X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi965%2Fintel_pixel_read.c;h=cf957378f925ea636d477c4730d652570fa210ed;hb=92f01fc5f914fd500497d0c3aed75f3ac8dc054d;hp=600d288994bcf8ee8d5daf9c620c51cd409f435a;hpb=877128505431adaf817dc8069172ebe4a1cdf5d8;p=mesa.git diff --git a/src/mesa/drivers/dri/i965/intel_pixel_read.c b/src/mesa/drivers/dri/i965/intel_pixel_read.c index 600d288994b..cf957378f92 100644 --- a/src/mesa/drivers/dri/i965/intel_pixel_read.c +++ b/src/mesa/drivers/dri/i965/intel_pixel_read.c @@ -1,5 +1,4 @@ -/************************************************************************** - * +/* * Copyright 2003 VMware, Inc. * All Rights Reserved. * @@ -7,7 +6,7 @@ * copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including * without limitation the rights to use, copy, modify, merge, publish, - * distribute, sub license, and/or sell copies of the Software, and to + * distribute, sublicense, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: * @@ -17,15 +16,13 @@ * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF - * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. - * - **************************************************************************/ + */ -#include "main/glheader.h" #include "main/enums.h" #include "main/mtypes.h" #include "main/macros.h" @@ -34,153 +31,260 @@ #include "main/bufferobj.h" #include "main/readpix.h" #include "main/state.h" +#include "main/glformats.h" +#include "program/prog_instruction.h" +#include "drivers/common/meta.h" #include "brw_context.h" +#include "brw_blorp.h" #include "intel_screen.h" +#include "intel_batchbuffer.h" #include "intel_blit.h" #include "intel_buffers.h" #include "intel_fbo.h" #include "intel_mipmap_tree.h" -#include "intel_regions.h" #include "intel_pixel.h" #include "intel_buffer_objects.h" +#include "intel_tiled_memcpy.h" #define FILE_DEBUG_FLAG DEBUG_PIXEL -/* For many applications, the new ability to pull the source buffers - * back out of the GTT and then do the packing/conversion operations - * in software will be as much of an improvement as trying to get the - * blitter and/or texture engine to do the work. - * - * This step is gated on private backbuffers. +/** + * \brief A fast path for glReadPixels * - * Obviously the frontbuffer can't be pulled back, so that is either - * an argument for blit/texture readpixels, or for blitting to a - * temporary and then pulling that back. + * This fast path is taken when the source format is BGRA, RGBA, + * A or L and when the texture memory is X- or Y-tiled. It downloads + * the source data by directly mapping the memory without a GTT fence. + * This then needs to be de-tiled on the CPU before presenting the data to + * the user in the linear fasion. * - * When the destination is a pbo, however, it's not clear if it is - * ever going to be pulled to main memory (though the access param - * will be a good hint). So it sounds like we do want to be able to - * choose between blit/texture implementation on the gpu and pullback - * and cpu-based copying. - * - * Unless you can magically turn client memory into a PBO for the - * duration of this call, there will be a cpu-based copying step in - * any case. + * This is a performance win over the conventional texture download path. + * In the conventional texture download path, the texture is either mapped + * through the GTT or copied to a linear buffer with the blitter before + * handing off to a software path. This allows us to avoid round-tripping + * through the GPU (in the case where we would be blitting) and do only a + * single copy operation. */ - static bool -do_blit_readpixels(struct gl_context * ctx, - GLint x, GLint y, GLsizei width, GLsizei height, - GLenum format, GLenum type, - const struct gl_pixelstore_attrib *pack, GLvoid * pixels) +intel_readpixels_tiled_memcpy(struct gl_context * ctx, + GLint xoffset, GLint yoffset, + GLsizei width, GLsizei height, + GLenum format, GLenum type, + GLvoid * pixels, + const struct gl_pixelstore_attrib *pack) { struct brw_context *brw = brw_context(ctx); - struct intel_buffer_object *dst = intel_buffer_object(pack->BufferObj); - GLuint dst_offset; - drm_intel_bo *dst_buffer; - GLint dst_x, dst_y; - GLuint dirty; - - DBG("%s\n", __FUNCTION__); + struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer; + const struct gen_device_info *devinfo = &brw->screen->devinfo; - assert(_mesa_is_bufferobj(pack->BufferObj)); + /* This path supports reading from color buffers only */ + if (rb == NULL) + return false; - struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer; struct intel_renderbuffer *irb = intel_renderbuffer(rb); + int dst_pitch; + + /* The miptree's buffer. */ + struct brw_bo *bo; + + uint32_t cpp; + mem_copy_fn mem_copy = NULL; - if (ctx->_ImageTransferState || - !