Diego Biurrun
12 years ago
1 changed files with 49 additions and 55 deletions
Unified View
Diff Options
-
+49 -55libavcodec/x86/dsputil_mmx.c
+ 49
- 55
libavcodec/x86/dsputil_mmx.c
View File
| @@ -57,8 +57,8 @@ void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, | |||||
| "movq %%mm2, (%0, %1) \n\t" | "movq %%mm2, (%0, %1) \n\t" | ||||
| "movq %%mm4, (%0, %1, 2) \n\t" | "movq %%mm4, (%0, %1, 2) \n\t" | ||||
| "movq %%mm6, (%0, %2) \n\t" | "movq %%mm6, (%0, %2) \n\t" | ||||
| :: "r"(pix), "r"((x86_reg)line_size), "r"((x86_reg)line_size * 3), | |||||
| "r"(p) | |||||
| :: "r" (pix), "r" ((x86_reg) line_size), "r" ((x86_reg) line_size * 3), | |||||
| "r" (p) | |||||
| : "memory"); | : "memory"); | ||||
| pix += line_size * 4; | pix += line_size * 4; | ||||
| p += 32; | p += 32; | ||||
| @@ -83,7 +83,8 @@ void ff_put_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, | |||||
| "movq %%mm2, (%0, %1) \n\t" | "movq %%mm2, (%0, %1) \n\t" | ||||
| "movq %%mm4, (%0, %1, 2) \n\t" | "movq %%mm4, (%0, %1, 2) \n\t" | ||||
| "movq %%mm6, (%0, %2) \n\t" | "movq %%mm6, (%0, %2) \n\t" | ||||
| :: "r"(pix), "r"((x86_reg)line_size), "r"((x86_reg)line_size * 3), "r"(p) | |||||
| :: "r" (pix), "r" ((x86_reg) line_size), "r" ((x86_reg) line_size * 3), | |||||
| "r" (p) | |||||
| : "memory"); | : "memory"); | ||||
| } | } | ||||
| @@ -117,8 +118,8 @@ void ff_put_signed_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, | |||||
| put_signed_pixels_clamped_mmx_half(0) | put_signed_pixels_clamped_mmx_half(0) | ||||
| "lea (%0, %3, 4), %0 \n\t" | "lea (%0, %3, 4), %0 \n\t" | ||||
| put_signed_pixels_clamped_mmx_half(64) | put_signed_pixels_clamped_mmx_half(64) | ||||
| : "+&r"(pixels), "=&r"(line_skip3) | |||||
| : "r"(block), "r"(line_skip) | |||||
| : "+&r" (pixels), "=&r" (line_skip3) | |||||
| : "r" (block), "r" (line_skip) | |||||
| : "memory"); | : "memory"); | ||||
| } | } | ||||
| @@ -156,8 +157,8 @@ void ff_add_pixels_clamped_mmx(const int16_t *block, uint8_t *pixels, | |||||
| "packuswb %%mm3, %%mm2 \n\t" | "packuswb %%mm3, %%mm2 \n\t" | ||||
| "movq %%mm0, %0 \n\t" | "movq %%mm0, %0 \n\t" | ||||
| "movq %%mm2, %1 \n\t" | "movq %%mm2, %1 \n\t" | ||||
| : "+m"(*pix), "+m"(*(pix + line_size)) | |||||
| : "r"(p) | |||||
| : "+m" (*pix), "+m" (*(pix + line_size)) | |||||
| : "r" (p) | |||||
| : "memory"); | : "memory"); | ||||
| pix += line_size * 2; | pix += line_size * 2; | ||||
| p += 16; | p += 16; | ||||
| @@ -177,10 +178,9 @@ void name(int16_t *blocks) \ | |||||
| "movq %%mm7, 24(%0, %%"REG_a") \n\t" \ | "movq %%mm7, 24(%0, %%"REG_a") \n\t" \ | ||||
| "add $32, %%"REG_a" \n\t" \ | "add $32, %%"REG_a" \n\t" \ | ||||
| "js 1b \n\t" \ | "js 1b \n\t" \ | ||||
| :: "r"(((uint8_t *)blocks) + 128 * n), \ | |||||
| :: "r"(((uint8_t *) blocks) + 128 * n), \ | |||||
| "i"(-128 * n) \ | "i"(-128 * n) \ | ||||
| : "%"REG_a \ | |||||
