yading@11: /* yading@11: * AltiVec acceleration for colorspace conversion yading@11: * yading@11: * copyright (C) 2004 Marc Hoffman yading@11: * yading@11: * This file is part of FFmpeg. yading@11: * yading@11: * FFmpeg is free software; you can redistribute it and/or yading@11: * modify it under the terms of the GNU Lesser General Public yading@11: * License as published by the Free Software Foundation; either yading@11: * version 2.1 of the License, or (at your option) any later version. yading@11: * yading@11: * FFmpeg is distributed in the hope that it will be useful, yading@11: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@11: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@11: * Lesser General Public License for more details. yading@11: * yading@11: * You should have received a copy of the GNU Lesser General Public yading@11: * License along with FFmpeg; if not, write to the Free Software yading@11: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@11: */ yading@11: yading@11: /* yading@11: * Convert I420 YV12 to RGB in various formats, yading@11: * it rejects images that are not in 420 formats, yading@11: * it rejects images that don't have widths of multiples of 16, yading@11: * it rejects images that don't have heights of multiples of 2. yading@11: * Reject defers to C simulation code. yading@11: * yading@11: * Lots of optimizations to be done here. yading@11: * yading@11: * 1. Need to fix saturation code. I just couldn't get it to fly with packs yading@11: * and adds, so we currently use max/min to clip. yading@11: * yading@11: * 2. The inefficient use of chroma loading needs a bit of brushing up. yading@11: * yading@11: * 3. Analysis of pipeline stalls needs to be done. Use shark to identify yading@11: * pipeline stalls. yading@11: * yading@11: * yading@11: * MODIFIED to calculate coeffs from currently selected color space. yading@11: * MODIFIED core to be a macro where you specify the output format. yading@11: * ADDED UYVY conversion which is never called due to some thing in swscale. yading@11: * CORRECTED algorithim selection to be strict on input formats. yading@11: * ADDED runtime detection of AltiVec. yading@11: * yading@11: * ADDED altivec_yuv2packedX vertical scl + RGB converter yading@11: * yading@11: * March 27,2004 yading@11: * PERFORMANCE ANALYSIS yading@11: * yading@11: * The C version uses 25% of the processor or ~250Mips for D1 video rawvideo yading@11: * used as test. yading@11: * The AltiVec version uses 10% of the processor or ~100Mips for D1 video yading@11: * same sequence. yading@11: * yading@11: * 720 * 480 * 30 ~10MPS yading@11: * yading@11: * so we have roughly 10 clocks per pixel. This is too high, something has yading@11: * to be wrong. yading@11: * yading@11: * OPTIMIZED clip codes to utilize vec_max and vec_packs removing the yading@11: * need for vec_min. yading@11: * yading@11: * OPTIMIZED DST OUTPUT cache/DMA controls. We are pretty much guaranteed to yading@11: * have the input video frame, it was just decompressed so it probably resides yading@11: * in L1 caches. However, we are creating the output video stream. This needs yading@11: * to use the DSTST instruction to optimize for the cache. We couple this with yading@11: * the fact that we are not going to be visiting the input buffer again so we yading@11: * mark it Least Recently Used. This shaves 25% of the processor cycles off. yading@11: * yading@11: * Now memcpy is the largest mips consumer in the system, probably due yading@11: * to the inefficient X11 stuff. yading@11: * yading@11: * GL libraries seem to be very slow on this machine 1.33Ghz PB running yading@11: * Jaguar, this is not the case for my 1Ghz PB. I thought it might be yading@11: * a versioning issue, however I have libGL.1.2.dylib for both yading@11: * machines. (We need to figure this out now.) yading@11: * yading@11: * GL2 libraries work now with patch for RGB32. yading@11: * yading@11: * NOTE: quartz vo driver ARGB32_to_RGB24 consumes 30% of the processor. yading@11: * yading@11: * Integrated luma