_mesa_format_matches_format_and_type(irb->mt->format, format, type, - false)) { - DBG("%s - bad format for blit\n", __FUNCTION__); + /* This fastpath is restricted to specific renderbuffer types: + * a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support + * more types. + */ + if (!devinfo->has_llc || + !(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) || + pixels == NULL || + _mesa_is_bufferobj(pack->BufferObj) || + pack->Alignment > 4 || + pack->SkipPixels > 0 || + pack->SkipRows > 0 || + (pack->RowLength != 0 && pack->RowLength != width) || + pack->SwapBytes || + pack->LsbFirst || + pack->Invert) return false; - } - if (pack->SwapBytes || pack->LsbFirst) { - DBG("%s: bad packing params\n", __FUNCTION__); + /* Only a simple blit, no scale, bias or other mapping. */ + if (ctx->_ImageTransferState) return false; - } - int dst_stride = _mesa_image_row_stride(pack, width, format, type); - bool dst_flip = false; - /* Mesa flips the dst_stride for pack->Invert, but we want our mt to have a - * normal dst_stride. + /* It is possible that the renderbuffer (or underlying texture) is + * multisampled. Since ReadPixels from a multisampled buffer requires a + * multisample resolve, we can't handle this here */ - struct gl_pixelstore_attrib uninverted_pack = *pack; - if (pack->Invert) { - dst_stride = -dst_stride; - dst_flip = true; - uninverted_pack.Invert = false; - } + if (rb->NumSamples > 1) + return false; + + /* We can't handle copying from RGBX or BGRX because the tiled_memcpy + * function doesn't set the last channel to 1. Note this checks BaseFormat + * rather than TexFormat in case the RGBX format is being simulated with an + * RGBA format. + */ + if (rb->_BaseFormat == GL_RGB) + return false; - dst_offset = (GLintptr)pixels; - dst_offset += _mesa_image_offset(2, &uninverted_pack, width, height, - format, type, 0, 0, 0); + if (!intel_get_memcpy(rb->Format, format, type, &mem_copy, &cpp)) + return false; - if (!_mesa_clip_copytexsubimage(ctx, - &dst_x, &dst_y, - &x, &y, - &width, &height)) { - return true; + if (!irb->mt || + (irb->mt->surf.tiling != ISL_TILING_X && + irb->mt->surf.tiling != ISL_TILING_Y0)) { + /* The algorithm is written only for X- or Y-tiled memory. */ + return false; } - dirty = brw->front_buffer_dirty; - intel_prepare_render(brw); - brw->front_buffer_dirty = dirty; + /* tiled_to_linear() assumes that if the object is swizzled, it is using + * I915_BIT6_SWIZZLE_9_10 for X and I915_BIT6_SWIZZLE_9 for Y. This is only + * true on gen5 and above. + * + * The killer on top is that some gen4 have an L-shaped swizzle mode, where + * parts of the memory aren't swizzled at all. Userspace just can't handle + * that. + */ + if (devinfo->gen < 5 && brw->has_swizzling) + return false; + + /* Since we are going to read raw data to the miptree, we need to resolve + * any pending fast color clears before we start. + */ + intel_miptree_access_raw(brw, irb->mt, irb->mt_level, irb->mt_layer, false); + + bo = irb->mt->bo; + + if (brw_batch_references(&brw->batch, bo)) { + perf_debug("Flushing before mapping a referenced bo.\n"); + intel_batchbuffer_flush(brw); + } - dst_buffer = intel_bufferobj_buffer(brw, dst, - dst_offset, height * dst_stride); - - struct intel_mipmap_tree *pbo_mt = - intel_miptree_create_for_bo(brw, - dst_buffer, - irb->mt->format, - dst_offset, - width, height, - dst_stride, I915_TILING_NONE); - - if (!intel_miptree_blit(brw, - irb->mt, irb->mt_level, irb->mt_layer, - x, y, _mesa_is_winsys_fbo(ctx->ReadBuffer), - pbo_mt, 0, 0, - 0, 0, dst_flip, - width, height, GL_COPY)) { + void *map = brw_bo_map(brw, bo, MAP_READ | MAP_RAW); + if (map == NULL) { + DBG("%s: failed to map bo\n", __func__); return false; } - intel_miptree_release(&pbo_mt); + unsigned