1506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/* 2506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Copyright © 2012 Intel Corporation 3506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 4506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Permission is hereby granted, free of charge, to any person obtaining a 5506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * copy of this software and associated documentation files (the "Software"), 6506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * to deal in the Software without restriction, including without limitation 7506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * and/or sell copies of the Software, and to permit persons to whom the 9506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Software is furnished to do so, subject to the following conditions: 10506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 11506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * The above copyright notice and this permission notice (including the next 12506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * paragraph) shall be included in all copies or substantial portions of the 13506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Software. 14506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 15506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * IN THE SOFTWARE. 22506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 23506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 24506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "main/teximage.h" 25c738ea1191cd1b5a0dc60b0e6d05fd918083e961Paul Berry#include "main/fbobject.h" 26506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 27506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "glsl/ralloc.h" 28506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 29506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "intel_fbo.h" 30506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 31506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "brw_blorp.h" 32506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "brw_context.h" 33506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "brw_eu.h" 34506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#include "brw_state.h" 35506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 36506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 37506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 38506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Helper function for handling mirror image blits. 39506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 40506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * If coord0 > coord1, swap them and invert the "mirror" boolean. 41506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 42506d70be21cd3469118de89297cba0c0f709c1aePaul Berrystatic inline void 43506d70be21cd3469118de89297cba0c0f709c1aePaul Berryfixup_mirroring(bool &mirror, GLint &coord0, GLint &coord1) 44506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 45506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (coord0 > coord1) { 46506d70be21cd3469118de89297cba0c0f709c1aePaul Berry mirror = !mirror; 47506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLint tmp = coord0; 48506d70be21cd3469118de89297cba0c0f709c1aePaul Berry coord0 = coord1; 49506d70be21cd3469118de89297cba0c0f709c1aePaul Berry coord1 = tmp; 50506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 51506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 52506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 53506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 5447b64c9290d54f78e5a20e378593977cd47e285fPaul Berry/** 5547b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * Adjust {src,dst}_x{0,1} to account for clipping and scissoring of 5647b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * destination coordinates. 5747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * 5847b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * Return true if there is still blitting to do, false if all pixels got 5947b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * rejected by the clip and/or scissor. 6047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * 6147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * For clarity, the nomenclature of this function assumes we are clipping and 6247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * scissoring the X coordinate; the exact same logic applies for Y 6347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * coordinates. 6475f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry * 6575f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry * Note: this function may also be used to account for clipping of source 6675f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry * coordinates, by swapping the roles of src and dst. 6747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry */ 6847b64c9290d54f78e5a20e378593977cd47e285fPaul Berrystatic inline bool 6947b64c9290d54f78e5a20e378593977cd47e285fPaul Berryclip_or_scissor(bool mirror, GLint &src_x0, GLint &src_x1, GLint &dst_x0, 7047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry GLint &dst_x1, GLint fb_xmin, GLint fb_xmax) 7147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry{ 7247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry /* If we are going to scissor everything away, stop. */ 7347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry if (!(fb_xmin < fb_xmax && 7447b64c9290d54f78e5a20e378593977cd47e285fPaul Berry dst_x0 < fb_xmax && 7547b64c9290d54f78e5a20e378593977cd47e285fPaul Berry fb_xmin < dst_x1 && 7647b64c9290d54f78e5a20e378593977cd47e285fPaul Berry dst_x0 < dst_x1)) { 7747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry return false; 7847b64c9290d54f78e5a20e378593977cd47e285fPaul Berry } 7947b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 8047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry /* Clip the destination rectangle, and keep track of how many pixels we 8147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * clipped off of the left and right sides of it. 8247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry */ 8347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry GLint pixels_clipped_left = 0; 8447b64c9290d54f78e5a20e378593977cd47e285fPaul Berry GLint pixels_clipped_right = 0; 8547b64c9290d54f78e5a20e378593977cd47e285fPaul Berry if (dst_x0 < fb_xmin) { 8647b64c9290d54f78e5a20e378593977cd47e285fPaul Berry pixels_clipped_left = fb_xmin - dst_x0; 8747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry dst_x0 = fb_xmin; 8847b64c9290d54f78e5a20e378593977cd47e285fPaul Berry } 8947b64c9290d54f78e5a20e378593977cd47e285fPaul Berry if (fb_xmax < dst_x1) { 9047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry pixels_clipped_right = dst_x1 - fb_xmax; 9147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry dst_x1 = fb_xmax; 9247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry } 9347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 9447b64c9290d54f78e5a20e378593977cd47e285fPaul Berry /* If we are mirrored, then before applying pixels_clipped_{left,right} to 9547b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * the source coordinates, we need to flip them to account for the 9647b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * mirroring. 9747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry */ 9847b64c9290d54f78e5a20e378593977cd47e285fPaul Berry if (mirror) { 9947b64c9290d54f78e5a20e378593977cd47e285fPaul Berry GLint tmp = pixels_clipped_left; 10047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry pixels_clipped_left = pixels_clipped_right; 10147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry pixels_clipped_right = tmp; 10247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry } 10347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 10447b64c9290d54f78e5a20e378593977cd47e285fPaul Berry /* Adjust the source rectangle to remove the pixels corresponding to those 10547b64c9290d54f78e5a20e378593977cd47e285fPaul Berry * that were clipped/scissored out of the destination rectangle. 10647b64c9290d54f78e5a20e378593977cd47e285fPaul Berry */ 10747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry src_x0 += pixels_clipped_left; 10847b64c9290d54f78e5a20e378593977cd47e285fPaul Berry src_x1 -= pixels_clipped_right; 10947b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 11047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry return true; 11147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry} 11247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 11347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 114fa1d267beb4adb542ea90b805306599f602c38d2Paul Berrystatic struct intel_mipmap_tree * 1155c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berryfind_miptree(GLbitfield buffer_bit, struct intel_renderbuffer *irb) 116fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry{ 117fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry struct intel_mipmap_tree *mt = irb->mt; 118fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry if (buffer_bit == GL_STENCIL_BUFFER_BIT && mt->stencil_mt) 119fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry mt = mt->stencil_mt; 120fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry return mt; 121fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry} 122fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 1236cc9df331b4799715b31d7ec606ad09fa914e260Chad Versacevoid 1246cc9df331b4799715b31d7ec606ad09fa914e260Chad Versacebrw_blorp_blit_miptrees(struct intel_context *intel, 1256cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace struct intel_mipmap_tree *src_mt, 126e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry unsigned src_level, unsigned src_layer, 1276cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace struct intel_mipmap_tree *dst_mt, 128e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry unsigned dst_level, unsigned dst_layer, 1296cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace int src_x0, int src_y0, 1306cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace int dst_x0, int dst_y0, 1316cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace int dst_x1, int dst_y1, 1326cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace bool mirror_x, bool mirror_y) 1336cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace{ 1346cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace brw_blorp_blit_params params(brw_context(&intel->ctx), 135e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry src_mt, src_level, src_layer, 136e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry dst_mt, dst_level, dst_layer, 1376cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace src_x0, src_y0, 1386cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace dst_x0, dst_y0, 1396cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace dst_x1, dst_y1, 1406cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace mirror_x, mirror_y); 1416cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace brw_blorp_exec(intel, ¶ms); 1426cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace} 143fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 144fa1d267beb4adb542ea90b805306599f602c38d2Paul Berrystatic void 145fa1d267beb4adb542ea90b805306599f602c38d2Paul Berrydo_blorp_blit(struct intel_context *intel, GLbitfield buffer_bit, 1465c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_renderbuffer *src_irb, 1475c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_renderbuffer *dst_irb, 148fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry GLint srcX0, GLint srcY0, 149fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, 150fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry bool mirror_x, bool mirror_y) 151fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry{ 152fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry /* Find source/dst miptrees */ 1535c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_mipmap_tree *src_mt = find_miptree(buffer_bit, src_irb); 1545c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_mipmap_tree *dst_mt = find_miptree(buffer_bit, dst_irb); 155fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 156fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry /* Get ready to blit. This includes depth resolving the src and dst 157fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry * buffers if necessary. 158fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry */ 1595c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer_resolve_depth(intel, src_irb); 1605c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer_resolve_depth(intel, dst_irb); 161fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 162fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry /* Do the blit */ 163e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry brw_blorp_blit_miptrees(intel, 164e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry src_mt, src_irb->mt_level, src_irb->mt_layer, 165e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry dst_mt, dst_irb->mt_level, dst_irb->mt_layer, 1666cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace srcX0, srcY0, dstX0, dstY0, dstX1, dstY1, 1676cc9df331b4799715b31d7ec606ad09fa914e260Chad Versace mirror_x, mirror_y); 168fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 1695c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer_set_needs_hiz_resolve(dst_irb); 1705c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer_set_needs_downsample(dst_irb); 171fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry} 172fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 173fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 174fa1d267beb4adb542ea90b805306599f602c38d2Paul Berrystatic bool 1755c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berryformats_match(GLbitfield buffer_bit, struct intel_renderbuffer *src_irb, 1765c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_renderbuffer *dst_irb) 177fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry{ 178fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry /* Note: don't just check gl_renderbuffer::Format, because in some cases 179fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry * multiple gl_formats resolve to the same native type in the miptree (for 180fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry * example MESA_FORMAT_X8_Z24 and MESA_FORMAT_S8_Z24), and we can blit 181fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry * between those formats. 182fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry */ 1838c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke gl_format src_format = find_miptree(buffer_bit, src_irb)->format; 1848c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke gl_format dst_format = find_miptree(buffer_bit, dst_irb)->format; 1858c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke 1868c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke return _mesa_get_srgb_format_linear(src_format) == 1878c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke _mesa_get_srgb_format_linear(dst_format); 188fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry} 189fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 190fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry 191506d70be21cd3469118de89297cba0c0f709c1aePaul Berrystatic bool 192506d70be21cd3469118de89297cba0c0f709c1aePaul Berrytry_blorp_blit(struct intel_context *intel, 193506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, 194506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, 195506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLenum filter, GLbitfield buffer_bit) 196506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 197506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct gl_context *ctx = &intel->ctx; 198506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 199506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Sync up the state of window system buffers. We need to do this before 200506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * we go looking for the buffers. 