| ); \ | |||||
| : "%"REG_a); \ | |||||
| } | } | ||||
| CLEAR_BLOCKS(ff_clear_blocks_mmx, 6) | CLEAR_BLOCKS(ff_clear_blocks_mmx, 6) | ||||
| CLEAR_BLOCKS(ff_clear_block_mmx, 1) | CLEAR_BLOCKS(ff_clear_block_mmx, 1) | ||||
| @@ -197,9 +197,8 @@ void ff_clear_block_sse(int16_t *block) | |||||
| "movaps %%xmm0, 80(%0) \n" | "movaps %%xmm0, 80(%0) \n" | ||||
| "movaps %%xmm0, 96(%0) \n" | "movaps %%xmm0, 96(%0) \n" | ||||
| "movaps %%xmm0, 112(%0) \n" | "movaps %%xmm0, 112(%0) \n" | ||||
| :: "r"(block) | |||||
| : "memory" | |||||
| ); | |||||
| :: "r" (block) | |||||
| : "memory"); | |||||
| } | } | ||||
| void ff_clear_blocks_sse(int16_t *blocks) | void ff_clear_blocks_sse(int16_t *blocks) | ||||
| @@ -218,15 +217,14 @@ void ff_clear_blocks_sse(int16_t *blocks) | |||||
| "movaps %%xmm0, 112(%0, %%"REG_a") \n" | "movaps %%xmm0, 112(%0, %%"REG_a") \n" | ||||
| "add $128, %%"REG_a" \n" | "add $128, %%"REG_a" \n" | ||||
| "js 1b \n" | "js 1b \n" | ||||
| :: "r"(((uint8_t *)blocks) + 128 * 6), | |||||
| "i"(-128 * 6) | |||||
| : "%"REG_a | |||||
| ); | |||||
| :: "r"(((uint8_t *) blocks) + 128 * 6), "i"(-128 * 6) | |||||
| : "%"REG_a); | |||||
| } | } | ||||
| void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, int w) | void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, int w) | ||||
| { | { | ||||
| x86_reg i = 0; | x86_reg i = 0; | ||||
| __asm__ volatile ( | __asm__ volatile ( | ||||
| "jmp 2f \n\t" | "jmp 2f \n\t" | ||||
| "1: \n\t" | "1: \n\t" | ||||
| @@ -242,10 +240,10 @@ void ff_add_bytes_mmx(uint8_t *dst, uint8_t *src, int w) | |||||
| "2: \n\t" | "2: \n\t" | ||||
| "cmp %3, %0 \n\t" | "cmp %3, %0 \n\t" | ||||
| "js 1b \n\t" | "js 1b \n\t" | ||||
| : "+r"(i) | |||||
| : "r"(src), "r"(dst), "r"((x86_reg)w - 15) | |||||
| ); | |||||
| for ( ; i < w; i++) | |||||
| : "+r" (i) | |||||
| : "r" (src), "r" (dst), "r" ((x86_reg) w - 15)); | |||||
| for (; i < w; i++) | |||||
| dst[i + 0] += src[i + 0]; | dst[i + 0] += src[i + 0]; | ||||
| } | } | ||||
| @@ -276,9 +274,9 @@ void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, | |||||
| "add %1, %0 \n\t" | "add %1, %0 \n\t" | ||||
| "cmp %3, %0 \n\t" | "cmp %3, %0 \n\t" | ||||
| "jb 1b \n\t" | "jb 1b \n\t" | ||||
| : "+r"(ptr) | |||||
| : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height) | |||||
| ); | |||||
| : "+r" (ptr) | |||||
| : "r" ((x86_reg) wrap), "r" ((x86_reg) width), | |||||
| "r" (ptr + wrap * height)); | |||||
| } else { | } else { | ||||
| __asm__ volatile ( | __asm__ volatile ( | ||||
| "1: \n\t" | "1: \n\t" | ||||
| @@ -297,9 +295,9 @@ void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, | |||||
| "add %1, %0 \n\t" | "add %1, %0 \n\t" | ||||
| "cmp %3, %0 \n\t" | "cmp %3, %0 \n\t" | ||||
| "jb 1b \n\t" | "jb 1b \n\t" | ||||
| : "+r"(ptr) | |||||
| : "r"((x86_reg)wrap), "r"((x86_reg)width), "r"(ptr + wrap * height) | |||||
| ); | |||||