prescaling adjustment for saturation/contrast/brightness yading@11: * adjustment. yading@11: */ yading@11: yading@11: #include yading@11: #include yading@11: #include yading@11: #include yading@11: #include yading@11: yading@11: #include "config.h" yading@11: #include "libswscale/rgb2rgb.h" yading@11: #include "libswscale/swscale.h" yading@11: #include "libswscale/swscale_internal.h" yading@11: #include "libavutil/attributes.h" yading@11: #include "libavutil/cpu.h" yading@11: #include "libavutil/pixdesc.h" yading@11: #include "yuv2rgb_altivec.h" yading@11: yading@11: #undef PROFILE_THE_BEAST yading@11: #undef INC_SCALING yading@11: yading@11: typedef unsigned char ubyte; yading@11: typedef signed char sbyte; yading@11: yading@11: /* RGB interleaver, 16 planar pels 8-bit samples per channel in yading@11: * homogeneous vector registers x0,x1,x2 are interleaved with the yading@11: * following technique: yading@11: * yading@11: * o0 = vec_mergeh(x0, x1); yading@11: * o1 = vec_perm(o0, x2, perm_rgb_0); yading@11: * o2 = vec_perm(o0, x2, perm_rgb_1); yading@11: * o3 = vec_mergel(x0, x1); yading@11: * o4 = vec_perm(o3, o2, perm_rgb_2); yading@11: * o5 = vec_perm(o3, o2, perm_rgb_3); yading@11: * yading@11: * perm_rgb_0: o0(RG).h v1(B) --> o1* yading@11: * 0 1 2 3 4 yading@11: * rgbr|gbrg|brgb|rgbr yading@11: * 0010 0100 1001 0010 yading@11: * 0102 3145 2673 894A yading@11: * yading@11: * perm_rgb_1: o0(RG).h v1(B) --> o2 yading@11: * 0 1 2 3 4 yading@11: * gbrg|brgb|bbbb|bbbb yading@11: * 0100 1001 1111 1111 yading@11: * B5CD 6EF7 89AB CDEF yading@11: * yading@11: * perm_rgb_2: o3(RG).l o2(rgbB.l) --> o4* yading@11: * 0 1 2 3 4 yading@11: * gbrg|brgb|rgbr|gbrg yading@11: * 1111 1111 0010 0100 yading@11: * 89AB CDEF 0182 3945 yading@11: * yading@11: * perm_rgb_2: o3(RG).l o2(rgbB.l) ---> o5* yading@11: * 0 1 2 3 4 yading@11: * brgb|rgbr|gbrg|brgb yading@11: * 1001 0010 0100 1001 yading@11: * a67b 89cA BdCD eEFf yading@11: * yading@11: */ yading@11: static const vector unsigned char yading@11: perm_rgb_0 = { 0x00, 0x01, 0x10, 0x02, 0x03, 0x11, 0x04, 0x05, yading@11: 0x12, 0x06, 0x07, 0x13, 0x08, 0x09, 0x14, 0x0a }, yading@11: perm_rgb_1 = { 0x0b, 0x15, 0x0c, 0x0d, 0x16, 0x0e, 0x0f, 0x17, yading@11: 0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f }, yading@11: perm_rgb_2 = { 0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, yading@11: 0x00, 0x01, 0x18, 0x02, 0x03, 0x19, 0x04, 0x05 }, yading@11: perm_rgb_3 = { 0x1a, 0x06, 0x07, 0x1b, 0x08, 0x09, 0x1c, 0x0a, yading@11: 0x0b, 0x1d, 0x0c, 0x0d, 0x1e, 0x0e, 0x0f, 0x1f }; yading@11: yading@11: #define vec_merge3(x2, x1, x0, y0, y1, y2) \ yading@11: do { \ yading@11: __typeof__(x0) o0, o2, o3; \ yading@11: o0 = vec_mergeh(x0, x1); \ yading@11: y0 = vec_perm(o0, x2, perm_rgb_0); \ yading@11: o2 = vec_perm(o0, x2, perm_rgb_1); \ yading@11: o3 = vec_mergel(x0, x1); \ yading@11: y1 = vec_perm(o3, o2, perm_rgb_2); \ yading@11: y2 = vec_perm(o3, o2, perm_rgb_3); \ yading@11: } while (0) yading@11: yading@11: #define vec_mstbgr24(x0, x1, x2, ptr) \ yading@11: do { \ yading@11: __typeof__(x0) _0, _1, _2; \ yading@11: vec_merge3(x0, x1, x2, _0, _1, _2); \ yading@11: vec_st(_0, 0, ptr++); \ yading@11: vec_st(_1, 0, ptr++); \ yading@11: vec_st(_2, 0, ptr++); \ yading@11: } while (0) yading@11: yading@11: #define vec_mstrgb24(x0, x1, x2, ptr) \ yading@11: do { \ yading@11: __typeof__(x0) _0, _1, _2; \ yading@11: vec_merge3(x2, x1, x0, _0, _1, _2); \ yading@11: vec_st(_0, 0, ptr++); \ yading@11: vec_st(_1, 0, ptr++); \ yading@11: vec_st(_2, 0, ptr++); \ yading@11: } while (0) yading@11: yading@11: /* pack the pixels in rgb0 format yading@11: * msb R yading@11: * lsb 0 yading@11: */ yading@11: #define vec_mstrgb32(T, x0, x1, x2, x3, ptr) \ yading@11: do { \ yading@11: T _0, _1, _2, _3; \ yading@11: _0 = vec_mergeh(x0, x1); \ yading@11: _1 = vec_mergeh(x2, x3); \ yading@11: _2 = (T) vec_mergeh((vector unsigned short) _0, \ yading@11: (vector unsigned short) _1); \ yading@11: _3 = (T) vec_mergel((vector unsigned short) _0, \ yading@11: (vector unsigned