201506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 202506d70be21cd3469118de89297cba0c0f709c1aePaul Berry intel_prepare_render(intel); 203506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 204506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const struct gl_framebuffer *read_fb = ctx->ReadBuffer; 205506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const struct gl_framebuffer *draw_fb = ctx->DrawBuffer; 206506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 207506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Detect if the blit needs to be mirrored */ 208506d70be21cd3469118de89297cba0c0f709c1aePaul Berry bool mirror_x = false, mirror_y = false; 209506d70be21cd3469118de89297cba0c0f709c1aePaul Berry fixup_mirroring(mirror_x, srcX0, srcX1); 210506d70be21cd3469118de89297cba0c0f709c1aePaul Berry fixup_mirroring(mirror_x, dstX0, dstX1); 211506d70be21cd3469118de89297cba0c0f709c1aePaul Berry fixup_mirroring(mirror_y, srcY0, srcY1); 212506d70be21cd3469118de89297cba0c0f709c1aePaul Berry fixup_mirroring(mirror_y, dstY0, dstY1); 213506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 214506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Make sure width and height match */ 215da54d2e576426122009be083ecbfb9eefd8a3799Paul Berry if (srcX1 - srcX0 != dstX1 - dstX0) return false; 216da54d2e576426122009be083ecbfb9eefd8a3799Paul Berry if (srcY1 - srcY0 != dstY1 - dstY0) return false; 217506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 21847b64c9290d54f78e5a20e378593977cd47e285fPaul Berry /* If the destination rectangle needs to be clipped or scissored, do so. 219506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 22047b64c9290d54f78e5a20e378593977cd47e285fPaul Berry if (!(clip_or_scissor(mirror_x, srcX0, srcX1, dstX0, dstX1, 22147b64c9290d54f78e5a20e378593977cd47e285fPaul Berry draw_fb->_Xmin, draw_fb->_Xmax) && 22247b64c9290d54f78e5a20e378593977cd47e285fPaul Berry clip_or_scissor(mirror_y, srcY0, srcY1, dstY0, dstY1, 22347b64c9290d54f78e5a20e378593977cd47e285fPaul Berry draw_fb->_Ymin, draw_fb->_Ymax))) { 22447b64c9290d54f78e5a20e378593977cd47e285fPaul Berry /* Everything got clipped/scissored away, so the blit was successful. */ 22547b64c9290d54f78e5a20e378593977cd47e285fPaul Berry return true; 22647b64c9290d54f78e5a20e378593977cd47e285fPaul Berry } 22747b64c9290d54f78e5a20e378593977cd47e285fPaul Berry 22875f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry /* If the source rectangle needs to be clipped or scissored, do so. */ 22975f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry if (!(clip_or_scissor(mirror_x, dstX0, dstX1, srcX0, srcX1, 23075f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry 0, read_fb->Width) && 23175f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry clip_or_scissor(mirror_y, dstY0, dstY1, srcY0, srcY1, 23275f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry 0, read_fb->Height))) { 23375f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry /* Everything got clipped/scissored away, so the blit was successful. */ 23475f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry return true; 23575f409d75cacf90df2d6f1d718251a5d5cd92f7fPaul Berry } 236506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 2370dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry /* Account for the fact that in the system framebuffer, the origin is at 2380dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry * the lower left. 2390dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry */ 240c738ea1191cd1b5a0dc60b0e6d05fd918083e961Paul Berry if (_mesa_is_winsys_fbo(read_fb)) { 2410dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry GLint tmp = read_fb->Height - srcY0; 2420dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry srcY0 = read_fb->Height - srcY1; 2430dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry srcY1 = tmp; 2440dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry mirror_y = !mirror_y; 2450dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry } 246c738ea1191cd1b5a0dc60b0e6d05fd918083e961Paul Berry if (_mesa_is_winsys_fbo(draw_fb)) { 2470dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry GLint tmp = draw_fb->Height - dstY0; 2480dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry dstY0 = draw_fb->Height - dstY1; 2490dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry dstY1 = tmp; 2500dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry mirror_y = !mirror_y; 2510dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry } 2520dbec6ae07e7b3d566cc397ab09caa413e412846Paul Berry 253fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry /* Find buffers */ 2545c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_renderbuffer *src_irb; 2555c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry struct intel_renderbuffer *dst_irb; 256fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry switch (buffer_bit) { 257fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry case GL_COLOR_BUFFER_BIT: 2585c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry src_irb = intel_renderbuffer(read_fb->_ColorReadBuffer); 259ff9313fac70fa85d051dd4d2b9d3402d39f67ceaPaul Berry for (unsigned i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; ++i) { 2605c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry dst_irb = intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[i]); 2615c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry if (dst_irb && !formats_match(buffer_bit, src_irb, dst_irb)) 262ff9313fac70fa85d051dd4d2b9d3402d39f67ceaPaul Berry return false; 263ff9313fac70fa85d051dd4d2b9d3402d39f67ceaPaul Berry } 264ff9313fac70fa85d051dd4d2b9d3402d39f67ceaPaul Berry for (unsigned i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; ++i) { 2655c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry dst_irb = intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[i]); 2665c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry do_blorp_blit(intel, buffer_bit, src_irb, dst_irb, srcX0, srcY0, 267ff9313fac70fa85d051dd4d2b9d3402d39f67ceaPaul Berry dstX0, dstY0, dstX1, dstY1, mirror_x, mirror_y); 268ff9313fac70fa85d051dd4d2b9d3402d39f67ceaPaul Berry } 269fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry break; 270fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry case GL_DEPTH_BUFFER_BIT: 2715c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry src_irb = 2725c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer(read_fb->Attachment[BUFFER_DEPTH].Renderbuffer); 2735c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry dst_irb = 2745c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer(draw_fb->Attachment[BUFFER_DEPTH].Renderbuffer); 2755c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry if (!formats_match(buffer_bit, src_irb, dst_irb)) 276fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry return false; 2775c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry do_blorp_blit(intel, buffer_bit, src_irb, dst_irb, srcX0, srcY0, 278fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry dstX0, dstY0, dstX1, dstY1, mirror_x, mirror_y); 279fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry break; 280fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry case GL_STENCIL_BUFFER_BIT: 2815c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry src_irb = 2825c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer(read_fb->Attachment[BUFFER_STENCIL].Renderbuffer); 2835c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry dst_irb = 2845c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry intel_renderbuffer(draw_fb->Attachment[BUFFER_STENCIL].Renderbuffer); 2855c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry if (!formats_match(buffer_bit, src_irb, dst_irb)) 286fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry return false; 2875c66640ac7c271a96f66f4cb49adad54eb58cc47Paul Berry do_blorp_blit(intel, buffer_bit, src_irb, dst_irb, srcX0, srcY0, 288fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry dstX0, dstY0, dstX1, dstY1, mirror_x, mirror_y); 289fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry break; 290fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry default: 291fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry assert(false); 292fa1d267beb4adb542ea90b805306599f602c38d2Paul Berry } 293506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 294506d70be21cd3469118de89297cba0c0f709c1aePaul Berry return true; 295506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 296506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 297506d70be21cd3469118de89297cba0c0f709c1aePaul BerryGLbitfield 298506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_framebuffer(struct intel_context *intel, 299506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1, 300506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1, 301506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLbitfield mask, GLenum filter) 302506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 303b08545199ac8a01392a805f158d22cc03060a6fbPaul Berry /* BLORP is not supported before Gen6. */ 304b08545199ac8a01392a805f158d22cc03060a6fbPaul Berry if (intel->gen < 6) 305506d70be21cd3469118de89297cba0c0f709c1aePaul Berry return mask; 306506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 307506d70be21cd3469118de89297cba0c0f709c1aePaul Berry static GLbitfield buffer_bits[] = { 308506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GL_COLOR_BUFFER_BIT, 309506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GL_DEPTH_BUFFER_BIT, 310506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GL_STENCIL_BUFFER_BIT, 311506d70be21cd3469118de89297cba0c0f709c1aePaul Berry }; 312506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 313506d70be21cd3469118de89297cba0c0f709c1aePaul Berry for (unsigned int i = 0; i < ARRAY_SIZE(buffer_bits); ++i) { 314506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if ((mask & buffer_bits[i]) && 315506d70be21cd3469118de89297cba0c0f709c1aePaul Berry try_blorp_blit(intel, 316506d70be21cd3469118de89297cba0c0f709c1aePaul Berry srcX0, srcY0, srcX1, srcY1, 317506d70be21cd3469118de89297cba0c0f709c1aePaul Berry dstX0, dstY0, dstX1, dstY1, 318506d70be21cd3469118de89297cba0c0f709c1aePaul Berry filter, buffer_bits[i])) { 319506d70be21cd3469118de89297cba0c0f709c1aePaul Berry mask &= ~buffer_bits[i]; 320506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 321506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 322506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 323506d70be21cd3469118de89297cba0c0f709c1aePaul Berry return mask; 324506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 325506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 326665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry 327665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry/** 328665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry * Enum to specify the order of arguments in a sampler message 329665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry */ 330665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berryenum sampler_message_arg 331665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry{ 332665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_U_FLOAT, 333665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_V_FLOAT, 334665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_U_INT, 335665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_V_INT, 336233c207e9e477b6b0a5c6705e727129b92989073Paul Berry SAMPLER_MESSAGE_ARG_SI_INT, 3374ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_MCS_INT, 338233c207e9e477b6b0a5c6705e727129b92989073Paul Berry SAMPLER_MESSAGE_ARG_ZERO_INT, 339665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry}; 340665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry 341506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 342506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Generator for WM programs used in BLORP blits. 343506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 344506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * The bulk of the work done by the WM program is to wrap and unwrap the 345506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * coordinate transformations used by the hardware to store surfaces in 34619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * memory. The hardware transforms a pixel location (X, Y, S) (where S is the 34719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * sample index for a multisampled surface) to a memory offset by the 34819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * following formulas: 349506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 3508b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * offset = tile(tiling_format, encode_msaa(num_samples, layout, X, Y, S)) 3518b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * (X, Y, S) = decode_msaa(num_samples, layout, detile(tiling_format, offset)) 35219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 3531bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * For a single-sampled surface, or for a multisampled surface using 3541bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * INTEL_MSAA_LAYOUT_UMS, encode_msaa() and decode_msaa are the identity 3551bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * function: 35619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 3571bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * encode_msaa(1, NONE, X, Y, 0) = (X, Y, 0) 3581bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * decode_msaa(1, NONE, X, Y, 0) = (X, Y, 0) 3591bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * encode_msaa(n, UMS, X, Y, S) = (X, Y, S) 3601bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * decode_msaa(n, UMS, X, Y, S) = (X, Y, S) 36119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 3621bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * For a 4x multisampled surface using INTEL_MSAA_LAYOUT_IMS, encode_msaa() 3631bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * embeds the sample number into bit 1 of the X and Y coordinates: 36419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 3651bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * encode_msaa(4, IMS, X, Y, S) = (X', Y', 0) 36619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1) 36719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Y' = (Y & ~0b1 ) << 1 | (S & 0b10) | (Y & 0b1) 3681bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * decode_msaa(4, IMS, X, Y, 0) = (X', Y', S) 36919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * where X' = (X & ~0b11) >> 1 | (X & 0b1) 37019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Y' = (Y & ~0b11) >> 1 | (Y & 0b1) 37119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * S = (Y & 0b10) | (X & 0b10) >> 1 372506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 3737fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * For an 8x multisampled surface using INTEL_MSAA_LAYOUT_IMS, encode_msaa() 3747fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * embeds the sample number into bits 1 and 2 of the X coordinate and bit 1 of 3757fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * the Y coordinate: 3767fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * 3777fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * encode_msaa(8, IMS, X, Y, S) = (X', Y', 0) 3787fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1 | (X & 0b1) 3797fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1) 