| : "+r" (ptr) | |||||
| : "r" ((x86_reg) wrap), "r" ((x86_reg) width), | |||||
| "r" (ptr + wrap * height)); | |||||
| } | } | ||||
| /* top and bottom (and hopefully also the corners) */ | /* top and bottom (and hopefully also the corners) */ | ||||
| @@ -316,10 +314,10 @@ void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, | |||||
| "add $8, %0 \n\t" | "add $8, %0 \n\t" | ||||
| "cmp %4, %0 \n\t" | "cmp %4, %0 \n\t" | ||||
| "jb 1b \n\t" | "jb 1b \n\t" | ||||
| : "+r"(ptr) | |||||
| : "r"((x86_reg)buf - (x86_reg)ptr - w), "r"((x86_reg) -wrap), | |||||
| "r"((x86_reg) -wrap * 3), "r"(ptr + width + 2 * w) | |||||
| ); | |||||
| : "+r" (ptr) | |||||
| : "r" ((x86_reg) buf - (x86_reg) ptr - w), | |||||
| "r" ((x86_reg) - wrap), "r" ((x86_reg) - wrap * 3), | |||||
| "r" (ptr + width + 2 * w)); | |||||
| } | } | ||||
| } | } | ||||
| @@ -336,11 +334,10 @@ void ff_draw_edges_mmx(uint8_t *buf, int wrap, int width, int height, | |||||
| "add $8, %0 \n\t" | "add $8, %0 \n\t" | ||||
| "cmp %4, %0 \n\t" | "cmp %4, %0 \n\t" | ||||
| "jb 1b \n\t" | "jb 1b \n\t" | ||||
| : "+r"(ptr) | |||||
| : "r"((x86_reg)last_line - (x86_reg)ptr - w), | |||||
| "r"((x86_reg)wrap), "r"((x86_reg)wrap * 3), | |||||
| "r"(ptr + width + 2 * w) | |||||
| ); | |||||
| : "+r" (ptr) | |||||
| : "r" ((x86_reg) last_line - (x86_reg) ptr - w), | |||||
| "r" ((x86_reg) wrap), "r" ((x86_reg) wrap * 3), | |||||
| "r" (ptr + width + 2 * w)); | |||||
| } | } | ||||
| } | } | ||||
| } | } | ||||
| @@ -362,20 +359,21 @@ void ff_gmc_mmx(uint8_t *dst, uint8_t *src, | |||||
| const uint16_t r4[4] = { r, r, r, r }; | const uint16_t r4[4] = { r, r, r, r }; | ||||
| const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys }; | const uint16_t dxy4[4] = { dxys, dxys, dxys, dxys }; | ||||
| const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys }; | const uint16_t dyy4[4] = { dyys, dyys, dyys, dyys }; | ||||
| const uint64_t shift2 = 2 * shift; | |||||
| const uint64_t shift2 = 2 * shift; | |||||
| int x, y; | int x, y; | ||||
| const int dxw = (dxx - (1 << (16 + shift))) * (w - 1); | const int dxw = (dxx - (1 << (16 + shift))) * (w - 1); | ||||
| const int dyh = (dyy - (1 << (16 + shift))) * (h - 1); | const int dyh = (dyy - (1 << (16 + shift))) * (h - 1); | ||||
| const int dxh = dxy * (h - 1); | const int dxh = dxy * (h - 1); | ||||
| const int dyw = dyx * (w - 1); | const int dyw = dyx * (w - 1); | ||||
| if ( // non-constant fullpel offset (3% of blocks) | if ( // non-constant fullpel offset (3% of blocks) | ||||
| ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) | | ((ox ^ (ox + dxw)) | (ox ^ (ox + dxh)) | (ox ^ (ox + dxw + dxh)) | | ||||
| (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) | |||||
| (oy ^ (oy + dyw)) | (oy ^ (oy + dyh)) | (oy ^ (oy + dyw + dyh))) >> (16 + shift) || | |||||