slice_offset_x, slice_offset_y; + intel_miptree_get_image_offset(irb->mt, irb->mt_level, irb->mt_layer, + &slice_offset_x, &slice_offset_y); + xoffset += slice_offset_x; + yoffset += slice_offset_y; - DBG("%s - DONE\n", __FUNCTION__); + dst_pitch = _mesa_image_row_stride(pack, width, format, type); + /* For a window-system renderbuffer, the buffer is actually flipped + * vertically, so we need to handle that. Since the detiling function + * can only really work in the forwards direction, we have to be a + * little creative. First, we compute the Y-offset of the first row of + * the renderbuffer (in renderbuffer coordinates). We then match that + * with the last row of the client's data. Finally, we give + * tiled_to_linear a negative pitch so that it walks through the + * client's data backwards as it walks through the renderbufer forwards. + */ + if (rb->Name == 0) { + yoffset = rb->Height - yoffset - height; + pixels += (ptrdiff_t) (height - 1) * dst_pitch; + dst_pitch = -dst_pitch; + } + + /* We postponed printing this message until having committed to executing + * the function. + */ + DBG("%s: x,y=(%d,%d) (w,h)=(%d,%d) format=0x%x type=0x%x " + "mesa_format=0x%x tiling=%d " + "pack=(alignment=%d row_length=%d skip_pixels=%d skip_rows=%d)\n", + __func__, xoffset, yoffset, width, height, + format, type, rb->Format, irb->mt->surf.tiling, + pack->Alignment, pack->RowLength, pack->SkipPixels, + pack->SkipRows); + + tiled_to_linear( + xoffset * cpp, (xoffset + width) * cpp, + yoffset, yoffset + height, + pixels, + map + irb->mt->offset, + dst_pitch, irb->mt->surf.row_pitch, + brw->has_swizzling, + irb->mt->surf.tiling, + mem_copy + ); + + brw_bo_unmap(bo); return true; } +static bool +intel_readpixels_blorp(struct gl_context *ctx, + unsigned x, unsigned y, + unsigned w, unsigned h, + GLenum format, GLenum type, const void *pixels, + const struct gl_pixelstore_attrib *packing) +{ + struct brw_context *brw = brw_context(ctx); + struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer; + if (!rb) + return false; + + struct intel_renderbuffer *irb = intel_renderbuffer(rb); + + /* _mesa_get_readpixels_transfer_ops() includes the cases of read + * color clamping along with the ctx->_ImageTransferState. + */ + if (_mesa_get_readpixels_transfer_ops(ctx, rb->Format, format, + type, GL_FALSE)) + return false; + + GLenum dst_base_format = _mesa_unpack_format_to_base_format(format); + if (_mesa_need_rgb_to_luminance_conversion(rb->_BaseFormat, + dst_base_format)) + return false; + + unsigned swizzle; + if (irb->Base.Base._BaseFormat == GL_RGB) { + swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE); + } else { + swizzle = SWIZZLE_XYZW; + } + + return brw_blorp_download_miptree(brw, irb->mt, rb->Format, swizzle, + irb->mt_level, x, y, irb->mt_layer, + w, h, 1, GL_TEXTURE_2D, format, type, + rb->Name == 0, pixels, packing); +} + void intelReadPixels(struct gl_context * ctx, GLint x, GLint y, GLsizei width, GLsizei height, GLenum format, GLenum type, const struct gl_pixelstore_attrib *pack, GLvoid * pixels) { + bool ok; + struct brw_context *brw = brw_context(ctx); bool dirty; - DBG("%s\n", __FUNCTION__); + DBG("%s\n", __func__); + + /* Reading pixels wont dirty the front buffer, so reset the dirty + * flag after calling intel_prepare_render(). + */ + dirty = brw->front_buffer_dirty; + intel_prepare_render(brw); + brw->front_buffer_dirty = dirty; if (_mesa_is_bufferobj(pack->BufferObj)) { - /* Using PBOs, so try the BLT based path. */ - if (do_blit_readpixels(ctx, x, y, width, height, format, type, pack, - pixels)) { + if (intel_readpixels_blorp(ctx, x, y, width, height, + format, type, pixels, pack)) return; - } - perf_debug("%s: fallback to CPU mapping in PBO case\n", __FUNCTION__); + perf_debug("%s: fallback to CPU mapping in PBO case\n", __func__); } - /* glReadPixels() wont dirty the front buffer, so reset the dirty - * flag after calling intel_prepare_render(). */ - dirty = brw->front_buffer_dirty; - intel_prepare_render(brw); - brw->front_buffer_dirty = dirty; + ok = intel_readpixels_tiled_memcpy(ctx, x, y, width, height, + format, type, pixels, pack); + if(ok) + return; /* Update Mesa state before calling _mesa_readpixels(). * XXX this may not be needed since ReadPixels no longer uses the