3807fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * decode_msaa(8, IMS, X, Y, 0) = (X', Y', S) 3817fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * where X' = (X & ~0b111) >> 2 | (X & 0b1) 3827fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * Y' = (Y & ~0b11) >> 1 | (Y & 0b1) 3837fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * S = (X & 0b100) | (Y & 0b10) | (X & 0b10) >> 1 3847fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * 385506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * For X tiling, tile() combines together the low-order bits of the X and Y 386506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * coordinates in the pattern 0byyyxxxxxxxxx, creating 4k tiles that are 512 387506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * bytes wide and 8 rows high: 388506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 3898b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile(x_tiled, X, Y, S) = A 390506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where A = tile_num << 12 | offset 3918b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile_num = (Y' >> 3) * tile_pitch + (X' >> 9) 3928b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * offset = (Y' & 0b111) << 9 393506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (X & 0b111111111) 394506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = X * cpp 3958b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = Y + S * qpitch 3968b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * detile(x_tiled, A) = (X, Y, S) 397506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where X = X' / cpp 3988b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y = Y' % qpitch 3998b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * S = Y' / qpitch 4008b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = (tile_num / tile_pitch) << 3 4018b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (A & 0b111000000000) >> 9 402506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = (tile_num % tile_pitch) << 9 403506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b111111111) 404506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 405506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (In all tiling formulas, cpp is the number of bytes occupied by a single 4068b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * sample ("chars per pixel"), tile_pitch is the number of 4k tiles required 4078b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * to fill the width of the surface, and qpitch is the spacing (in rows) 4088b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * between array slices). 409506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 410506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * For Y tiling, tile() combines together the low-order bits of the X and Y 411506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * coordinates in the pattern 0bxxxyyyyyxxxx, creating 4k tiles that are 128 412506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * bytes wide and 32 rows high: 413506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 4148b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile(y_tiled, X, Y, S) = A 415506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where A = tile_num << 12 | offset 4168b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile_num = (Y' >> 5) * tile_pitch + (X' >> 7) 417506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * offset = (X' & 0b1110000) << 5 418506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (Y' & 0b11111) << 4 419506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (X' & 0b1111) 420506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = X * cpp 4218b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = Y + S * qpitch 4228b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * detile(y_tiled, A) = (X, Y, S) 423506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where X = X' / cpp 4248b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y = Y' % qpitch 4258b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * S = Y' / qpitch 4268b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = (tile_num / tile_pitch) << 5 4278b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (A & 0b111110000) >> 4 428506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = (tile_num % tile_pitch) << 7 429506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b111000000000) >> 5 430506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b1111) 431506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 432506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * For W tiling, tile() combines together the low-order bits of the X and Y 433506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * coordinates in the pattern 0bxxxyyyyxyxyx, creating 4k tiles that are 64 434506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * bytes wide and 64 rows high (note that W tiling is only used for stencil 4358b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * buffers, which always have cpp = 1 and S=0): 436506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 4378b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile(w_tiled, X, Y, S) = A 438506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where A = tile_num << 12 | offset 4398b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile_num = (Y' >> 6) * tile_pitch + (X' >> 6) 440506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * offset = (X' & 0b111000) << 6 4418b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (Y' & 0b111100) << 3 442506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (X' & 0b100) << 2 4438b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (Y' & 0b10) << 2 444506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (X' & 0b10) << 1 4458b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (Y' & 0b1) << 1 446506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (X' & 0b1) 447506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = X * cpp = X 4488b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = Y + S * qpitch 4498b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * detile(w_tiled, A) = (X, Y, S) 450506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where X = X' / cpp = X' 4518b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y = Y' % qpitch = Y' 4528b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * S = Y / qpitch = 0 4538b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = (tile_num / tile_pitch) << 6 4548b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (A & 0b111100000) >> 3 4558b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (A & 0b1000) >> 2 4568b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * | (A & 0b10) >> 1 457506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = (tile_num % tile_pitch) << 6 458506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b111000000000) >> 6 459506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b10000) >> 2 460506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b100) >> 1 461506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * | (A & 0b1) 462506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 463506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Finally, for a non-tiled surface, tile() simply combines together the X and 464506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Y coordinates in the natural way: 465506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 4668b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * tile(untiled, X, Y, S) = A 467506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where A = Y * pitch + X' 468506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = X * cpp 4698b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = Y + S * qpitch 4708b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * detile(untiled, A) = (X, Y, S) 471506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * where X = X' / cpp 4728b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y = Y' % qpitch 4738b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * S = Y' / qpitch 474506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = A % pitch 4758b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * Y' = A / pitch 476506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 477506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (In these formulas, pitch is the number of bytes occupied by a single row 47819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * of samples). 479506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 480506d70be21cd3469118de89297cba0c0f709c1aePaul Berryclass brw_blorp_blit_program 481506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 482506d70be21cd3469118de89297cba0c0f709c1aePaul Berrypublic: 483506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_blorp_blit_program(struct brw_context *brw, 484506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const brw_blorp_blit_prog_key *key); 485506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ~brw_blorp_blit_program(); 486506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 487506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const GLuint *compile(struct brw_context *brw, GLuint *program_size); 488506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 489506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_blorp_prog_data prog_data; 490506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 491506d70be21cd3469118de89297cba0c0f709c1aePaul Berryprivate: 492506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void alloc_regs(); 493506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void alloc_push_const_regs(int base_reg); 494506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void compute_frag_coords(); 495506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void translate_tiling(bool old_tiled_w, bool new_tiled_w); 4961bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry void encode_msaa(unsigned num_samples, intel_msaa_layout layout); 4971bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry void decode_msaa(unsigned num_samples, intel_msaa_layout layout); 498506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void kill_if_outside_dst_rect(); 499506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void translate_dst_to_src(); 50019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry void single_to_blend(); 50117eae9762cdd6cfa69a060001e26113dfc0d7c86Paul Berry void manual_blend(unsigned num_samples); 5024725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry void sample(struct brw_reg dst); 5034725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry void texel_fetch(struct brw_reg dst); 5044ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry void mcs_fetch(); 505665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry void expand_to_32_bits(struct brw_reg src, struct brw_reg dst); 5064725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry void texture_lookup(struct brw_reg dst, GLuint msg_type, 5074725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry const sampler_message_arg *args, int num_args); 508506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void render_target_write(); 509506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 510b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /** 511b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * Base-2 logarithm of the maximum number of samples that can be blended. 512b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry */ 51317eae9762cdd6cfa69a060001e26113dfc0d7c86Paul Berry static const unsigned LOG2_MAX_BLEND_SAMPLES = 3; 514b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 515506d70be21cd3469118de89297cba0c0f709c1aePaul Berry void *mem_ctx; 516506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_context *brw; 517506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const brw_blorp_blit_prog_key *key; 518506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_compile func; 519506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 520506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Thread dispatch header */ 521506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg R0; 522506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 523506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Pixel X/Y coordinates (always in R1). */ 524506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg R1; 525506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 526506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Push constants */ 527506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg dst_x0; 528506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg dst_x1; 529506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg dst_y0; 530506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg dst_y1; 531506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct { 532506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg multiplier; 533506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg offset; 534506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } x_transform, y_transform; 535506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 536b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* Data read from texture (4 vec16's per array element) */ 537b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry struct brw_reg texture_data[LOG2_MAX_BLEND_SAMPLES + 1]; 538506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 5394ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry /* Auxiliary storage for the contents of the MCS surface. 5404ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry * 5414ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry * Since the sampler always returns 8 registers worth of data, this is 8 5424ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry * registers wide, even though we only use the first 2 registers of it. 5434ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry */ 5444ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry struct brw_reg mcs_data; 5454ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry 546506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* X coordinates. We have two of them so that we can perform coordinate 547506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * transformations easily. 548506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 549506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg x_coords[2]; 550506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 551506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Y coordinates. We have two of them so that we can perform coordinate 552506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * transformations easily. 