg = _mm_xor_si128(neg, x1); 137 pxor xmm7, %37 ; neg = _mm_xor_si128(neg, x1); 138 movdqa XMMWORD [esp + t1 + %1 * SIZEOF_WORD], %34 ; _mm_storeu_si128((__m128i *)(t1 + ko), x1); 139 movdqa XMMWORD [esp + t1 + (%1 + 8) * SIZEOF_WORD], %35 ; _mm_storeu_si128((__m128i *)(t1 + ko + 8), x1); 140 movdqa XMMWORD [esp + t1 + (%1 + 16) * SIZEOF_WORD], %36 ; _mm_storeu_si128((__m128i *)(t1 + ko + 16), x1); 141 movdqa XMMWORD [esp + t1 + (%1 + 24) * SIZEOF_WORD], %37 ; _mm_storeu_si128((__m128i *)(t1 + ko + 24), x1); 142 movdqa XMMWORD [esp + t2 + %1 * SIZEOF_WORD], xmm4 ; _mm_storeu_si128((__m128i *)(t2 + ko), neg); 143 movdqa XMMWORD [esp + t2 + (%1 + 8) * SIZEOF_WORD], xmm5 ; _mm_storeu_si128((__m128i *)(t2 + ko + 8), neg); 144 movdqa XMMWORD [esp + t2 + (%1 + 16) * SIZEOF_WORD], xmm6 ; _mm_storeu_si128((__m128i *)(t2 + ko + 16), neg); 145 movdqa XMMWORD [esp + t2 + (%1 + 24) * SIZEOF_WORD], xmm7 ; _mm_storeu_si128((__m128i *)(t2 + ko + 24), neg); 146%endmacro 147 148; 149; Encode a single block's worth of coefficients. 150; 151; GLOBAL(JOCTET *) 152; jsimd_huff_encode_one_block_sse2(working_state *state, JOCTET *buffer, 153; JCOEFPTR block, int last_dc_val, 154; c_derived_tbl *dctbl, c_derived_tbl *actbl) 155; 156 157; eax + 8 = working_state *state 158; eax + 12 = JOCTET *buffer 159; eax + 16 = JCOEFPTR block 160; eax + 20 = int last_dc_val 161; eax + 24 = c_derived_tbl *dctbl 162; eax + 28 = c_derived_tbl *actbl 163 164%define pad 6 * SIZEOF_DWORD ; Align to 16 bytes 165%define t1 pad 166%define t2 t1 + (DCTSIZE2 * SIZEOF_WORD) 167%define block t2 + (DCTSIZE2 * SIZEOF_WORD) 168%define actbl block + SIZEOF_DWORD 169%define buffer actbl + SIZEOF_DWORD 170%define temp buffer + SIZEOF_DWORD 171%define temp2 temp + SIZEOF_DWORD 172%define temp3 temp2 + SIZEOF_DWORD 173%define temp4 temp3 + SIZEOF_DWORD 174%define temp5 temp4 + SIZEOF_DWORD 175%define gotptr temp5 + SIZEOF_DWORD ; void *gotptr 176%define put_buffer ebx 177%define put_bits edi 178 179 align 32 180 GLOBAL_FUNCTION(jsimd_huff_encode_one_block_sse2) 181 182EXTN(jsimd_huff_encode_one_block_sse2): 183 push ebp 184 mov eax, esp ; eax = original ebp 185 sub esp, byte 4 186 and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits 187 mov [esp], eax 188 mov ebp, esp ; ebp = aligned ebp 189 sub esp, temp5+9*SIZEOF_DWORD-pad 190 push ebx 191 push ecx 192; push edx ; need not be preserved 193 push esi 194 push edi 195 push ebp 196 197 mov esi, POINTER [eax+8] ; (working_state *state) 198 mov put_buffer, dword [esi+8] ; put_buffer = state->cur.put_buffer; 199 mov put_bits, dword [esi+12] ; put_bits = state->cur.put_bits; 200 push esi ; esi is now scratch 201 202 get_GOT edx ; get GOT address 203 movpic POINTER [esp+gotptr], edx ; save GOT address 204 205 mov ecx, POINTER [eax+28] 206 mov edx, POINTER [eax+16] 207 mov esi, POINTER [eax+12] 208 mov POINTER [esp+actbl], ecx 209 mov POINTER [esp+block], edx 210 mov POINTER [esp+buffer], esi 211 212 ; Encode the DC coefficient difference per section F.1.2.1 213 mov esi, POINTER [esp+block] ; block 214 movsx ecx, word [esi] ; temp = temp2 = block[0] - last_dc_val; 215 sub ecx, dword [eax+20] 216 mov esi, ecx 217 218 ; This is a well-known technique for obtaining the absolute value 219 ; with out a branch. It is derived from an assembly language technique 220 ; presented in "How to Optimize for the Pentium Processors", 221 ; Copyright (c) 1996, 1997 by Agner Fog. 222 mov edx, ecx 223 sar edx, 31 ; temp3 = temp >> (CHAR_BIT * sizeof(int) - 1); 224 xor ecx, edx ; temp ^= temp3; 225 sub ecx, edx ; temp -= temp3; 226 227 ; For a negative input, want temp2 = bitwise complement of abs(input) 228 ; This code assumes we are on a two's complement machine 229 add esi, edx ; temp2 += temp3; 230 mov dword [esp+temp], esi ; backup temp2 in temp 231 232 ; Find the number of bits needed for the magnitude of the coefficient 233 movpic ebp, POINTER [esp+gotptr] ; load GOT address (ebp) 234 movzx edx, byte [GOTOFF(ebp, jpeg_nbits_table + ecx)] ; nbits = JPEG_NBITS(temp); 235 mov dword [esp+temp2], edx ; backup nbits in temp2 236 237 ; Emit the Huffman-coded symbol for the number of bits 238 mov ebp, POINTER [eax+24] ; After this point, arguments are not accessible anymore 239 mov eax, INT [ebp + edx * 4] ; code = dctbl->ehufco[nbits]; 240 movzx ecx, byte [ebp + edx + 1024] ; size = dctbl->ehufsi[nbits]; 241 EMIT_BITS eax ; EMIT_BITS(code, size) 242 243 mov ecx, dword [esp+temp2] ; restore nbits 244 245 ; Mask off any extra bits in code 246 mov eax, 1 247 shl eax, cl 248 dec eax 249 and eax, dword [esp+temp] ; temp2 &= (((JLONG)1)<<nbits) - 1; 250 251 ; Emit that number of bits of the value, if positive, 252 ; or the complement of its magnitude, if negative. 253 EMIT_BITS eax ; EMIT_BITS(temp2, nbits) 254 255 ; Prepare data 256 xor ecx, ecx 257 mov esi, POINTER [esp+block] 258 kloop_prepare 0, 1, 8, 16, 9, 2, 3, 10, 17, 24, 32, 25, \ 259 18, 11, 4, 5, 12, 19, 26, 33, 40, 48, 41, 34, \ 260 27, 20, 13, 6, 7, 14, 21, 28, 35, \ 261 xmm0, xmm1, xmm2, xmm3 262 kloop_prepare 32, 42, 49, 56, 57, 50, 43, 36, 29, 22, 15, 23, \ 263 30, 37, 44, 51, 58, 59, 52, 45, 38, 31, 39, 46, \ 264 53, 60, 61, 54, 47, 55, 62, 63, 63, \ 265 xmm0, xmm1, xmm2, xmm3 266 267 pxor xmm7, xmm7 268 movdqa xmm0, XMMWORD [esp + t1 + 0 * SIZEOF_WORD] ; __m128i tmp0 = _mm_loadu_si128((__m128i *)(t1 + 0)); 269 movdqa xmm1, XMMWORD [esp + t1 + 8 * SIZEOF_WORD] ; __m128i tmp1 = _mm_loadu_si128((__m128i *)(t1 + 8)); 270 movdqa xmm2, XMMWORD [esp + t1 + 16 * SIZEOF_WORD] ; __m128i tmp2 = _mm_loadu_si128((__m128i *)(t1 + 16)); 271 movdqa xmm3, XMMWORD [esp + t1 + 24 * SIZEOF_WORD] ; __m128i tmp3 = _mm_loadu_si128((__m128i *)(t1 + 24)); 272 pcmpeqw xmm0, xmm7 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero); 273 pcmpeqw xmm1, xmm7 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero); 274 pcmpeqw xmm2, xmm7 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero); 275 pcmpeqw xmm3, xmm7 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero); 276 packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1); 277 packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3); 278 pmovmskb edx, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0; 279 pmovmskb ecx, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16; 280 shl ecx, 16 281 or edx, ecx 282 not edx ; index = ~index; 283 284 lea esi, [esp+t1] 285 mov ebp, POINTER [esp+actbl] ; ebp = actbl 286 287.BLOOP: 288 bsf ecx, edx ; r = __builtin_ctzl(index); 289 jz near .ELOOP 290 lea esi, [esi+ecx*2] ; k += r; 291 shr edx, cl ; index >>= r; 292 mov dword [esp+temp3], edx 293.BRLOOP: 294 cmp ecx, 16 ; while (r > 15) { 295 jl near .ERLOOP 296 sub ecx, 16 ; r -= 16; 297 mov dword [esp+temp], ecx 298 mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0]; 299 movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0]; 300 EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0) 301 mov ecx, dword [esp+temp] 302 jmp .BRLOOP 303.ERLOOP: 304 movsx eax, word [esi] ; temp = t1[k]; 305 movpic edx, POINTER [esp+gotptr] ; load GOT address (edx) 306 movzx eax, byte [GOTOFF(edx, jpeg_nbits_table + eax)] ; nbits = JPEG_NBITS(temp); 307 mov dword [esp+temp2], eax 308 ; Emit Huffman symbol for run length / number of bits 309 shl ecx, 4 ; temp3 = (r << 4) + nbits; 310 add ecx, eax 311 mov eax, INT [ebp + ecx * 4] ; code = actbl->ehufco[temp3]; 312 movzx ecx, byte [ebp + ecx + 1024] ; size = actbl->ehufsi[temp3]; 313 EMIT_BITS eax 314 315 movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k]; 316 ; Mask off any extra bits in code 317 mov ecx, dword [esp+temp2] 318 mov eax, 1 319 shl eax, cl 320 dec eax 321 and eax, edx ; temp2 &= (((JLONG)1)<<nbits) - 1; 322 EMIT_BITS eax ; PUT_BITS(temp2, nbits) 323 mov edx, dword [esp+temp3] 324 add esi, 2 ; ++k; 325 shr edx, 1 ; index >>= 1; 326 327 jmp .BLOOP 328.ELOOP: 329 movdqa xmm0, XMMWORD [esp + t1 + 32 * SIZEOF_WORD] ; __m128i tmp0 = _mm_loadu_si128((__m128i *)(t1 + 0)); 330 movdqa xmm1, XMMWORD [esp + t1 + 40 * SIZEOF_WORD] ; __m128i tmp1 = _mm_loadu_si128((__m128i *)(t1 + 8)); 331 movdqa xmm2, XMMWORD [esp + t1 + 48 * SIZEOF_WORD] ; __m128i tmp2 = _mm_loadu_si128((__m128i *)(t1 + 16)); 332 movdqa xmm3, XMMWORD [esp + t1 + 56 * SIZEOF_WORD] ; __m128i tmp3 = _mm_loadu_si128((__m128i *)(t1 + 24)); 333 pcmpeqw xmm0, xmm7 ; tmp0 = _mm_cmpeq_epi16(tmp0, zero); 334 pcmpeqw xmm1, xmm7 ; tmp1 = _mm_cmpeq_epi16(tmp1, zero); 335 pcmpeqw xmm2, xmm7 ; tmp2 = _mm_cmpeq_epi16(tmp2, zero); 336 pcmpeqw xmm3, xmm7 ; tmp3 = _mm_cmpeq_epi16(tmp3, zero); 337 packsswb xmm0, xmm1 ; tmp0 = _mm_packs_epi16(tmp0, tmp1); 338 packsswb xmm2, xmm3 ; tmp2 = _mm_packs_epi16(tmp2, tmp3); 339 pmovmskb edx, xmm0 ; index = ((uint64_t)_mm_movemask_epi8(tmp0)) << 0; 340 pmovmskb ecx, xmm2 ; index = ((uint64_t)_mm_movemask_epi8(tmp2)) << 16; 341 shl ecx, 16 342 or edx, ecx 343 not edx ; index = ~index; 344 345 lea eax, [esp + t1 + (DCTSIZE2/2) * 2] 346 sub eax, esi 347 shr eax, 1 348 bsf ecx, edx ; r = __builtin_ctzl(index); 349 jz near .ELOOP2 350 shr edx, cl ; index >>= r; 351 add ecx, eax 352 lea esi, [esi+ecx*2] ; k += r; 353 mov dword [esp+temp3], edx 354 jmp .BRLOOP2 355.BLOOP2: 356 bsf ecx, edx ; r = __builtin_ctzl(index); 357 jz near .ELOOP2 358 lea esi, [esi+ecx*2] ; k += r; 359 shr edx, cl ; index >>= r; 360 mov dword [esp+temp3], edx 361.BRLOOP2: 362 cmp ecx, 16 ; while (r > 15) { 363 jl near .ERLOOP2 364 sub ecx, 16 ; r -= 16; 365 mov dword [esp+temp], ecx 366 mov eax, INT [ebp + 240 * 4] ; code_0xf0 = actbl->ehufco[0xf0]; 367 movzx ecx, byte [ebp + 1024 + 240] ; size_0xf0 = actbl->ehufsi[0xf0]; 368 EMIT_BITS eax ; EMIT_BITS(code_0xf0, size_0xf0) 369 mov ecx, dword [esp+temp] 370 jmp .BRLOOP2 371.ERLOOP2: 372 movsx eax, word [esi] ; temp = t1[k]; 373 bsr eax, eax ; nbits = 32 - __builtin_clz(temp); 374 inc eax 375 mov dword [esp+temp2], eax 376 ; Emit Huffman symbol for run length / number of bits 377 shl ecx, 4 ; temp3 = (r << 4) + nbits; 378 add ecx, eax 379 mov eax, INT [ebp + ecx * 4] ; code = actbl->ehufco[temp3]; 380 movzx ecx, byte [ebp + ecx + 1024] ; size = actbl->ehufsi[temp3]; 381 EMIT_BITS eax 382 383 movsx edx, word [esi+DCTSIZE2*2] ; temp2 = t2[k]; 384 ; Mask off any extra bits in code 385 mov ecx, dword [esp+temp2] 386 mov eax, 1 387 shl eax, cl 388 dec eax 389 and eax, edx ; temp2 &= (((JLONG)1)<<nbits) - 1; 390 EMIT_BITS eax ; PUT_BITS(temp2, nbits) 391 mov edx, dword [esp+temp3] 392 add esi, 2 ; ++k; 393 shr edx, 1 ; index >>= 1; 394 395 jmp .BLOOP2 396.ELOOP2: 397 ; If the last coef(s) were zero, emit an end-of-block code 398 lea edx, [esp + t1 + (DCTSIZE2-1) * 2] ; r = DCTSIZE2-1-k; 399 cmp edx, esi ; if (r > 0) { 400 je .EFN 401 mov eax, INT [ebp] ; code = actbl->ehufco[0]; 402 movzx ecx, byte [ebp + 1024] ; size = actbl->ehufsi[0]; 403 EMIT_BITS eax 404.EFN: 405 mov eax, [esp+buffer] 406 pop esi 407 ; Save put_buffer & put_bits 408 mov dword [esi+8], put_buffer ; state->cur.put_buffer = put_buffer; 409 mov dword [esi+12], put_bits ; state->cur.put_bits = put_bits; 410 411 pop ebp 412 pop edi 413 pop esi 414; pop edx ; need not be preserved 415 pop ecx 416 pop ebx 417 mov esp, ebp ; esp <- aligned ebp 418 pop esp ; esp <- original ebp 419 pop ebp 420 ret 421 422; For some reason, the OS X linker does not honor the request to align the 423; segment unless we do this. 424 align 32 425