This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project gfxprim.git. The branch, generate has been updated via 78f907f5c2ec5e07373ddfc04774c9a78639a44d (commit) from 2fae7f19796fbac3a1e8a05891838595669c9407 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- http://repo.or.cz/w/gfxprim.git/commit/78f907f5c2ec5e07373ddfc04774c9a78639a... commit 78f907f5c2ec5e07373ddfc04774c9a78639a44d Author: Cyril Hrubis <metan@ucw.cz> Date: Wed Sep 28 12:52:27 2011 +0200 Unrolled inner loops and used gcc vector aritmetics. diff --git a/libs/filters/GP_ScaleDown.c b/libs/filters/GP_ScaleDown.c index da5d9d3..90ba812 100644 --- a/libs/filters/GP_ScaleDown.c +++ b/libs/filters/GP_ScaleDown.c @@ -69,11 +69,21 @@ float cubic(float x) return 0; } +typedef float v4sf __attribute__ ((vector_size (sizeof(float) * 4))); + +typedef union v4f { + v4sf v; + float f[4]; +} v4f; + +#define MUL_V4SF(a, b) ((union v4f)((a).v * (b).v)) +#define SUM_V4SF(a) ((a).f[0] + (a).f[1] + (a).f[2] + (a).f[3]) + GP_Context *GP_Scale(GP_Context *src, GP_Size w, GP_Size h) { GP_Context *dst; float col_r[src->h], col_g[src->h], col_b[src->h]; - uint32_t i, j, k; + uint32_t i, j; int idx; if (src->pixel_type != GP_PIXEL_RGB888) @@ -89,55 +99,93 @@ GP_Context *GP_Scale(GP_Context *src, GP_Size w, GP_Size h) /* Generate interpolated column */ for (j = 0; j < src->h; j++) { - uint8_t r[4], g[4], b[4]; - float c[4]; - - for (k = 0; k < 4; k++) { - GP_Pixel pix; - idx = x + k - 1; + v4f cv, rv, gv, bv; + GP_Pixel pix[4]; + + idx = x - 1; - if (idx < 0) - idx = 0; + if (idx < 0) + idx = 0; - if (idx > (int)src->w - 1) - idx = src->w - 1; + if (idx > (int)src->w - 4) + idx = src->w - 4; - pix = GP_GetPixel_Raw_24BPP(src, idx, j); + pix[0] = GP_GetPixel_Raw_24BPP(src, idx, j); + pix[1] = GP_GetPixel_Raw_24BPP(src, idx + 1, j); + pix[2] = GP_GetPixel_Raw_24BPP(src, idx + 2, j); + pix[3] = GP_GetPixel_Raw_24BPP(src, idx + 3, j); - r[k] = GP_Pixel_GET_R_RGB888(pix); - g[k] = GP_Pixel_GET_G_RGB888(pix); - b[k] = GP_Pixel_GET_B_RGB888(pix); + rv.f[0] = GP_Pixel_GET_R_RGB888(pix[0]); + rv.f[1] = GP_Pixel_GET_R_RGB888(pix[1]); + rv.f[2] = GP_Pixel_GET_R_RGB888(pix[2]); + rv.f[3] = GP_Pixel_GET_R_RGB888(pix[3]); - c[k] = cubic(x - idx); - } - - col_r[j] = r[0] * c[0] + r[1] * c[1] + r[2] * c[2] + r[3] * c[3]; - col_g[j] = g[0] * c[0] + g[1] * c[1] + g[2] * c[2] + g[3] * c[3]; - col_b[j] = b[0] * c[0] + b[1] * c[1] + b[2] * c[2] + b[3] * c[3]; - + gv.f[0] = GP_Pixel_GET_G_RGB888(pix[0]); + gv.f[1] = GP_Pixel_GET_G_RGB888(pix[1]); + gv.f[2] = GP_Pixel_GET_G_RGB888(pix[2]); + gv.f[3] = GP_Pixel_GET_G_RGB888(pix[3]); + + bv.f[0] = GP_Pixel_GET_B_RGB888(pix[0]); + bv.f[1] = GP_Pixel_GET_B_RGB888(pix[1]); + bv.f[2] = GP_Pixel_GET_B_RGB888(pix[2]); + bv.f[3] = GP_Pixel_GET_B_RGB888(pix[3]); + + cv.f[0] = cubic(x - idx); + cv.f[1] = cubic(x - idx - 1); + cv.f[2] = cubic(x - idx - 2); + cv.f[3] = cubic(x - idx - 3); + + rv = MUL_V4SF(rv, cv); + gv = MUL_V4SF(gv, cv); + bv = MUL_V4SF(bv, cv); + + col_r[j] = SUM_V4SF(rv); + col_g[j] = SUM_V4SF(gv); + col_b[j] = SUM_V4SF(bv); } /* now interpolate column for new image */ for (j = 0; j < h; j++) { float y = (1.00 * j / h) * src->h + 0.5; - float r = 0, g = 0, b = 0, c; - - for (k = 0; k < 4; k++) { - idx = y + k - 1; + v4f cv, rv, gv, bv; + float r, g, b; + + idx = y - 1; - if (idx < 0) - idx = 0; - - if (idx > (int)src->h - 1) - idx = src->h - 1; - - c = cubic(y - idx); - - r += col_r[idx] * c; - g += col_g[idx] * c; - b += col_b[idx] * c; - } + if (idx < 0) + idx = 0; + + if (idx > (int)src->h - 4) + idx = src->h - 4; + + rv.f[0] = col_r[idx]; + rv.f[1] = col_r[idx + 1]; + rv.f[2] = col_r[idx + 2]; + rv.f[3] = col_r[idx + 3]; + + gv.f[0] = col_g[idx]; + gv.f[1] = col_g[idx + 1]; + gv.f[2] = col_g[idx + 2]; + gv.f[3] = col_g[idx + 3]; + bv.f[0] = col_b[idx]; + bv.f[1] = col_b[idx + 1]; + bv.f[2] = col_b[idx + 2]; + bv.f[3] = col_b[idx + 3]; + + cv.f[0] = cubic(y - idx); + cv.f[1] = cubic(y - idx - 1); + cv.f[2] = cubic(y - idx - 2); + cv.f[3] = cubic(y - idx - 3); + + rv = MUL_V4SF(rv, cv); + gv = MUL_V4SF(gv, cv); + bv = MUL_V4SF(bv, cv); + + r = SUM_V4SF(rv); + g = SUM_V4SF(gv); + b = SUM_V4SF(bv); + if (r > 255) r = 255; ----------------------------------------------------------------------- Summary of changes: libs/filters/GP_ScaleDown.c | 124 ++++++++++++++++++++++++++++++------------- 1 files changed, 86 insertions(+), 38 deletions(-) repo.or.cz automatic notification. Contact project admin jiri.bluebear.dluhos@gmail.com if you want to unsubscribe, or site admin admin@repo.or.cz if you receive no reply. -- gfxprim.git ("A simple 2D graphics library with emphasis on correctness and well-defined operation.")