553506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 554506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg y_coords[2]; 555506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 556506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Which element of x_coords and y_coords is currently in use. 557506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 558506d70be21cd3469118de89297cba0c0f709c1aePaul Berry int xy_coord_index; 559506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 56019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* True if, at the point in the program currently being compiled, the 56119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * sample index is known to be zero. 56219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 56319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry bool s_is_zero; 56419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 56519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* Register storing the sample index when s_is_zero is false. */ 56619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry struct brw_reg sample_index; 56719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 568506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Temporaries */ 569506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg t1; 570506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg t2; 571506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 572665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry /* MRF used for sampling and render target writes */ 573506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLuint base_mrf; 574506d70be21cd3469118de89297cba0c0f709c1aePaul Berry}; 575506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 576506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::brw_blorp_blit_program( 577506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_context *brw, 578506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const brw_blorp_blit_prog_key *key) 579506d70be21cd3469118de89297cba0c0f709c1aePaul Berry : mem_ctx(ralloc_context(NULL)), 580506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw(brw), 581506d70be21cd3469118de89297cba0c0f709c1aePaul Berry key(key) 582506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 583506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_init_compile(brw, &func, mem_ctx); 584506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 585506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 586506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::~brw_blorp_blit_program() 587506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 588506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ralloc_free(mem_ctx); 589506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 590506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 591506d70be21cd3469118de89297cba0c0f709c1aePaul Berryconst GLuint * 592506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::compile(struct brw_context *brw, 593506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLuint *program_size) 594506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 59519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* Sanity checks */ 59634a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry if (key->dst_tiled_w && key->rt_samples > 0) { 59734a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry /* If the destination image is W tiled and multisampled, then the thread 59834a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * must be dispatched once per sample, not once per pixel. This is 59934a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * necessary because after conversion between W and Y tiling, there's no 60019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * guarantee that all samples corresponding to a single pixel will still 60119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * be together. 60219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 60334a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry assert(key->persample_msaa_dispatch); 60419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 60519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 60619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry if (key->blend) { 6074725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry /* We are blending, which means we won't have an opportunity to 6084725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry * translate the tiling and sample count for the texture surface. So 60919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the surface state for the texture must be configured with the correct 61019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * tiling and sample count. 61119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 61219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry assert(!key->src_tiled_w); 61319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry assert(key->tex_samples == key->src_samples); 6141bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert(key->tex_layout == key->src_layout); 61519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry assert(key->tex_samples > 0); 61619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 61719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 61834a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry if (key->persample_msaa_dispatch) { 61934a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry /* It only makes sense to do persample dispatch if the render target is 62034a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * configured as multisampled. 62134a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry */ 62234a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry assert(key->rt_samples > 0); 62334a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry } 62434a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry 6251bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* Make sure layout is consistent with sample count */ 6261bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert((key->tex_layout == INTEL_MSAA_LAYOUT_NONE) == 6271bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry (key->tex_samples == 0)); 6281bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert((key->rt_layout == INTEL_MSAA_LAYOUT_NONE) == 6291bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry (key->rt_samples == 0)); 6301bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert((key->src_layout == INTEL_MSAA_LAYOUT_NONE) == 6311bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry (key->src_samples == 0)); 6321bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert((key->dst_layout == INTEL_MSAA_LAYOUT_NONE) == 6331bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry (key->dst_samples == 0)); 6348b1f467cce34340637e9baca4847fc5273cf7541Paul Berry 63534a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry /* Set up prog_data */ 63634a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry memset(&prog_data, 0, sizeof(prog_data)); 63734a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry prog_data.persample_msaa_dispatch = key->persample_msaa_dispatch; 63834a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry 639506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_set_compression_control(&func, BRW_COMPRESSION_NONE); 640506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 641506d70be21cd3469118de89297cba0c0f709c1aePaul Berry alloc_regs(); 642506d70be21cd3469118de89297cba0c0f709c1aePaul Berry compute_frag_coords(); 643506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 644506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Render target and texture hardware don't support W tiling. */ 645506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const bool rt_tiled_w = false; 646506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const bool tex_tiled_w = false; 647506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 648506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* The address that data will be written to is determined by the 64919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * coordinates supplied to the WM thread and the tiling and sample count of 65019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the render target, according to the formula: 651506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 65219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * (X, Y, S) = decode_msaa(rt_samples, detile(rt_tiling, offset)) 653506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 65419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * If the actual tiling and sample count of the destination surface are not 65519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the same as the configuration of the render target, then these 65619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * coordinates are wrong and we have to adjust them to compensate for the 65719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * difference. 658506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 65919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry if (rt_tiled_w != key->dst_tiled_w || 6608b1f467cce34340637e9baca4847fc5273cf7541Paul Berry key->rt_samples != key->dst_samples || 6611bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry key->rt_layout != key->dst_layout) { 6621bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry encode_msaa(key->rt_samples, key->rt_layout); 6638b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* Now (X, Y, S) = detile(rt_tiling, offset) */ 664506d70be21cd3469118de89297cba0c0f709c1aePaul Berry translate_tiling(rt_tiled_w, key->dst_tiled_w); 6658b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* Now (X, Y, S) = detile(dst_tiling, offset) */ 6661bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry decode_msaa(key->dst_samples, key->dst_layout); 66719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 668506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 66919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* Now (X, Y, S) = decode_msaa(dst_samples, detile(dst_tiling, offset)). 670506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 67119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * That is: X, Y and S now contain the true coordinates and sample index of 67219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the data that the WM thread should output. 673506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 674506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * If we need to kill pixels that are outside the destination rectangle, 675506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * now is the time to do it. 676506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 677506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 678506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (key->use_kill) 679506d70be21cd3469118de89297cba0c0f709c1aePaul Berry kill_if_outside_dst_rect(); 680506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 681506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Next, apply a translation to obtain coordinates in the source image. */ 682506d70be21cd3469118de89297cba0c0f709c1aePaul Berry translate_dst_to_src(); 683506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 68419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* If the source image is not multisampled, then we want to fetch sample 68519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * number 0, because that's the only sample there is. 686506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 68719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry if (key->src_samples == 0) 68819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry s_is_zero = true; 689506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 69019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* X, Y, and S are now the coordinates of the pixel in the source image 69119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * that we want to texture from. Exception: if we are blending, then S is 69219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * irrelevant, because we are going to fetch all samples. 693506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 69419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry if (key->blend) { 6954725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry if (brw->intel.gen == 6) { 6964725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry /* Gen6 hardware an automatically blend using the SAMPLE message */ 6974725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry single_to_blend(); 698b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry sample(texture_data[0]); 6994725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry } else { 7004725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry /* Gen7+ hardware doesn't automaticaly blend. */ 70117eae9762cdd6cfa69a060001e26113dfc0d7c86Paul Berry manual_blend(key->src_samples); 7024725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry } 70319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } else { 70419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* We aren't blending, which means we just want to fetch a single sample 70519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * from the source surface. The address that we want to fetch from is 70619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * related to the X, Y and S values according to the formula: 70719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 70819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * (X, Y, S) = decode_msaa(src_samples, detile(src_tiling, offset)). 70919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 71019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * If the actual tiling and sample count of the source surface are not 71119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the same as the configuration of the texture, then we need to adjust 71219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the coordinates to compensate for the difference. 71319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 71419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry if (tex_tiled_w != key->src_tiled_w || 7158b1f467cce34340637e9baca4847fc5273cf7541Paul Berry key->tex_samples != key->src_samples || 7161bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry key->tex_layout != key->src_layout) { 7171bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry encode_msaa(key->src_samples, key->src_layout); 7188b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* Now (X, Y, S) = detile(src_tiling, offset) */ 71919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry translate_tiling(key->src_tiled_w, tex_tiled_w); 7208b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* Now (X, Y, S) = detile(tex_tiling, offset) */ 7211bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry decode_msaa(key->tex_samples, key->tex_layout); 72219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 723506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 72419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* Now (X, Y, S) = decode_msaa(tex_samples, detile(tex_tiling, offset)). 72519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 72619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * In other words: X, Y, and S now contain values which, when passed to 72719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the texturing unit, will cause data to be read from the correct 72819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * memory location. So we can fetch the texel now. 72919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 7304ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS) 7314ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry mcs_fetch(); 732b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry texel_fetch(texture_data[0]); 73319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 73419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 73519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* Finally, write the fetched (or blended) value to the render target and 73619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * terminate the thread. 