From: Eric Anholt Date: Fri, 21 Nov 2008 09:09:47 +0000 (+0800) Subject: i965: Add support for accelerated CopyTexSubImage. X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=7e97362884bcaa564730b05c28fb9f33734c7449;p=mesa.git i965: Add support for accelerated CopyTexSubImage. There were hacks in EmitCopyBlit before to adjust offsets so that y=0 after the offsets had been adjusted for a negative pitch. It appears that those hacks were due to an unclear and surprising aspect of the hardware: inverting the pitch results in the blit into the specified rectangle being inverted, without the user needing to adjust y and base offset. Tested with piglit copytexsubimage test on 915GM and GM965. Should fix serious performance issues with ETQW and other applications. --- diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c index e1046f4a5de..ab12aae6c71 100644 --- a/src/mesa/drivers/dri/intel/intel_blit.c +++ b/src/mesa/drivers/dri/intel/intel_blit.c @@ -40,6 +40,7 @@ #include "intel_reg.h" #include "intel_regions.h" #include "intel_batchbuffer.h" +#include "intel_chipset.h" #define FILE_DEBUG_FLAG DEBUG_BLIT @@ -359,51 +360,24 @@ intelEmitCopyBlit(struct intel_context *intel, return; } - /* Initial y values don't seem to work with negative pitches. If - * we adjust the offsets manually (below), it seems to work fine. - * - * On the other hand, if we always adjust, the hardware doesn't - * know which blit directions to use, so overlapping copypixels get - * the wrong result. - */ - if (dst_pitch > 0 && src_pitch > 0) { - assert(dst_x < dst_x2); - assert(dst_y < dst_y2); - - BEGIN_BATCH(8, NO_LOOP_CLIPRECTS); - OUT_BATCH(CMD); - OUT_BATCH(BR13 | dst_pitch); - OUT_BATCH((dst_y << 16) | dst_x); - OUT_BATCH((dst_y2 << 16) | dst_x2); - OUT_RELOC(dst_buffer, - I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, - dst_offset); - OUT_BATCH((src_y << 16) | src_x); - OUT_BATCH(src_pitch); - OUT_RELOC(src_buffer, - I915_GEM_DOMAIN_RENDER, 0, - src_offset); - ADVANCE_BATCH(); - } - else { - assert(dst_x < dst_x2); - assert(h > 0); - - BEGIN_BATCH(8, NO_LOOP_CLIPRECTS); - OUT_BATCH(CMD); - OUT_BATCH(BR13 | ((uint16_t)dst_pitch)); - OUT_BATCH((0 << 16) | dst_x); - OUT_BATCH((h << 16) | dst_x2); - OUT_RELOC(dst_buffer, - I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, - dst_offset + dst_y * dst_pitch); - OUT_BATCH((0 << 16) | src_x); - OUT_BATCH(src_pitch); - OUT_RELOC(src_buffer, - I915_GEM_DOMAIN_RENDER, 0, - src_offset + src_y * src_pitch); - ADVANCE_BATCH(); - } + assert(dst_x < dst_x2); + assert(dst_y < dst_y2); + + BEGIN_BATCH(8, NO_LOOP_CLIPRECTS); + OUT_BATCH(CMD); + OUT_BATCH(BR13 | (uint16_t)dst_pitch); + OUT_BATCH((dst_y << 16) | dst_x); + OUT_BATCH((dst_y2 << 16) | dst_x2); + OUT_RELOC(dst_buffer, + I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, + dst_offset); + OUT_BATCH((src_y << 16) | src_x); + OUT_BATCH((uint16_t)src_pitch); + OUT_RELOC(src_buffer, + I915_GEM_DOMAIN_RENDER, 0, + src_offset); + ADVANCE_BATCH(); + intel_batchbuffer_emit_mi_flush(intel->batch); } diff --git a/src/mesa/drivers/dri/intel/intel_tex.c b/src/mesa/drivers/dri/intel/intel_tex.c index 23455a4c745..82f8b870095 100644 --- a/src/mesa/drivers/dri/intel/intel_tex.c +++ b/src/mesa/drivers/dri/intel/intel_tex.c @@ -222,17 +222,10 @@ intelInitTextureFuncs(struct dd_function_table *functions) functions->TexSubImage1D = intelTexSubImage1D; functions->TexSubImage2D = intelTexSubImage2D; functions->TexSubImage3D = intelTexSubImage3D; -#ifdef I915 functions->CopyTexImage1D = intelCopyTexImage1D; functions->CopyTexImage2D = intelCopyTexImage2D; functions->CopyTexSubImage1D = intelCopyTexSubImage1D; functions->CopyTexSubImage2D = intelCopyTexSubImage2D; -#else - functions->CopyTexImage1D = _swrast_copy_teximage1d; - functions->CopyTexImage2D = _swrast_copy_teximage2d; - functions->CopyTexSubImage1D = _swrast_copy_texsubimage1d; - functions->CopyTexSubImage2D = _swrast_copy_texsubimage2d; -#endif functions->GetTexImage = intelGetTexImage; functions->GenerateMipmap = intelGenerateMipmap; diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c index f4cb4a781c6..36446efde71 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_copy.c +++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c @@ -114,43 +114,49 @@ do_copy_texsubimage(struct intel_context *intel, if (_mesa_clip_to_region(fb->_Xmin, fb->_Ymin, fb->_Xmax, fb->_Ymax, &x, &y, &width, &height)) { + GLshort src_pitch; + /* Update dst for clipped src. Need to also clip the source rect. */ dstx += x - orig_x; dsty += y - orig_y; + /* image_offset may be non-page-aligned, but that's illegal for tiling. + */ + assert(intelImage->mt->region->tiling == I915_TILING_NONE); + if (ctx->ReadBuffer->Name == 0) { /* reading from a window, adjust x, y */ __DRIdrawablePrivate *dPriv = intel->driDrawable; - GLuint window_y; - /* window_y = position of window on screen if y=0=bottom */ - window_y = intel->intelScreen->height - (dPriv->y + dPriv->h); - y = window_y + y; + y = dPriv->y + (dPriv->h - (y + height)); x += dPriv->x; + + /* Invert the data coming from the source rectangle due to GL + * and hardware disagreeing on where y=0 is. + * + * It appears that our offsets and pitches get mangled + * appropriately by the hardware, and we don't need to adjust them + * on our own. + */ + src_pitch = -src->pitch; } else { - /* reading from a FBO */ - /* invert Y */ - y = ctx->ReadBuffer->Height - y - 1; + /* reading from a FBO, y is already oriented the way we like */ + src_pitch = src->pitch; } - - /* A bit of fiddling to get the blitter to work with -ve - * pitches. But we get a nice inverted blit this way, so it's - * worth it: - */ intelEmitCopyBlit(intel, intelImage->mt->cpp, - -src->pitch, + src_pitch, src->buffer, - src->height * src->pitch * src->cpp, + 0, src->tiling, intelImage->mt->pitch, intelImage->mt->region->buffer, image_offset, intelImage->mt->region->tiling, - x, y + height, dstx, dsty, width, height, - GL_COPY); /* ? */ + x, y, dstx, dsty, width, height, + GL_COPY); } }