short) _1); \ yading@11: vec_st(_2, 0 * 16, (T *) ptr); \ yading@11: vec_st(_3, 1 * 16, (T *) ptr); \ yading@11: _0 = vec_mergel(x0, x1); \ yading@11: _1 = vec_mergel(x2, x3); \ yading@11: _2 = (T) vec_mergeh((vector unsigned short) _0, \ yading@11: (vector unsigned short) _1); \ yading@11: _3 = (T) vec_mergel((vector unsigned short) _0, \ yading@11: (vector unsigned short) _1); \ yading@11: vec_st(_2, 2 * 16, (T *) ptr); \ yading@11: vec_st(_3, 3 * 16, (T *) ptr); \ yading@11: ptr += 4; \ yading@11: } while (0) yading@11: yading@11: /* yading@11: * 1 0 1.4021 | | Y | yading@11: * 1 -0.3441 -0.7142 |x| Cb| yading@11: * 1 1.7718 0 | | Cr| yading@11: * yading@11: * yading@11: * Y: [-128 127] yading@11: * Cb/Cr : [-128 127] yading@11: * yading@11: * typical YUV conversion works on Y: 0-255 this version has been yading@11: * optimized for JPEG decoding. yading@11: */ yading@11: yading@11: #define vec_unh(x) \ yading@11: (vector signed short) \ yading@11: vec_perm(x, (__typeof__(x)) { 0 }, \ yading@11: ((vector unsigned char) { \ yading@11: 0x10, 0x00, 0x10, 0x01, 0x10, 0x02, 0x10, 0x03, \ yading@11: 0x10, 0x04, 0x10, 0x05, 0x10, 0x06, 0x10, 0x07 })) yading@11: yading@11: #define vec_unl(x) \ yading@11: (vector signed short) \ yading@11: vec_perm(x, (__typeof__(x)) { 0 }, \ yading@11: ((vector unsigned char) { \ yading@11: 0x10, 0x08, 0x10, 0x09, 0x10, 0x0A, 0x10, 0x0B, \ yading@11: 0x10, 0x0C, 0x10, 0x0D, 0x10, 0x0E, 0x10, 0x0F })) yading@11: yading@11: #define vec_clip_s16(x) \ yading@11: vec_max(vec_min(x, ((vector signed short) { \ yading@11: 235, 235, 235, 235, 235, 235, 235, 235 })), \ yading@11: ((vector signed short) { 16, 16, 16, 16, 16, 16, 16, 16 })) yading@11: yading@11: #define vec_packclp(x, y) \ yading@11: (vector unsigned char) \ yading@11: vec_packs((vector unsigned short) \ yading@11: vec_max(x, ((vector signed short) { 0 })), \ yading@11: (vector unsigned short) \ yading@11: vec_max(y, ((vector signed short) { 0 }))) yading@11: yading@11: //#define out_pixels(a, b, c, ptr) vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, a, a, ptr) yading@11: yading@11: static inline void cvtyuvtoRGB(SwsContext *c, vector signed short Y, yading@11: vector signed short U, vector signed short V, yading@11: vector signed short *R, vector signed short *G, yading@11: vector signed short *B) yading@11: { yading@11: vector signed short vx, ux, uvx; yading@11: yading@11: Y = vec_mradds(Y, c->CY, c->OY); yading@11: U = vec_sub(U, (vector signed short) yading@11: vec_splat((vector signed short) { 128 }, 0)); yading@11: V = vec_sub(V, (vector signed short) yading@11: vec_splat((vector signed short) { 128 }, 0)); yading@11: yading@11: // ux = (CBU * (u << c->CSHIFT) + 0x4000) >> 15; yading@11: ux = vec_sl(U, c->CSHIFT); yading@11: *B = vec_mradds(ux, c->CBU, Y); yading@11: yading@11: // vx = (CRV * (v << c->CSHIFT) + 0x4000) >> 15; yading@11: vx = vec_sl(V, c->CSHIFT); yading@11: *R = vec_mradds(vx, c->CRV, Y); yading@11: yading@11: // uvx = ((CGU * u) + (CGV * v)) >> 15; yading@11: uvx = vec_mradds(U, c->CGU, Y); yading@11: *G = vec_mradds(V, c->CGV, uvx); yading@11: } yading@11: yading@11: /* yading@11: * ------------------------------------------------------------------------------ yading@11: * CS converters yading@11: * ------------------------------------------------------------------------------ yading@11: */ yading@11: yading@11: #define DEFCSP420_CVT(name, out_pixels) \ yading@11: static int altivec_ ## name(SwsContext *c, const unsigned char **in, \ yading@11: int *instrides, int srcSliceY, int srcSliceH, \ yading@11: unsigned char **oplanes, int *outstrides) \ yading@11: { \ yading@11: int w = c->srcW; \ yading@11: int h = srcSliceH; \ yading@11: int i, j; \ yading@11: int instrides_scl[3]; \ yading@11: vector unsigned char y0, y1; \ yading@11: \ yading@11: vector signed char u, v; \ yading@11: \ yading@11: vector signed short Y0, Y1, Y2, Y3; \ yading@11: vector signed short U, V; \ yading@11: vector signed short vx, ux, uvx; \ yading@11: vector signed short vx0, ux0, uvx0; \ yading@11: vector signed short vx1, ux1, uvx1; \ yading@11: vector signed short R0, G0, B0; \ yading@11: vector signed short R1, G1, B1; \ yading@11: vector unsigned char R, G, B; \ yading@11: \ yading@11: const vector unsigned char *y1ivP, *y2ivP, *uivP, *vivP; \ yading@11: vector unsigned char align_perm; \ yading@11: \ yading@11: vector signed short lCY = c->CY; \ yading@11: vector signed short lOY = c->OY; \ yading@11: vector signed short lCRV = c->CRV; \ yading@11: vector signed short lCBU = c->CBU; \ yading@11: vector signed short lCGU = c->CGU; \ yading@11: vector signed short lCGV = c->CGV; \ yading@11: vector unsigned short lCSHIFT = c->CSHIFT; \ yading@11: \ yading@11: const ubyte *y1i = in[0]; \ yading@11: const ubyte *y2i = in[0] + instrides[0]; \ yading@11: const ubyte *ui = in[1]; \ yading@11: const ubyte *vi = in[2]; \ yading@11: \ yading@11: vector unsigned char *oute, *outo; \ yading@11: \ yading@11: /* loop moves y{1, 2}i by w */ \ yading@11: instrides_scl[0] = instrides[0] * 2 - w; \ yading@11: /* loop moves ui by w / 2 */ \ yading@11: instrides_scl[1] = instrides[1] - w / 2; \ yading@11: /* loop moves vi by w / 2 */ \ yading@11: instrides_scl[2] = instrides[2] - w / 2; \ yading@11: \ yading@11: for (i = 0; i < h / 2; i++) { \ yading@11: oute = (vector unsigned char *)(oplanes[0] + outstrides[0] * \ yading@11: (srcSliceY + i * 2)); \ yading@11: outo = oute + (outstrides[0] >> 4); \ yading@11: vec_dstst(outo, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 0); \ yading@11: vec_dstst(oute, (0x02000002 | (((w * 3 + 32) / 32) << 16)), 1); \ yading@11: \ yading@11: for (j = 0; j < w / 16; j++) { \ yading@11: y1ivP = (const vector unsigned char *) y1i; \ yading@11: y2ivP = (const vector unsigned char *) y2i; \ yading@11: uivP = (const vector unsigned char *) ui; \ yading@11: vivP = (const vector unsigned char *) vi; \ yading@11: \ yading@11: align_perm = vec_lvsl(0, y1i); \ yading@11: y0 = (vector unsigned char) \ yading@11: vec_perm(y1ivP[0], y1ivP[1], align_perm); \ yading@11: \ yading@11: align_perm = vec_lvsl(0, y2i); \ yading@11: y1 = (vector unsigned char) \ yading@11: vec_perm(y2ivP[0], y2ivP[1], align_perm); \ yading@11: \ yading@11: align_perm = vec_lvsl(0, ui); \ yading@11: u = (vector signed char) \ yading@11: vec_perm(uivP[0], uivP[1], align_perm); \ yading@11: \ yading@11: align_perm = vec_lvsl(0, vi); \ yading@11: v = (vector signed char) \ yading@11: vec_perm(vivP[0], vivP[1], align_perm); \ yading@11: \ yading@11: u = (vector signed char) \ yading@11: vec_sub(u, \ yading@11: (vector signed char) \ yading@11: vec_splat((vector signed char) { 128 }, 0)); \ yading@11: v = (vector signed char) \ yading@11: vec_sub(v, \ yading@11: (vector signed char) \ yading@11: vec_splat((vector signed char) { 128 }, 0)); \ yading@11: \ yading@11: U = vec_unpackh(u); \ yading@11: V = vec_unpackh(v); \ yading@11: \ yading@11: Y0 = vec_unh(y0); \ yading@11: Y1 = vec_unl(y0); \ yading@11: Y2 = vec_unh(y1); \ yading@11: Y3 = vec_unl(y1); \ yading@11: \ yading@11: Y0 = vec_mradds(Y0, lCY, lOY); \ yading@11: Y1 = vec_mradds(Y1, lCY, lOY); \ yading@11: Y2 = vec_mradds(Y2, lCY, lOY); \ yading@11: Y3 = vec_mradds(Y3, lCY, lOY); \ yading@11: \ yading@11: /* ux = (CBU * (u << CSHIFT) + 0x4000) >> 15 */ \ yading@11: ux = vec_sl(U, lCSHIFT); \ yading@11: ux = vec_mradds(ux, lCBU, (vector signed short) { 0 }); \ yading@11: ux0 = vec_mergeh(ux, ux); \ yading@11: ux1 = vec_mergel(ux, ux); \ yading@11: \ yading@11: /* vx = (CRV * (v << CSHIFT) + 0x4000) >> 15; */ \ yading@11: vx = vec_sl(V, lCSHIFT); \ yading@11: vx = vec_mradds(vx, lCRV, (vector signed short) { 0 }); \ yading@11: vx0 = vec_mergeh(vx, vx); \ yading@11: vx1 = vec_mergel(vx, vx); \ yading@11: \ yading@11: /* uvx = ((CGU * u) + (CGV * v)) >> 15 */ \ yading@11: uvx = vec_mradds(U, lCGU, (vector signed short) { 0 }); \ yading@11: uvx = vec_mradds(V, lCGV, uvx); \ yading@11: uvx0 = vec_mergeh(uvx, uvx); \ yading@11: uvx1 = vec_mergel(uvx, uvx); \ yading@11: \ yading@11: R0 = vec_add(Y0, vx0); \ yading@11: G0 = vec_add(Y0, uvx0); \ yading@11: B0 = vec_add(Y0, ux0); \ yading@11: R1 = vec_add(Y1, vx1); \ yading@11: G1 = vec_add(Y1, uvx1); \ yading@11: B1 = vec_add(Y1, ux1); \ yading@11: \ yading@11: R = vec_packclp(R0, R1); \ yading@11: G = vec_packclp(G0, G1); \ yading@11: B = vec_packclp(B0, B1); \ yading@11: \ yading@11: out_pixels(R, G, B, oute); \ yading@11: \ yading@11: R0 = vec_add(Y2, vx0); \ yading@11: G0 = vec_add(Y2, uvx0); \ yading@11: B0 = vec_add(Y2, ux0); \ yading@11: R1 = vec_add(Y3, vx1); \ yading@11: G1 = vec_add(Y3, uvx1); \ yading@11: B1 = vec_add(Y3, ux1); \ yading@11: R = vec_packclp(R0, R1); \ yading@11: G = vec_packclp(G0, G1); \ yading@11: B = vec_packclp(B0, B1); \ yading@11: \ yading@11: \ yading@11: out_pixels(R, G, B, outo); \ yading@11: \ yading@11: y1i += 16; \ yading@11: y2i += 16; \ yading@11: ui += 8; \ yading@11: vi += 8; \ yading@11: } \ yading@11: \ yading@11: ui += instrides_scl[1]; \ yading@11: vi += instrides_scl[2]; \ yading@11: y1i += instrides_scl[0]; \ yading@11: y2i += instrides_scl[0]; \ yading@11: } \ yading@11: return srcSliceH; \ yading@11: } yading@11: yading@11: #define out_abgr(a, b, c, ptr) \ yading@11: vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), c, b, a, ptr) yading@11: #define out_bgra(a, b, c, ptr) \ yading@11: vec_mstrgb32(__typeof__(a), c, b, a, ((__typeof__(a)) { 255 }), ptr) yading@11: #define out_rgba(a, b, c, ptr) \ yading@11: vec_mstrgb32(__typeof__(a), a, b, c, ((__typeof__(a)) { 255 }), ptr) yading@11: #define out_argb(a, b, c, ptr) \ yading@11: vec_mstrgb32(__typeof__(a), ((__typeof__(a)) { 255 }), a, b, c, ptr) yading@11: #define out_rgb24(a, b, c, ptr) vec_mstrgb24(a, b, c, ptr) yading@11: #define out_bgr24(a, b, c, ptr) vec_mstbgr24(a, b, c, ptr) yading@11: yading@11: DEFCSP420_CVT(yuv2_abgr, out_abgr) yading@11: DEFCSP420_CVT(yuv2_bgra, out_bgra) yading@11: DEFCSP420_CVT(yuv2_rgba, out_rgba) yading@11: DEFCSP420_CVT(yuv2_argb, out_argb) yading@11: DEFCSP420_CVT(yuv2_rgb24, out_rgb24) yading@11: DEFCSP420_CVT(yuv2_bgr24, out_bgr24) yading@11: yading@11: // uyvy|uyvy|uyvy|uyvy yading@11: // 0123 4567 89ab cdef yading@11: static const vector unsigned char yading@11: demux_u = { 0x10, 0x00, 0x10, 0x00, yading@11: 0x10, 0x04, 0x10, 0x04, yading@11: 0x10, 0x08, 0x10, 0x08, yading@11: 0x10, 0x0c, 0x10, 0x0c }, yading@11: demux_v = { 0x10, 0x02, 0x10, 0x02, yading@11: 0x10, 0x06, 0x10, 0x06, yading@11: 0x10, 0x0A, 0x10, 0x0A, yading@11: 0x10, 0x0E, 0x10, 0x0E }, yading@11: demux_y = { 0x10, 0x01, 0x10, 0x03, yading@11: 0x10, 0x05, 0x10, 0x07, yading@11: 0x10, 0x09, 0x10, 0x0B, yading@11: 0x10, 0x0D, 0x10, 0x0F }; yading@11: yading@11: /* yading@11: * this is so I can play live CCIR raw video yading@11: */ yading@11: static int altivec_uyvy_rgb32(SwsContext *c, const unsigned char **in, yading@11: int *instrides, int srcSliceY, int srcSliceH, yading@11: unsigned char **oplanes, int *outstrides) yading@11: { yading@11: int w = c->srcW; yading@11: int h = srcSliceH; yading@11: int i, j; yading@11: vector unsigned char uyvy; yading@11: vector signed short Y, U, V; yading@11: vector signed short R0, G0, B0, R1, G1, B1; yading@11: vector unsigned char R, G, B; yading@11: vector unsigned char *out; yading@11: const ubyte *img; yading@11: yading@11: img = in[0]; yading@11: out = (vector unsigned char *) (oplanes[0] + srcSliceY * outstrides[0]); yading@11: yading@11: for (i = 0; i < h; i++) yading@11: for (j = 0; j < w / 16; j++) { yading@11: uyvy = vec_ld(0, img); yading@11: yading@11: U = (vector signed short) yading@11: vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u); yading@11: V = (vector signed short) yading@11: vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v); yading@11: Y = (vector signed short) yading@11: vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y); yading@11: yading@11: cvtyuvtoRGB(c, Y, U, V, &R0, &G0, &B0); yading@11: yading@11: uyvy = vec_ld(16, img); yading@11: yading@11: U = (vector signed short) yading@11: vec_perm(uyvy, (vector unsigned char) { 0 }, demux_u); yading@11: V = (vector signed