737506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 738506d70be21cd3469118de89297cba0c0f709c1aePaul Berry render_target_write(); 739506d70be21cd3469118de89297cba0c0f709c1aePaul Berry return brw_get_program(&func, program_size); 740506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 741506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 742506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 743506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::alloc_push_const_regs(int base_reg) 744506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 745506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define CONST_LOC(name) offsetof(brw_blorp_wm_push_constants, name) 746506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define ALLOC_REG(name) \ 747506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->name = \ 748506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, base_reg, CONST_LOC(name) / 2) 749506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 750506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(dst_x0); 751506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(dst_x1); 752506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(dst_y0); 753506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(dst_y1); 754506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(x_transform.multiplier); 755506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(x_transform.offset); 756506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(y_transform.multiplier); 757506d70be21cd3469118de89297cba0c0f709c1aePaul Berry ALLOC_REG(y_transform.offset); 758506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef CONST_LOC 759506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef ALLOC_REG 760506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 761506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 762506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 763506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::alloc_regs() 764506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 765506d70be21cd3469118de89297cba0c0f709c1aePaul Berry int reg = 0; 766506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->R0 = retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW); 767506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->R1 = retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW); 768506d70be21cd3469118de89297cba0c0f709c1aePaul Berry prog_data.first_curbe_grf = reg; 769506d70be21cd3469118de89297cba0c0f709c1aePaul Berry alloc_push_const_regs(reg); 770506d70be21cd3469118de89297cba0c0f709c1aePaul Berry reg += BRW_BLORP_NUM_PUSH_CONST_REGS; 771b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry for (unsigned i = 0; i < ARRAY_SIZE(texture_data); ++i) { 772e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry this->texture_data[i] = 773e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry retype(vec16(brw_vec8_grf(reg, 0)), key->texture_data_type); 774e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry reg += 8; 775b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry } 7764ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry this->mcs_data = 7774ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry retype(brw_vec8_grf(reg, 0), BRW_REGISTER_TYPE_UD); reg += 8; 778506d70be21cd3469118de89297cba0c0f709c1aePaul Berry for (int i = 0; i < 2; ++i) { 779506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->x_coords[i] 780506d70be21cd3469118de89297cba0c0f709c1aePaul Berry = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 781506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->y_coords[i] 782506d70be21cd3469118de89297cba0c0f709c1aePaul Berry = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 783506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 784506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->xy_coord_index = 0; 78519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry this->sample_index 78619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 787506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->t1 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 788506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->t2 = vec16(retype(brw_vec8_grf(reg++, 0), BRW_REGISTER_TYPE_UW)); 789506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 790b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* Make sure we didn't run out of registers */ 791b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry assert(reg <= GEN7_MRF_HACK_START); 792b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 793506d70be21cd3469118de89297cba0c0f709c1aePaul Berry int mrf = 2; 794506d70be21cd3469118de89297cba0c0f709c1aePaul Berry this->base_mrf = mrf; 795506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 796506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 797506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/* In the code that follows, X and Y can be used to quickly refer to the 798506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * active elements of x_coords and y_coords, and Xp and Yp ("X prime" and "Y 799506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * prime") to the inactive elements. 80019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 80119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * S can be used to quickly refer to sample_index. 802506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 803506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define X x_coords[xy_coord_index] 804506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define Y y_coords[xy_coord_index] 805506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define Xp x_coords[!xy_coord_index] 806506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define Yp y_coords[!xy_coord_index] 80719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry#define S sample_index 808506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 809506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/* Quickly swap the roles of (X, Y) and (Xp, Yp). Saves us from having to do 810506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * MOVs to transfor (Xp, Yp) to (X, Y) after a coordinate transformation. 811506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 812506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#define SWAP_XY_AND_XPYP() xy_coord_index = !xy_coord_index; 813506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 814506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 815506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Emit code to compute the X and Y coordinates of the pixels being rendered 816506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * by this WM invocation. 817506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 818506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Assuming the render target is set up for Y tiling, these (X, Y) values are 819506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * related to the address offset where outputs will be written by the formula: 820506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 821506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (X, Y, S) = decode_msaa(detile(offset)). 822506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 823506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (See brw_blorp_blit_program). 824506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 825506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 826506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::compute_frag_coords() 827506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 828506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* R1.2[15:0] = X coordinate of upper left pixel of subspan 0 (pixel 0) 829506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * R1.3[15:0] = X coordinate of upper left pixel of subspan 1 (pixel 4) 830506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * R1.4[15:0] = X coordinate of upper left pixel of subspan 2 (pixel 8) 831506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * R1.5[15:0] = X coordinate of upper left pixel of subspan 3 (pixel 12) 832506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 833506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Pixels within a subspan are laid out in this arrangement: 834506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 0 1 835506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 2 3 836506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 837506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * So, to compute the coordinates of each pixel, we need to read every 2nd 838506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 16-bit value (vstride=2) from R1, starting at the 4th 16-bit value 839506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (suboffset=4), and duplicate each value 4 times (hstride=0, width=4). 840506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * In other words, the data we want to access is R1.4<2;4,0>UW. 841506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 842506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Then, we need to add the repeating sequence (0, 1, 0, 1, ...) to the 843506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * result, since pixels n+1 and n+3 are in the right half of the subspan. 844506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 845506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_ADD(&func, X, stride(suboffset(R1, 4), 2, 4, 0), brw_imm_v(0x10101010)); 846506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 847506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Similarly, Y coordinates for subspans come from R1.2[31:16] through 848506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * R1.5[31:16], so to get pixel Y coordinates we need to start at the 5th 849506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 16-bit value instead of the 4th (R1.5<2;4,0>UW instead of 850506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * R1.4<2;4,0>UW). 851506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 852506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * And we need to add the repeating sequence (0, 0, 1, 1, ...), since 853506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * pixels n+2 and n+3 are in the bottom half of the subspan. 854506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 855506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_ADD(&func, Y, stride(suboffset(R1, 5), 2, 4, 0), brw_imm_v(0x11001100)); 85619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 85734a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry if (key->persample_msaa_dispatch) { 858619471dc322de80942f7dbb29a437890e48155c6Paul Berry switch (key->rt_samples) { 859619471dc322de80942f7dbb29a437890e48155c6Paul Berry case 4: 860619471dc322de80942f7dbb29a437890e48155c6Paul Berry /* The WM will be run in MSDISPMODE_PERSAMPLE with num_samples == 4. 861619471dc322de80942f7dbb29a437890e48155c6Paul Berry * Therefore, subspan 0 will represent sample 0, subspan 1 will 862619471dc322de80942f7dbb29a437890e48155c6Paul Berry * represent sample 1, and so on. 863619471dc322de80942f7dbb29a437890e48155c6Paul Berry * 864619471dc322de80942f7dbb29a437890e48155c6Paul Berry * So we need to populate S with the sequence (0, 0, 0, 0, 1, 1, 1, 865619471dc322de80942f7dbb29a437890e48155c6Paul Berry * 1, 2, 2, 2, 2, 3, 3, 3, 3). The easiest way to do this is to 866619471dc322de80942f7dbb29a437890e48155c6Paul Berry * populate a temporary variable with the sequence (0, 1, 2, 3), and 867619471dc322de80942f7dbb29a437890e48155c6Paul Berry * then copy from it using vstride=1, width=4, hstride=0. 868619471dc322de80942f7dbb29a437890e48155c6Paul Berry */ 869619471dc322de80942f7dbb29a437890e48155c6Paul Berry brw_MOV(&func, t1, brw_imm_v(0x3210)); 870619471dc322de80942f7dbb29a437890e48155c6Paul Berry brw_MOV(&func, S, stride(t1, 1, 4, 0)); 871619471dc322de80942f7dbb29a437890e48155c6Paul Berry break; 872619471dc322de80942f7dbb29a437890e48155c6Paul Berry case 8: { 873619471dc322de80942f7dbb29a437890e48155c6Paul Berry /* The WM will be run in MSDISPMODE_PERSAMPLE with num_samples == 8. 874619471dc322de80942f7dbb29a437890e48155c6Paul Berry * Therefore, subspan 0 will represent sample N (where N is 0 or 4), 875619471dc322de80942f7dbb29a437890e48155c6Paul Berry * subspan 1 will represent sample 1, and so on. We can find the 876619471dc322de80942f7dbb29a437890e48155c6Paul Berry * value of N by looking at R0.0 bits 7:6 ("Starting Sample Pair 877619471dc322de80942f7dbb29a437890e48155c6Paul Berry * Index") and multiplying by two (since samples are always delivered 878619471dc322de80942f7dbb29a437890e48155c6Paul Berry * in pairs). That is, we compute 2*((R0.0 & 0xc0) >> 6) == (R0.0 & 879619471dc322de80942f7dbb29a437890e48155c6Paul Berry * 0xc0) >> 5. 880619471dc322de80942f7dbb29a437890e48155c6Paul Berry * 881619471dc322de80942f7dbb29a437890e48155c6Paul Berry * Then we need to add N to the sequence (0, 0, 0, 0, 1, 1, 1, 1, 2, 882619471dc322de80942f7dbb29a437890e48155c6Paul Berry * 2, 2, 2, 3, 3, 3, 3), which we compute by populating a temporary 883619471dc322de80942f7dbb29a437890e48155c6Paul Berry * variable with the sequence (0, 1, 2, 3), and then reading from it 884619471dc322de80942f7dbb29a437890e48155c6Paul Berry * using vstride=1, width=4, hstride=0. 885619471dc322de80942f7dbb29a437890e48155c6Paul Berry */ 886619471dc322de80942f7dbb29a437890e48155c6Paul Berry struct brw_reg t1_ud1 = vec1(retype(t1, BRW_REGISTER_TYPE_UD)); 887619471dc322de80942f7dbb29a437890e48155c6Paul Berry struct brw_reg r0_ud1 = vec1(retype(R0, BRW_REGISTER_TYPE_UD)); 888619471dc322de80942f7dbb29a437890e48155c6Paul Berry brw_AND(&func, t1_ud1, r0_ud1, brw_imm_ud(0xc0)); 889619471dc322de80942f7dbb29a437890e48155c6Paul Berry brw_SHR(&func, t1_ud1, t1_ud1, brw_imm_ud(5)); 890619471dc322de80942f7dbb29a437890e48155c6Paul Berry brw_MOV(&func, t2, brw_imm_v(0x3210)); 891619471dc322de80942f7dbb29a437890e48155c6Paul Berry brw_ADD(&func, S, retype(t1_ud1, BRW_REGISTER_TYPE_UW), 892619471dc322de80942f7dbb29a437890e48155c6Paul Berry stride(t2, 1, 4, 0)); 893619471dc322de80942f7dbb29a437890e48155c6Paul Berry break; 894619471dc322de80942f7dbb29a437890e48155c6Paul Berry } 895619471dc322de80942f7dbb29a437890e48155c6Paul Berry default: 896619471dc322de80942f7dbb29a437890e48155c6Paul Berry assert(!"Unrecognized sample count in " 897619471dc322de80942f7dbb29a437890e48155c6Paul Berry "brw_blorp_blit_program::compute_frag_coords()"); 898619471dc322de80942f7dbb29a437890e48155c6Paul Berry break; 899619471dc322de80942f7dbb29a437890e48155c6Paul Berry } 90034a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry s_is_zero = false; 90134a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry } else { 90234a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry /* Either the destination surface is single-sampled, or the WM will be 90334a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * run in MSDISPMODE_PERPIXEL (which causes a single fragment dispatch 90434a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * per pixel). In either case, it's not meaningful to compute a sample 90534a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * value. Just set it to 0. 90634a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry */ 90734a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry s_is_zero = true; 90834a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry } 909506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 910506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 911506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 912506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Emit code to compensate for the difference between Y and W tiling. 913506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 914506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * This code modifies the X and Y coordinates according to the formula: 915506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 9168b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * (X', Y', S') = detile(new_tiling, tile(old_tiling, X, Y, S)) 917506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 918506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (See brw_blorp_blit_program). 