Mercurial > libavcodec.hg
comparison i386/fdct_mmx.c @ 2979:bfabfdf9ce55 libavcodec
COSMETICS: tabs --> spaces, some prettyprinting
| author | diego |
|---|---|
| date | Thu, 22 Dec 2005 01:10:11 +0000 |
| parents | ef2149182f1c |
| children | 96f9bd6a9ea9 |
comparison
equal
deleted
inserted
replaced
| 2978:403183bbb505 | 2979:bfabfdf9ce55 |
|---|---|
| 28 // (8-byte) memory boundaries! Otherwise the unaligned memory access will | 28 // (8-byte) memory boundaries! Otherwise the unaligned memory access will |
| 29 // severely stall MMX execution. | 29 // severely stall MMX execution. |
| 30 // | 30 // |
| 31 ////////////////////////////////////////////////////////////////////// | 31 ////////////////////////////////////////////////////////////////////// |
| 32 | 32 |
| 33 #define BITS_FRW_ACC 3 //; 2 or 3 for accuracy | 33 #define BITS_FRW_ACC 3 //; 2 or 3 for accuracy |
| 34 #define SHIFT_FRW_COL BITS_FRW_ACC | 34 #define SHIFT_FRW_COL BITS_FRW_ACC |
| 35 #define SHIFT_FRW_ROW (BITS_FRW_ACC + 17 - 3) | 35 #define SHIFT_FRW_ROW (BITS_FRW_ACC + 17 - 3) |
| 36 #define RND_FRW_ROW (1 << (SHIFT_FRW_ROW-1)) | 36 #define RND_FRW_ROW (1 << (SHIFT_FRW_ROW-1)) |
| 37 //#define RND_FRW_COL (1 << (SHIFT_FRW_COL-1)) | 37 //#define RND_FRW_COL (1 << (SHIFT_FRW_COL-1)) |
| 38 | 38 |
| 39 //concatenated table, for forward DCT transformation | 39 //concatenated table, for forward DCT transformation |
| 40 static const int16_t fdct_tg_all_16[] ATTR_ALIGN(8) = { | 40 static const int16_t fdct_tg_all_16[] ATTR_ALIGN(8) = { |
| 41 13036, 13036, 13036, 13036, // tg * (2<<16) + 0.5 | 41 13036, 13036, 13036, 13036, // tg * (2<<16) + 0.5 |
| 42 27146, 27146, 27146, 27146, // tg * (2<<16) + 0.5 | 42 27146, 27146, 27146, 27146, // tg * (2<<16) + 0.5 |
| 43 -21746, -21746, -21746, -21746, // tg * (2<<16) + 0.5 | 43 -21746, -21746, -21746, -21746, // tg * (2<<16) + 0.5 |
| 44 }; | 44 }; |
| 45 | 45 |
| 46 static const int16_t ocos_4_16[4] ATTR_ALIGN(8) = { | 46 static const int16_t ocos_4_16[4] ATTR_ALIGN(8) = { |
| 47 23170, 23170, 23170, 23170, //cos * (2<<15) + 0.5 | 47 23170, 23170, 23170, 23170, //cos * (2<<15) + 0.5 |
| 48 }; | 48 }; |
| 49 | 49 |
| 50 static const int64_t fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL; | 50 static const int64_t fdct_one_corr ATTR_ALIGN(8) = 0x0001000100010001LL; |