short) yading@11: vec_perm(uyvy, (vector unsigned char) { 0 }, demux_v); yading@11: Y = (vector signed short) yading@11: vec_perm(uyvy, (vector unsigned char) { 0 }, demux_y); yading@11: yading@11: cvtyuvtoRGB(c, Y, U, V, &R1, &G1, &B1); yading@11: yading@11: R = vec_packclp(R0, R1); yading@11: G = vec_packclp(G0, G1); yading@11: B = vec_packclp(B0, B1); yading@11: yading@11: // vec_mstbgr24 (R,G,B, out); yading@11: out_rgba(R, G, B, out); yading@11: yading@11: img += 32; yading@11: } yading@11: return srcSliceH; yading@11: } yading@11: yading@11: /* Ok currently the acceleration routine only supports yading@11: * inputs of widths a multiple of 16 yading@11: * and heights a multiple 2 yading@11: * yading@11: * So we just fall back to the C codes for this. yading@11: */ yading@11: av_cold SwsFunc ff_yuv2rgb_init_altivec(SwsContext *c) yading@11: { yading@11: if (!(av_get_cpu_flags() & AV_CPU_FLAG_ALTIVEC)) yading@11: return NULL; yading@11: yading@11: /* yading@11: * and this seems not to matter too much I tried a bunch of yading@11: * videos with abnormal widths and MPlayer crashes elsewhere. yading@11: * mplayer -vo x11 -rawvideo on:w=350:h=240 raw-350x240.eyuv yading@11: * boom with X11 bad match. yading@11: * yading@11: */ yading@11: if ((c->srcW & 0xf) != 0) yading@11: return NULL; yading@11: yading@11: switch (c->srcFormat) { yading@11: case AV_PIX_FMT_YUV410P: yading@11: case AV_PIX_FMT_YUV420P: yading@11: /*case IMGFMT_CLPL: ??? */ yading@11: case AV_PIX_FMT_GRAY8: yading@11: case AV_PIX_FMT_NV12: yading@11: case AV_PIX_FMT_NV21: yading@11: if ((c->srcH & 0x1) != 0) yading@11: return NULL; yading@11: yading@11: switch (c->dstFormat) { yading@11: case AV_PIX_FMT_RGB24: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGB24\n"); yading@11: return altivec_yuv2_rgb24; yading@11: case AV_PIX_FMT_BGR24: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGR24\n"); yading@11: return altivec_yuv2_bgr24; yading@11: case AV_PIX_FMT_ARGB: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ARGB\n"); yading@11: return altivec_yuv2_argb; yading@11: case AV_PIX_FMT_ABGR: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space ABGR\n"); yading@11: return altivec_yuv2_abgr; yading@11: case AV_PIX_FMT_RGBA: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space RGBA\n"); yading@11: return altivec_yuv2_rgba; yading@11: case AV_PIX_FMT_BGRA: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space BGRA\n"); yading@11: return altivec_yuv2_bgra; yading@11: default: return NULL; yading@11: } yading@11: break; yading@11: yading@11: case AV_PIX_FMT_UYVY422: yading@11: switch (c->dstFormat) { yading@11: case AV_PIX_FMT_BGR32: yading@11: av_log(c, AV_LOG_WARNING, "ALTIVEC: Color Space UYVY -> RGB32\n"); yading@11: return altivec_uyvy_rgb32; yading@11: default: return NULL; yading@11: } yading@11: break; yading@11: } yading@11: return NULL; yading@11: } yading@11: yading@11: av_cold void ff_yuv2rgb_init_tables_altivec(SwsContext *c, yading@11: const int inv_table[4], yading@11: int brightness, yading@11: int contrast, yading@11: int saturation) yading@11: { yading@11: union { yading@11: DECLARE_ALIGNED(16, signed short, tmp)[8]; yading@11: vector signed short vec; yading@11: } buf; yading@11: yading@11: buf.tmp[0] = ((0xffffLL) * contrast >> 8) >> 9; // cy yading@11: buf.tmp[1] = -256 * brightness; // oy yading@11: buf.tmp[2] = (inv_table[0] >> 3) * (contrast >> 16) * (saturation >> 16); // crv yading@11: buf.tmp[3] = (inv_table[1] >> 3) * (contrast >> 16) * (saturation >> 16); // cbu yading@11: buf.tmp[4] = -((inv_table[2] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgu yading@11: buf.tmp[5] = -((inv_table[3] >> 1) * (contrast >> 16) * (saturation >> 16)); // cgv yading@11: yading@11: c->CSHIFT = (vector unsigned short) vec_splat_u16(2); yading@11: c->CY = vec_splat((vector signed short) buf.vec, 0); yading@11: c->OY = vec_splat((vector signed short) buf.vec, 1); yading@11: c->CRV = vec_splat((vector signed short) buf.vec, 2); yading@11: c->CBU = vec_splat((vector signed short) buf.vec, 