| // uses more than 16 bits of subpel mv (only at huge resolution) | // uses more than 16 bits of subpel mv (only at huge resolution) | ||||
| || (dxx | dxy | dyx | dyy) & 15 || | |||||
| (unsigned)ix >= width - w || | |||||
| (unsigned)iy >= height - h) { | |||||
| (dxx | dxy | dyx | dyy) & 15 || | |||||
| (unsigned) ix >= width - w || | |||||
| (unsigned) iy >= height - h) { | |||||
| // FIXME could still use mmx for some of the rows | // FIXME could still use mmx for some of the rows | ||||
| ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, | ff_gmc_c(dst, src, stride, h, ox, oy, dxx, dxy, dyx, dyy, | ||||
| shift, r, width, height); | shift, r, width, height); | ||||
| @@ -389,8 +387,7 @@ void ff_gmc_mmx(uint8_t *dst, uint8_t *src, | |||||
| "pxor %%mm7, %%mm7 \n\t" | "pxor %%mm7, %%mm7 \n\t" | ||||
| "punpcklwd %%mm6, %%mm6 \n\t" | "punpcklwd %%mm6, %%mm6 \n\t" | ||||
| "punpcklwd %%mm6, %%mm6 \n\t" | "punpcklwd %%mm6, %%mm6 \n\t" | ||||
| :: "r"(1<<shift) | |||||
| ); | |||||
| :: "r" (1 << shift)); | |||||
| for (x = 0; x < w; x += 4) { | for (x = 0; x < w; x += 4) { | ||||
| uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0), | uint16_t dx4[4] = { oxs - dxys + dxxs * (x + 0), | ||||
| @@ -412,9 +409,8 @@ void ff_gmc_mmx(uint8_t *dst, uint8_t *src, | |||||
| "movq %%mm5, %1 \n\t" | "movq %%mm5, %1 \n\t" | ||||
| "psrlw $12, %%mm4 \n\t" | "psrlw $12, %%mm4 \n\t" | ||||
| "psrlw $12, %%mm5 \n\t" | "psrlw $12, %%mm5 \n\t" | ||||
| : "+m"(*dx4), "+m"(*dy4) | |||||
| : "m"(*dxy4), "m"(*dyy4) | |||||
| ); | |||||
| : "+m" (*dx4), "+m" (*dy4) | |||||
| : "m" (*dxy4), "m" (*dyy4)); | |||||
| __asm__ volatile ( | __asm__ volatile ( | ||||
| "movq %%mm6, %%mm2 \n\t" | "movq %%mm6, %%mm2 \n\t" | ||||
| @@ -450,11 +446,10 @@ void ff_gmc_mmx(uint8_t *dst, uint8_t *src, | |||||
| "packuswb %%mm0, %%mm0 \n\t" | "packuswb %%mm0, %%mm0 \n\t" | ||||
| "movd %%mm0, %0 \n\t" | "movd %%mm0, %0 \n\t" | ||||
| : "=m"(dst[x + y * stride]) | |||||
| : "m"(src[0]), "m"(src[1]), | |||||
| "m"(src[stride]), "m"(src[stride + 1]), | |||||
| "m"(*r4), "m"(shift2) | |||||
| ); | |||||
| : "=m" (dst[x + y * stride]) | |||||
| : "m" (src[0]), "m" (src[1]), | |||||
| "m" (src[stride]), "m" (src[stride + 1]), | |||||
| "m" (*r4), "m" (shift2)); | |||||
| src += stride; | src += stride; | ||||
| } | } | ||||
| src += 4 - h * stride; | src += 4 - h * stride; | ||||
| @@ -489,10 +484,9 @@ void ff_vector_clipf_sse(float *dst, const float *src, | |||||
| "movaps %%xmm3, 48(%1, %0) \n\t" | "movaps %%xmm3, 48(%1, %0) \n\t" | ||||
| "sub $64, %0 \n\t" | "sub $64, %0 \n\t" | ||||
| "jge 1b \n\t" | "jge 1b \n\t" | ||||
| : "+&r"(i) | |||||
| : "r"(dst), "r"(src), "m"(min), "m"(max) | |||||
| : "memory" | |||||
| ); | |||||
| : "+&r" (i) | |||||
| : "r" (dst), "r" (src), "m" (min), "m" (max) | |||||
| : "memory"); | |||||
| } | } | ||||
| #endif /* HAVE_INLINE_ASM */ | #endif /* HAVE_INLINE_ASM */ | ||||