| 51 | 51 |
| 52 static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW }; | 52 static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW }; |
| 349 | 349 |
| 350 | 350 |
| 351 static always_inline void fdct_row_sse2(const int16_t *in, int16_t *out) | 351 static always_inline void fdct_row_sse2(const int16_t *in, int16_t *out) |
| 352 { | 352 { |
| 353 asm volatile( | 353 asm volatile( |
| 354 ".macro FDCT_ROW_SSE2_H1 i t \n\t" | 354 ".macro FDCT_ROW_SSE2_H1 i t \n\t" |
| 355 "movq \\i(%0), %%xmm2 \n\t" | 355 "movq \\i(%0), %%xmm2 \n\t" |
| 356 "movq \\i+8(%0), %%xmm0 \n\t" | 356 "movq \\i+8(%0), %%xmm0 \n\t" |
| 357 "movdqa \\t+32(%1), %%xmm3 \n\t" | 357 "movdqa \\t+32(%1), %%xmm3 \n\t" |
| 358 "movdqa \\t+48(%1), %%xmm7 \n\t" | 358 "movdqa \\t+48(%1), %%xmm7 \n\t" |
| 359 "movdqa \\t(%1), %%xmm4 \n\t" | 359 "movdqa \\t(%1), %%xmm4 \n\t" |
| 360 "movdqa \\t+16(%1), %%xmm5 \n\t" | 360 "movdqa \\t+16(%1), %%xmm5 \n\t" |
| 361 ".endm \n\t" | 361 ".endm \n\t" |
| 362 ".macro FDCT_ROW_SSE2_H2 i t \n\t" | 362 ".macro FDCT_ROW_SSE2_H2 i t \n\t" |
| 363 "movq \\i(%0), %%xmm2 \n\t" | 363 "movq \\i(%0), %%xmm2 \n\t" |
| 364 "movq \\i+8(%0), %%xmm0 \n\t" | 364 "movq \\i+8(%0), %%xmm0 \n\t" |
| 365 "movdqa \\t+32(%1), %%xmm3 \n\t" | 365 "movdqa \\t+32(%1), %%xmm3 \n\t" |
| 366 "movdqa \\t+48(%1), %%xmm7 \n\t" | 366 "movdqa \\t+48(%1), %%xmm7 \n\t" |
| 367 ".endm \n\t" | 367 ".endm \n\t" |
| 368 ".macro FDCT_ROW_SSE2 i \n\t" | 368 ".macro FDCT_ROW_SSE2 i \n\t" |
| 369 "movq %%xmm2, %%xmm1 \n\t" | 369 "movq %%xmm2, %%xmm1 \n\t" |
| 370 "pshuflw $27, %%xmm0, %%xmm0 \n\t" | 370 "pshuflw $27, %%xmm0, %%xmm0 \n\t" |
| 371 "paddsw %%xmm0, %%xmm1 \n\t" | 371 "paddsw %%xmm0, %%xmm1 \n\t" |
| 372 "psubsw %%xmm0, %%xmm2 \n\t" | 372 "psubsw %%xmm0, %%xmm2 \n\t" |
| 373 "punpckldq %%xmm2, %%xmm1 \n\t" | 373 "punpckldq %%xmm2, %%xmm1 \n\t" |
| 374 "pshufd $78, %%xmm1, %%xmm2 \n\t" | 374 "pshufd $78, %%xmm1, %%xmm2 \n\t" |
| 375 "pmaddwd %%xmm2, %%xmm3 \n\t" | 375 "pmaddwd %%xmm2, %%xmm3 \n\t" |
| 376 "pmaddwd %%xmm1, %%xmm7 \n\t" | 376 "pmaddwd %%xmm1, %%xmm7 \n\t" |
| 377 "pmaddwd %%xmm5, %%xmm2 \n\t" | 377 "pmaddwd %%xmm5, %%xmm2 \n\t" |
| 378 "pmaddwd %%xmm4, %%xmm1 \n\t" | 378 "pmaddwd %%xmm4, %%xmm1 \n\t" |