919506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 920506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * It can only translate between W and Y tiling, so new_tiling and old_tiling 921506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * are booleans where true represents W tiling and false represents Y tiling. 922506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 923506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 924506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::translate_tiling(bool old_tiled_w, bool new_tiled_w) 925506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 926506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (old_tiled_w == new_tiled_w) 927506d70be21cd3469118de89297cba0c0f709c1aePaul Berry return; 928506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 9298b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* In the code that follows, we can safely assume that S = 0, because W 9301bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * tiling formats always use IMS layout. 9318b1f467cce34340637e9baca4847fc5273cf7541Paul Berry */ 9328b1f467cce34340637e9baca4847fc5273cf7541Paul Berry assert(s_is_zero); 9338b1f467cce34340637e9baca4847fc5273cf7541Paul Berry 934506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (new_tiled_w) { 935506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Given X and Y coordinates that describe an address using Y tiling, 936506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * translate to the X and Y coordinates that describe the same address 937506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * using W tiling. 938506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 939506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * If we break down the low order bits of X and Y, using a 940506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * single letter to represent each low-order bit: 941506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 942506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X = A << 7 | 0bBCDEFGH 943506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Y = J << 5 | 0bKLMNP (1) 944506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 945506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Then we can apply the Y tiling formula to see the memory offset being 946506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * addressed: 947506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 948506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * offset = (J * tile_pitch + A) << 12 | 0bBCDKLMNPEFGH (2) 949506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 950506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * If we apply the W detiling formula to this memory location, that the 951506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * corresponding X' and Y' coordinates are: 952506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 953506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = A << 6 | 0bBCDPFH (3) 954506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Y' = J << 6 | 0bKLMNEG 955506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 956506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Combining (1) and (3), we see that to transform (X, Y) to (X', Y'), 957506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * we need to make the following computation: 958506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 959506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = (X & ~0b1011) >> 1 | (Y & 0b1) << 2 | X & 0b1 (4) 960506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Y' = (Y & ~0b1) << 1 | (X & 0b1000) >> 2 | (X & 0b10) >> 1 961506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 962506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t1, X, brw_imm_uw(0xfff4)); /* X & ~0b1011 */ 963506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b1011) >> 1 */ 964506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 965506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHL(&func, t2, t2, brw_imm_uw(2)); /* (Y & 0b1) << 2 */ 966506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, t1, t1, t2); /* (X & ~0b1011) >> 1 | (Y & 0b1) << 2 */ 967506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 968506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, Xp, t1, t2); 969506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */ 970506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */ 971506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, X, brw_imm_uw(8)); /* X & 0b1000 */ 972506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHR(&func, t2, t2, brw_imm_uw(2)); /* (X & 0b1000) >> 2 */ 973506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (X & 0b1000) >> 2 */ 974506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */ 975506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */ 976506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, Yp, t1, t2); 977506d70be21cd3469118de89297cba0c0f709c1aePaul Berry SWAP_XY_AND_XPYP(); 978506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } else { 979506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Applying the same logic as above, but in reverse, we obtain the 980506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * formulas: 981506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * 982506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * X' = (X & ~0b101) << 1 | (Y & 0b10) << 2 | (Y & 0b1) << 1 | X & 0b1 983506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Y' = (Y & ~0b11) >> 1 | (X & 0b100) >> 2 984506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 985506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t1, X, brw_imm_uw(0xfffa)); /* X & ~0b101 */ 986506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b101) << 1 */ 987506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, Y, brw_imm_uw(2)); /* Y & 0b10 */ 988506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHL(&func, t2, t2, brw_imm_uw(2)); /* (Y & 0b10) << 2 */ 989506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, t1, t1, t2); /* (X & ~0b101) << 1 | (Y & 0b10) << 2 */ 990506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 991506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHL(&func, t2, t2, brw_imm_uw(1)); /* (Y & 0b1) << 1 */ 992506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, t1, t1, t2); /* (X & ~0b101) << 1 | (Y & 0b10) << 2 993506d70be21cd3469118de89297cba0c0f709c1aePaul Berry | (Y & 0b1) << 1 */ 994506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 995506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, Xp, t1, t2); 996506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */ 997506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */ 998506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, t2, X, brw_imm_uw(4)); /* X & 0b100 */ 999506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SHR(&func, t2, t2, brw_imm_uw(2)); /* (X & 0b100) >> 2 */ 1000506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_OR(&func, Yp, t1, t2); 1001506d70be21cd3469118de89297cba0c0f709c1aePaul Berry SWAP_XY_AND_XPYP(); 1002506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 1003506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1004506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1005506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 100619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Emit code to compensate for the difference between MSAA and non-MSAA 100719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * surfaces. 100819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 100919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * This code modifies the X and Y coordinates according to the formula: 101019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 10117fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * (X', Y', S') = encode_msaa(num_samples, IMS, X, Y, S) 101219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 101319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * (See brw_blorp_blit_program). 101419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 101519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berryvoid 10161bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berrybrw_blorp_blit_program::encode_msaa(unsigned num_samples, 10171bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry intel_msaa_layout layout) 101819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry{ 10191bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry switch (layout) { 10201bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_NONE: 10218b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* No translation necessary, and S should already be zero. */ 10228b1f467cce34340637e9baca4847fc5273cf7541Paul Berry assert(s_is_zero); 10231bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 10241bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_CMS: 10251bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* We can't compensate for compressed layout since at this point in the 10261bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * program we haven't read from the MCS buffer. 10271bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry */ 10281bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert(!"Bad layout in encode_msaa"); 10291bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 10301bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_UMS: 103119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* No translation necessary. */ 10321bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 10331bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_IMS: 10347fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry switch (num_samples) { 10357fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry case 4: 10367fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry /* encode_msaa(4, IMS, X, Y, S) = (X', Y', 0) 10377fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * where X' = (X & ~0b1) << 1 | (S & 0b1) << 1 | (X & 0b1) 10387fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1) 10397fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry */ 10407fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, X, brw_imm_uw(0xfffe)); /* X & ~0b1 */ 10417fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry if (!s_is_zero) { 10427fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, S, brw_imm_uw(1)); /* S & 0b1 */ 10437fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, t1, t1, t2); /* (X & ~0b1) | (S & 0b1) */ 10447fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry } 10457fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b1) << 1 10467fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry | (S & 0b1) << 1 */ 10477fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 10487fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Xp, t1, t2); 10497fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */ 10507fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */ 10517fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry if (!s_is_zero) { 10527fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, S, brw_imm_uw(2)); /* S & 0b10 */ 10537fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (S & 0b10) */ 10547fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry } 10557fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 10567fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Yp, t1, t2); 10577fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry break; 10587fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry case 8: 10597fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry /* encode_msaa(8, IMS, X, Y, S) = (X', Y', 0) 10607fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * where X' = (X & ~0b1) << 2 | (S & 0b100) | (S & 0b1) << 1 10617fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * | (X & 0b1) 10627fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * Y' = (Y & ~0b1) << 1 | (S & 0b10) | (Y & 0b1) 10637fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry */ 10647fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, X, brw_imm_uw(0xfffe)); /* X & ~0b1 */ 10657fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHL(&func, t1, t1, brw_imm_uw(2)); /* (X & ~0b1) << 2 */ 10667fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry if (!s_is_zero) { 10677fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, S, brw_imm_uw(4)); /* S & 0b100 */ 10687fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, t1, t1, t2); /* (X & ~0b1) << 2 | (S & 0b100) */ 10697fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, S, brw_imm_uw(1)); /* S & 0b1 */ 10707fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHL(&func, t2, t2, brw_imm_uw(1)); /* (S & 0b1) << 1 */ 10717fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, t1, t1, t2); /* (X & ~0b1) << 2 | (S & 0b100) 10727fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry | (S & 0b1) << 1 */ 10737fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry } 10747fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 10757fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Xp, t1, t2); 10767fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, Y, brw_imm_uw(0xfffe)); /* Y & ~0b1 */ 10777fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHL(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b1) << 1 */ 10787fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry if (!s_is_zero) { 10797fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, S, brw_imm_uw(2)); /* S & 0b10 */ 10807fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, t1, t1, t2); /* (Y & ~0b1) << 1 | (S & 0b10) */ 10817fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry } 10827fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 10837fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Yp, t1, t2); 10847fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry break; 108519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 108619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry SWAP_XY_AND_XPYP(); 10878b1f467cce34340637e9baca4847fc5273cf7541Paul Berry s_is_zero = true; 10881bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 108919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 109019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry} 109119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 109219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry/** 109319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Emit code to compensate for the difference between MSAA and non-MSAA 109419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * surfaces. 109519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 109619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * This code modifies the X and Y coordinates according to the formula: 109719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 10987fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * (X', Y', S) = decode_msaa(num_samples, IMS, X, Y, S) 109919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * 110019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * (See brw_blorp_blit_program). 110119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 110219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berryvoid 11031bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berrybrw_blorp_blit_program::decode_msaa(unsigned num_samples, 11041bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry intel_msaa_layout layout) 110519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry{ 11061bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry switch (layout) { 11071bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_NONE: 11088b1f467cce34340637e9baca4847fc5273cf7541Paul Berry /* No translation necessary, and S should already be zero. */ 11098b1f467cce34340637e9baca4847fc5273cf7541Paul Berry assert(s_is_zero); 11101bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 11111bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_CMS: 11121bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* We can't compensate for compressed layout since at this point in the 11131bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * program we don't have access to the MCS buffer. 