3); yading@11: c->CGU = vec_splat((vector signed short) buf.vec, 4); yading@11: c->CGV = vec_splat((vector signed short) buf.vec, 5); yading@11: return; yading@11: } yading@11: yading@11: static av_always_inline void ff_yuv2packedX_altivec(SwsContext *c, yading@11: const int16_t *lumFilter, yading@11: const int16_t **lumSrc, yading@11: int lumFilterSize, yading@11: const int16_t *chrFilter, yading@11: const int16_t **chrUSrc, yading@11: const int16_t **chrVSrc, yading@11: int chrFilterSize, yading@11: const int16_t **alpSrc, yading@11: uint8_t *dest, yading@11: int dstW, int dstY, yading@11: enum AVPixelFormat target) yading@11: { yading@11: int i, j; yading@11: vector signed short X, X0, X1, Y0, U0, V0, Y1, U1, V1, U, V; yading@11: vector signed short R0, G0, B0, R1, G1, B1; yading@11: yading@11: vector unsigned char R, G, B; yading@11: vector unsigned char *out, *nout; yading@11: yading@11: vector signed short RND = vec_splat_s16(1 << 3); yading@11: vector unsigned short SCL = vec_splat_u16(4); yading@11: DECLARE_ALIGNED(16, unsigned int, scratch)[16]; yading@11: yading@11: vector signed short *YCoeffs, *CCoeffs; yading@11: yading@11: YCoeffs = c->vYCoeffsBank + dstY * lumFilterSize; yading@11: CCoeffs = c->vCCoeffsBank + dstY * chrFilterSize; yading@11: yading@11: out = (vector unsigned char *) dest; yading@11: yading@11: for (i = 0; i < dstW; i += 16) { yading@11: Y0 = RND; yading@11: Y1 = RND; yading@11: /* extract 16 coeffs from lumSrc */ yading@11: for (j = 0; j < lumFilterSize; j++) { yading@11: X0 = vec_ld(0, &lumSrc[j][i]); yading@11: X1 = vec_ld(16, &lumSrc[j][i]); yading@11: Y0 = vec_mradds(X0, YCoeffs[j], Y0); yading@11: Y1 = vec_mradds(X1, YCoeffs[j], Y1); yading@11: } yading@11: yading@11: U = RND; yading@11: V = RND; yading@11: /* extract 8 coeffs from U,V */ yading@11: for (j = 0; j < chrFilterSize; j++) { yading@11: X = vec_ld(0, &chrUSrc[j][i / 2]); yading@11: U = vec_mradds(X, CCoeffs[j], U); yading@11: X = vec_ld(0, &chrVSrc[j][i / 2]); yading@11: V = vec_mradds(X, CCoeffs[j], V); yading@11: } yading@11: yading@11: /* scale and clip signals */ yading@11: Y0 = vec_sra(Y0, SCL); yading@11: Y1 = vec_sra(Y1, SCL); yading@11: U = vec_sra(U, SCL); yading@11: V = vec_sra(V, SCL); yading@11: yading@11: Y0 = vec_clip_s16(Y0); yading@11: Y1 = vec_clip_s16(Y1); yading@11: U = vec_clip_s16(U); yading@11: V = vec_clip_s16(V); yading@11: yading@11: /* now we have yading@11: * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 yading@11: * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7 yading@11: * yading@11: * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 yading@11: * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7 yading@11: * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7 yading@11: */ yading@11: yading@11: U0 = vec_mergeh(U, U); yading@11: V0 = vec_mergeh(V, V); yading@11: yading@11: U1 = vec_mergel(U, U); yading@11: V1 = vec_mergel(V, V); yading@11: yading@11: cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0); yading@11: cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1); yading@11: yading@11: R = vec_packclp(R0, R1); yading@11: G = vec_packclp(G0, G1); yading@11: B = vec_packclp(B0, B1); yading@11: yading@11: switch (target) { yading@11: case AV_PIX_FMT_ABGR: yading@11: out_abgr(R, G, B, out); yading@11: break; yading@11: case AV_PIX_FMT_BGRA: yading@11: out_bgra(R, G, B, out); yading@11: break; yading@11: case AV_PIX_FMT_RGBA: yading@11: out_rgba(R, G, B, out); yading@11: break; yading@11: case AV_PIX_FMT_ARGB: yading@11: out_argb(R, G, B, out); yading@11: break; yading@11: case AV_PIX_FMT_RGB24: yading@11: out_rgb24(R, G, B, out); yading@11: break; yading@11: case AV_PIX_FMT_BGR24: yading@11: out_bgr24(R, G, B, out); yading@11: break; yading@11: default: yading@11: { yading@11: /* If this is reached, the caller should have called yuv2packedXinC yading@11: * instead. */ yading@11: static int printed_error_message; yading@11: if (!printed_error_message) { yading@11: av_log(c, AV_LOG_ERROR, yading@11: "altivec_yuv2packedX doesn't