| 379 "paddd %%xmm7, %%xmm3 \n\t" | 379 "paddd %%xmm7, %%xmm3 \n\t" |
| 380 "paddd %%xmm2, %%xmm1 \n\t" | 380 "paddd %%xmm2, %%xmm1 \n\t" |
| 381 "paddd %%xmm6, %%xmm3 \n\t" | 381 "paddd %%xmm6, %%xmm3 \n\t" |
| 382 "paddd %%xmm6, %%xmm1 \n\t" | 382 "paddd %%xmm6, %%xmm1 \n\t" |
| 383 "psrad %3, %%xmm3 \n\t" | 383 "psrad %3, %%xmm3 \n\t" |
| 384 "psrad %3, %%xmm1 \n\t" | 384 "psrad %3, %%xmm1 \n\t" |
| 385 "packssdw %%xmm3, %%xmm1 \n\t" | 385 "packssdw %%xmm3, %%xmm1 \n\t" |
| 386 "movdqa %%xmm1, \\i(%4) \n\t" | 386 "movdqa %%xmm1, \\i(%4) \n\t" |
| 387 ".endm \n\t" | 387 ".endm \n\t" |
| 388 "movdqa (%2), %%xmm6 \n\t" | 388 "movdqa (%2), %%xmm6 \n\t" |
| 389 "FDCT_ROW_SSE2_H1 0 0 \n\t" | 389 "FDCT_ROW_SSE2_H1 0 0 \n\t" |
| 390 "FDCT_ROW_SSE2 0 \n\t" | 390 "FDCT_ROW_SSE2 0 \n\t" |
| 391 "FDCT_ROW_SSE2_H2 64 0 \n\t" | 391 "FDCT_ROW_SSE2_H2 64 0 \n\t" |
| 392 "FDCT_ROW_SSE2 64 \n\t" | 392 "FDCT_ROW_SSE2 64 \n\t" |
| 393 | 393 |
| 394 "FDCT_ROW_SSE2_H1 16 64 \n\t" | 394 "FDCT_ROW_SSE2_H1 16 64 \n\t" |
| 395 "FDCT_ROW_SSE2 16 \n\t" | 395 "FDCT_ROW_SSE2 16 \n\t" |
| 396 "FDCT_ROW_SSE2_H2 112 64 \n\t" | 396 "FDCT_ROW_SSE2_H2 112 64 \n\t" |
| 397 "FDCT_ROW_SSE2 112 \n\t" | 397 "FDCT_ROW_SSE2 112 \n\t" |
| 398 | 398 |
| 399 "FDCT_ROW_SSE2_H1 32 128 \n\t" | 399 "FDCT_ROW_SSE2_H1 32 128 \n\t" |
| 400 "FDCT_ROW_SSE2 32 \n\t" | 400 "FDCT_ROW_SSE2 32 \n\t" |
| 401 "FDCT_ROW_SSE2_H2 96 128 \n\t" | 401 "FDCT_ROW_SSE2_H2 96 128 \n\t" |
| 402 "FDCT_ROW_SSE2 96 \n\t" | 402 "FDCT_ROW_SSE2 96 \n\t" |
| 403 | 403 |
| 404 "FDCT_ROW_SSE2_H1 48 192 \n\t" | 404 "FDCT_ROW_SSE2_H1 48 192 \n\t" |
| 405 "FDCT_ROW_SSE2 48 \n\t" | 405 "FDCT_ROW_SSE2 48 \n\t" |
| 406 "FDCT_ROW_SSE2_H2 80 192 \n\t" | 406 "FDCT_ROW_SSE2_H2 80 192 \n\t" |
| 407 "FDCT_ROW_SSE2 80 \n\t" | 407 "FDCT_ROW_SSE2 80 \n\t" |
| 408 : | 408 : |
| 409 : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out) | 409 : "r" (in), "r" (tab_frw_01234567_sse2.tab_frw_01234567_sse2), "r" (fdct_r_row_sse2.fdct_r_row_sse2), "i" (SHIFT_FRW_ROW), "r" (out) |
| 410 ); | 410 ); |
| 411 } | 411 } |
| 412 | 412 |
| 413 static always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table) | 413 static always_inline void fdct_row_mmx2(const int16_t *in, int16_t *out, const int16_t *table) |
| 414 { | 414 { |