11141bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry */ 11151bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert(!"Bad layout in encode_msaa"); 11161bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 11171bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_UMS: 111819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* No translation necessary. */ 11191bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 11201bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry case INTEL_MSAA_LAYOUT_IMS: 11218b1f467cce34340637e9baca4847fc5273cf7541Paul Berry assert(s_is_zero); 11227fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry switch (num_samples) { 11237fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry case 4: 11247fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry /* decode_msaa(4, IMS, X, Y, 0) = (X', Y', S) 11257fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * where X' = (X & ~0b11) >> 1 | (X & 0b1) 11267fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * Y' = (Y & ~0b11) >> 1 | (Y & 0b1) 11277fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * S = (Y & 0b10) | (X & 0b10) >> 1 11287fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry */ 11297fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, X, brw_imm_uw(0xfffc)); /* X & ~0b11 */ 11307fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (X & ~0b11) >> 1 */ 11317fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 11327fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Xp, t1, t2); 11337fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */ 11347fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */ 11357fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 11367fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Yp, t1, t2); 11377fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, Y, brw_imm_uw(2)); /* Y & 0b10 */ 11387fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */ 11397fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */ 11407fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, S, t1, t2); 11417fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry break; 11427fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry case 8: 11437fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry /* decode_msaa(8, IMS, X, Y, 0) = (X', Y', S) 11447fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * where X' = (X & ~0b111) >> 2 | (X & 0b1) 11457fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * Y' = (Y & ~0b11) >> 1 | (Y & 0b1) 11467fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry * S = (X & 0b100) | (Y & 0b10) | (X & 0b10) >> 1 11477fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry */ 11487fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, X, brw_imm_uw(0xfff8)); /* X & ~0b111 */ 11497fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHR(&func, t1, t1, brw_imm_uw(2)); /* (X & ~0b111) >> 2 */ 11507fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, X, brw_imm_uw(1)); /* X & 0b1 */ 11517fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Xp, t1, t2); 11527fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, Y, brw_imm_uw(0xfffc)); /* Y & ~0b11 */ 11537fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHR(&func, t1, t1, brw_imm_uw(1)); /* (Y & ~0b11) >> 1 */ 11547fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, Y, brw_imm_uw(1)); /* Y & 0b1 */ 11557fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, Yp, t1, t2); 11567fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t1, X, brw_imm_uw(4)); /* X & 0b100 */ 11577fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, Y, brw_imm_uw(2)); /* Y & 0b10 */ 11587fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, t1, t1, t2); /* (X & 0b100) | (Y & 0b10) */ 11597fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_AND(&func, t2, X, brw_imm_uw(2)); /* X & 0b10 */ 11607fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_SHR(&func, t2, t2, brw_imm_uw(1)); /* (X & 0b10) >> 1 */ 11617fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry brw_OR(&func, S, t1, t2); 11627fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry break; 11637fae97c98bfa13bff0e9da857e86eefdb625584cPaul Berry } 116419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry s_is_zero = false; 116519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry SWAP_XY_AND_XPYP(); 11661bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry break; 116719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 116819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry} 116919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 117019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry/** 1171506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Emit code that kills pixels whose X and Y coordinates are outside the 1172506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * boundary of the rectangle defined by the push constants (dst_x0, dst_y0, 1173506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * dst_x1, dst_y1). 1174506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 1175506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 1176506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::kill_if_outside_dst_rect() 1177506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1178506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg f0 = brw_flag_reg(); 1179506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg g1 = retype(brw_vec1_grf(1, 7), BRW_REGISTER_TYPE_UW); 1180506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct brw_reg null16 = vec16(retype(brw_null_reg(), BRW_REGISTER_TYPE_UW)); 1181506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1182506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_CMP(&func, null16, BRW_CONDITIONAL_GE, X, dst_x0); 1183506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_CMP(&func, null16, BRW_CONDITIONAL_GE, Y, dst_y0); 1184506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_CMP(&func, null16, BRW_CONDITIONAL_L, X, dst_x1); 1185506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_CMP(&func, null16, BRW_CONDITIONAL_L, Y, dst_y1); 1186506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1187506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_set_predicate_control(&func, BRW_PREDICATE_NONE); 1188506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_push_insn_state(&func); 1189506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_set_mask_control(&func, BRW_MASK_DISABLE); 1190506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_AND(&func, g1, f0, g1); 1191506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_pop_insn_state(&func); 1192506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1193506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1194506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 1195506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Emit code to translate from destination (X, Y) coordinates to source (X, Y) 1196506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * coordinates. 1197506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 1198506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 1199506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::translate_dst_to_src() 1200506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1201506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_MUL(&func, Xp, X, x_transform.multiplier); 1202506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_MUL(&func, Yp, Y, y_transform.multiplier); 1203506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_ADD(&func, Xp, Xp, x_transform.offset); 1204506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_ADD(&func, Yp, Yp, y_transform.offset); 1205506d70be21cd3469118de89297cba0c0f709c1aePaul Berry SWAP_XY_AND_XPYP(); 1206506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1207506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1208506d70be21cd3469118de89297cba0c0f709c1aePaul Berry/** 120919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Emit code to transform the X and Y coordinates as needed for blending 121019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * together the different samples in an MSAA texture. 121119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 121219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berryvoid 121319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berrybrw_blorp_blit_program::single_to_blend() 121419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry{ 121519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* When looking up samples in an MSAA texture using the SAMPLE message, 121619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Gen6 requires the texture coordinates to be odd integers (so that they 121719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * correspond to the center of a 2x2 block representing the four samples 121819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * that maxe up a pixel). So we need to multiply our X and Y coordinates 121919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * each by 2 and then add 1. 122019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 122119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry brw_SHL(&func, t1, X, brw_imm_w(1)); 122219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry brw_SHL(&func, t2, Y, brw_imm_w(1)); 122319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry brw_ADD(&func, Xp, t1, brw_imm_w(1)); 122419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry brw_ADD(&func, Yp, t2, brw_imm_w(1)); 122519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry SWAP_XY_AND_XPYP(); 122619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry} 122719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 1228b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 1229b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry/** 1230b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * Count the number of trailing 1 bits in the given value. For example: 1231b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * 1232b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * count_trailing_one_bits(0) == 0 1233b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * count_trailing_one_bits(7) == 3 1234b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * count_trailing_one_bits(11) == 2 1235b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry */ 1236b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berryinline int count_trailing_one_bits(unsigned value) 1237b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry{ 1238b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry#if defined(__GNUC__) && ((__GNUC__ * 100 + __GNUC_MINOR__) >= 304) /* gcc 3.4 or later */ 1239b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry return __builtin_ctz(~value); 1240b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry#else 1241b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry return _mesa_bitcount(value & ~(value + 1)); 1242b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry#endif 1243b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry} 1244b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 1245b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 12464725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berryvoid 124717eae9762cdd6cfa69a060001e26113dfc0d7c86Paul Berrybrw_blorp_blit_program::manual_blend(unsigned num_samples) 12484725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry{ 12494ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS) 12504ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry mcs_fetch(); 12514ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry 1252b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* We add together samples using a binary tree structure, e.g. for 4x MSAA: 1253b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * 1254b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * result = ((sample[0] + sample[1]) + (sample[2] + sample[3])) / 4 1255b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * 1256b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * This ensures that when all samples have the same value, no numerical 1257b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * precision is lost, since each addition operation always adds two equal 1258b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * values, and summing two equal floating point values does not lose 1259b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * precision. 1260b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * 1261b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * We perform this computation by treating the texture_data array as a 1262b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * stack and performing the following operations: 1263b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * 1264b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - push sample 0 onto stack 1265b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - push sample 1 onto stack 1266b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - add top two stack entries 1267b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - push sample 2 onto stack 1268b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - push sample 3 onto stack 1269b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - add top two stack entries 1270b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - add top two stack entries 1271b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * - divide top stack entry by 4 1272b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * 1273b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * Note that after pushing sample i onto the stack, the number of add 1274b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * operations we do is equal to the number of trailing 1 bits in i. This 1275b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * works provided the total number of samples is a power of two, which it 1276b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry * always is for i965. 1277e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry * 1278e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry * For integer formats, we replace the add operations with average 1279e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry * operations and skip the final division. 