support %s output\n", yading@11: av_get_pix_fmt_name(c->dstFormat)); yading@11: printed_error_message = 1; yading@11: } yading@11: return; yading@11: } yading@11: } yading@11: } yading@11: yading@11: if (i < dstW) { yading@11: i -= 16; yading@11: yading@11: Y0 = RND; yading@11: Y1 = RND; yading@11: /* extract 16 coeffs from lumSrc */ yading@11: for (j = 0; j < lumFilterSize; j++) { yading@11: X0 = vec_ld(0, &lumSrc[j][i]); yading@11: X1 = vec_ld(16, &lumSrc[j][i]); yading@11: Y0 = vec_mradds(X0, YCoeffs[j], Y0); yading@11: Y1 = vec_mradds(X1, YCoeffs[j], Y1); yading@11: } yading@11: yading@11: U = RND; yading@11: V = RND; yading@11: /* extract 8 coeffs from U,V */ yading@11: for (j = 0; j < chrFilterSize; j++) { yading@11: X = vec_ld(0, &chrUSrc[j][i / 2]); yading@11: U = vec_mradds(X, CCoeffs[j], U); yading@11: X = vec_ld(0, &chrVSrc[j][i / 2]); yading@11: V = vec_mradds(X, CCoeffs[j], V); yading@11: } yading@11: yading@11: /* scale and clip signals */ yading@11: Y0 = vec_sra(Y0, SCL); yading@11: Y1 = vec_sra(Y1, SCL); yading@11: U = vec_sra(U, SCL); yading@11: V = vec_sra(V, SCL); yading@11: yading@11: Y0 = vec_clip_s16(Y0); yading@11: Y1 = vec_clip_s16(Y1); yading@11: U = vec_clip_s16(U); yading@11: V = vec_clip_s16(V); yading@11: yading@11: /* now we have yading@11: * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 yading@11: * U = u0 u1 u2 u3 u4 u5 u6 u7 V = v0 v1 v2 v3 v4 v5 v6 v7 yading@11: * yading@11: * Y0 = y0 y1 y2 y3 y4 y5 y6 y7 Y1 = y8 y9 y10 y11 y12 y13 y14 y15 yading@11: * U0 = u0 u0 u1 u1 u2 u2 u3 u3 U1 = u4 u4 u5 u5 u6 u6 u7 u7 yading@11: * V0 = v0 v0 v1 v1 v2 v2 v3 v3 V1 = v4 v4 v5 v5 v6 v6 v7 v7 yading@11: */ yading@11: yading@11: U0 = vec_mergeh(U, U); yading@11: V0 = vec_mergeh(V, V); yading@11: yading@11: U1 = vec_mergel(U, U); yading@11: V1 = vec_mergel(V, V); yading@11: yading@11: cvtyuvtoRGB(c, Y0, U0, V0, &R0, &G0, &B0); yading@11: cvtyuvtoRGB(c, Y1, U1, V1, &R1, &G1, &B1); yading@11: yading@11: R = vec_packclp(R0, R1); yading@11: G = vec_packclp(G0, G1); yading@11: B = vec_packclp(B0, B1); yading@11: yading@11: nout = (vector unsigned char *) scratch; yading@11: switch (target) { yading@11: case AV_PIX_FMT_ABGR: yading@11: out_abgr(R, G, B, nout); yading@11: break; yading@11: case AV_PIX_FMT_BGRA: yading@11: out_bgra(R, G, B, nout); yading@11: break; yading@11: case AV_PIX_FMT_RGBA: yading@11: out_rgba(R, G, B, nout); yading@11: break; yading@11: case AV_PIX_FMT_ARGB: yading@11: out_argb(R, G, B, nout); yading@11: break; yading@11: case AV_PIX_FMT_RGB24: yading@11: out_rgb24(R, G, B, nout); yading@11: break; yading@11: case AV_PIX_FMT_BGR24: yading@11: out_bgr24(R, G, B, nout); yading@11: break; yading@11: default: yading@11: /* Unreachable, I think. */ yading@11: av_log(c, AV_LOG_ERROR, yading@11: "altivec_yuv2packedX doesn't support %s output\n", yading@11: av_get_pix_fmt_name(c->dstFormat)); yading@11: return; yading@11: } yading@11: yading@11: memcpy(&((uint32_t *) dest)[i], scratch, (dstW - i) / 4); yading@11: } yading@11: } yading@11: yading@11: #define YUV2PACKEDX_WRAPPER(suffix, pixfmt) \ yading@11: void ff_yuv2 ## suffix ## _X_altivec(SwsContext *c, \ yading@11: const int16_t *lumFilter, \ yading@11: const int16_t **lumSrc, \ yading@11: int lumFilterSize, \ yading@11: const int16_t *chrFilter, \ yading@11: const int16_t **chrUSrc, \ yading@11: const int16_t **chrVSrc, \ yading@11: int chrFilterSize, \ yading@11: const int16_t **alpSrc, \ yading@11: uint8_t *dest, int dstW, int dstY) \ yading@11: { \ yading@11: ff_yuv2packedX_altivec(c, lumFilter, lumSrc, lumFilterSize, \ yading@11: chrFilter, chrUSrc, chrVSrc, \ yading@11: chrFilterSize, alpSrc, \ yading@11: dest, dstW, dstY, pixfmt); \ yading@11: } yading@11: yading@11: YUV2PACKEDX_WRAPPER(abgr, AV_PIX_FMT_ABGR); yading@11: YUV2PACKEDX_WRAPPER(bgra, AV_PIX_FMT_BGRA); yading@11: YUV2PACKEDX_WRAPPER(argb, AV_PIX_FMT_ARGB); yading@11: YUV2PACKEDX_WRAPPER(rgba, AV_PIX_FMT_RGBA); yading@11: YUV2PACKEDX_WRAPPER(rgb24, AV_PIX_FMT_RGB24); yading@11: YUV2PACKEDX_WRAPPER(bgr24, AV_PIX_FMT_BGR24);