1280b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry */ 1281e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry typedef struct brw_instruction *(*brw_op2_ptr)(struct brw_compile *, 1282e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry struct brw_reg, 1283e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry struct brw_reg, 1284e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry struct brw_reg); 1285e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry brw_op2_ptr combine_op = 1286e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry key->texture_data_type == BRW_REGISTER_TYPE_F ? brw_ADD : brw_AVG; 1287b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry unsigned stack_depth = 0; 128817eae9762cdd6cfa69a060001e26113dfc0d7c86Paul Berry for (unsigned i = 0; i < num_samples; ++i) { 1289b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry assert(stack_depth == _mesa_bitcount(i)); /* Loop invariant */ 1290b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 1291b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* Push sample i onto the stack */ 1292b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry assert(stack_depth < ARRAY_SIZE(texture_data)); 1293b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry if (i == 0) { 1294b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry s_is_zero = true; 1295b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry } else { 1296b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry s_is_zero = false; 1297b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry brw_MOV(&func, S, brw_imm_uw(i)); 1298b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry } 1299b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry texel_fetch(texture_data[stack_depth++]); 1300b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 1301f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry if (i == 0 && key->tex_layout == INTEL_MSAA_LAYOUT_CMS) { 1302f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry /* The Ivy Bridge PRM, Vol4 Part1 p27 (Multisample Control Surface) 1303f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * suggests an optimization: 1304f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * 1305f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * "A simple optimization with probable large return in 1306f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * performance is to compare the MCS value to zero (indicating 1307f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * all samples are on sample slice 0), and sample only from 1308f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * sample slice 0 using ld2dss if MCS is zero." 1309f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * 1310f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * Note that in the case where the MCS value is zero, sampling from 1311f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * sample slice 0 using ld2dss and sampling from sample 0 using 1312f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * ld2dms are equivalent (since all samples are on sample slice 0). 1313f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * Since we have already sampled from sample 0, all we need to do is 1314f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry * skip the remaining fetches and averaging if MCS is zero. 1315f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry */ 1316f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry brw_CMP(&func, vec16(brw_null_reg()), BRW_CONDITIONAL_NZ, 1317f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry mcs_data, brw_imm_ud(0)); 1318f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry brw_IF(&func, BRW_EXECUTE_16); 1319f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry } 1320f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry 1321b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* Do count_trailing_one_bits(i) times */ 1322b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry for (int j = count_trailing_one_bits(i); j-- > 0; ) { 1323b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry assert(stack_depth >= 2); 1324b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry --stack_depth; 1325b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 1326b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* TODO: should use a smaller loop bound for non_RGBA formats */ 1327b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry for (int k = 0; k < 4; ++k) { 1328e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry combine_op(&func, offset(texture_data[stack_depth - 1], 2*k), 1329e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry offset(vec8(texture_data[stack_depth - 1]), 2*k), 1330e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry offset(vec8(texture_data[stack_depth]), 2*k)); 1331b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry } 13324725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry } 13334725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry } 13344725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry 1335b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry /* We should have just 1 sample on the stack now. */ 1336b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry assert(stack_depth == 1); 1337b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry 1338e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry if (key->texture_data_type == BRW_REGISTER_TYPE_F) { 1339e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry /* Scale the result down by a factor of num_samples */ 1340e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry /* TODO: should use a smaller loop bound for non-RGBA formats */ 1341e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry for (int j = 0; j < 4; ++j) { 1342e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry brw_MUL(&func, offset(texture_data[0], 2*j), 1343e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry offset(vec8(texture_data[0]), 2*j), 1344e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry brw_imm_f(1.0/num_samples)); 1345e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry } 13464725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry } 1347f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry 1348f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry if (key->tex_layout == INTEL_MSAA_LAYOUT_CMS) 1349f91b4d92b97664e6354f66138705e93bec363ba0Paul Berry brw_ENDIF(&func); 13504725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry} 13514725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry 135219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry/** 135319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * Emit code to look up a value in the texture using the SAMPLE message (which 135419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * does blending of MSAA surfaces). 135519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 135619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berryvoid 13574725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berrybrw_blorp_blit_program::sample(struct brw_reg dst) 135819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry{ 1359665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry static const sampler_message_arg args[2] = { 1360665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_U_FLOAT, 1361665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_V_FLOAT 1362665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry }; 1363665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry 13644725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE, args, ARRAY_SIZE(args)); 136519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry} 136619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 136719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry/** 1368506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Emit code to look up a value in the texture using the SAMPLE_LD message 1369506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * (which does a simple texel fetch). 1370506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 1371506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 13724725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berrybrw_blorp_blit_program::texel_fetch(struct brw_reg dst) 1373506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 13741c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry static const sampler_message_arg gen6_args[5] = { 1375665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry SAMPLER_MESSAGE_ARG_U_INT, 1376233c207e9e477b6b0a5c6705e727129b92989073Paul Berry SAMPLER_MESSAGE_ARG_V_INT, 1377233c207e9e477b6b0a5c6705e727129b92989073Paul Berry SAMPLER_MESSAGE_ARG_ZERO_INT, /* R */ 1378233c207e9e477b6b0a5c6705e727129b92989073Paul Berry SAMPLER_MESSAGE_ARG_ZERO_INT, /* LOD */ 1379233c207e9e477b6b0a5c6705e727129b92989073Paul Berry SAMPLER_MESSAGE_ARG_SI_INT 1380665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry }; 13811c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry static const sampler_message_arg gen7_ld_args[3] = { 13821c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry SAMPLER_MESSAGE_ARG_U_INT, 13831c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry SAMPLER_MESSAGE_ARG_ZERO_INT, /* LOD */ 13841c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry SAMPLER_MESSAGE_ARG_V_INT 13851c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry }; 13861c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry static const sampler_message_arg gen7_ld2dss_args[3] = { 13871c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry SAMPLER_MESSAGE_ARG_SI_INT, 13881c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry SAMPLER_MESSAGE_ARG_U_INT, 13891c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry SAMPLER_MESSAGE_ARG_V_INT 13901c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry }; 13914ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry static const sampler_message_arg gen7_ld2dms_args[4] = { 13924ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_SI_INT, 13934ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_MCS_INT, 13944ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_U_INT, 13954ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_V_INT 13964ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry }; 1397665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry 13981c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry switch (brw->intel.gen) { 13991c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry case 6: 14004725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE_LD, gen6_args, 14011c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry s_is_zero ? 2 : 5); 14021c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry break; 14031c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry case 7: 140460c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry switch (key->tex_layout) { 140560c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry case INTEL_MSAA_LAYOUT_IMS: 140660c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry /* From the Ivy Bridge PRM, Vol4 Part1 p72 (Multisampled Surface Storage 140760c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * Format): 140860c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * 140960c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * If this field is MSFMT_DEPTH_STENCIL 141060c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * [a.k.a. INTEL_MSAA_LAYOUT_IMS], the only sampling engine 141160c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * messages allowed are "ld2dms", "resinfo", and "sampleinfo". 141260c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * 141360c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * So fall through to emit the same message as we use for 141460c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * INTEL_MSAA_LAYOUT_CMS. 141560c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry */ 141660c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry case INTEL_MSAA_LAYOUT_CMS: 141760c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry texture_lookup(dst, GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS, 141860c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry gen7_ld2dms_args, ARRAY_SIZE(gen7_ld2dms_args)); 141960c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry break; 142060c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry case INTEL_MSAA_LAYOUT_UMS: 142160c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry texture_lookup(dst, GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DSS, 142260c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry gen7_ld2dss_args, ARRAY_SIZE(gen7_ld2dss_args)); 142360c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry break; 142460c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry case INTEL_MSAA_LAYOUT_NONE: 14251c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry assert(s_is_zero); 14264725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry texture_lookup(dst, GEN5_SAMPLER_MESSAGE_SAMPLE_LD, gen7_ld_args, 14271c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry ARRAY_SIZE(gen7_ld_args)); 142860c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry break; 14291c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry } 14301c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry break; 14311c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry default: 14321c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry assert(!"Should not get here."); 14331c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry break; 14341c73c705fadf164d61003415e3380f2d06f2e7b3Paul Berry }; 1435506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1436506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1437506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 14384ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berrybrw_blorp_blit_program::mcs_fetch() 14394ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry{ 14404ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry static const sampler_message_arg gen7_ld_mcs_args[2] = { 14414ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_U_INT, 14424ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry SAMPLER_MESSAGE_ARG_V_INT 14434ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry }; 14444ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry texture_lookup(vec16(mcs_data), GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS, 14454ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry gen7_ld_mcs_args, ARRAY_SIZE(gen7_ld_mcs_args)); 14464ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry} 14474ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry 14484ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berryvoid 1449665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berrybrw_blorp_blit_program::expand_to_32_bits(struct brw_reg src, 1450665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry struct brw_reg dst) 1451506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1452665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry brw_MOV(&func, vec8(dst), vec8(src)); 1453506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_set_compression_control(&func, BRW_COMPRESSION_2NDHALF); 1454665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry brw_MOV(&func, offset(vec8(dst), 1), suboffset(vec8(src), 8)); 1455506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_set_compression_control(&func, BRW_COMPRESSION_NONE); 1456665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry} 1457665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry 1458665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berryvoid 14594725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berrybrw_blorp_blit_program::texture_lookup(struct brw_reg dst, 14604725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry GLuint msg_type, 1461665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry const sampler_message_arg *args, 1462665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry int num_args) 1463665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry{ 1464665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry struct brw_reg mrf = 1465665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry retype(vec16(brw_message_reg(base_mrf)), BRW_REGISTER_TYPE_UD); 1466665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry for (int arg = 0; arg < num_args; ++arg) { 1467665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry switch (args[arg]) { 1468665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry case SAMPLER_MESSAGE_ARG_U_FLOAT: 1469665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry expand_to_32_bits(X, retype(mrf, BRW_REGISTER_TYPE_F)); 1470665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry break; 1471665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry case SAMPLER_MESSAGE_ARG_V_FLOAT: 1472665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry expand_to_32_bits(Y, retype(mrf, BRW_REGISTER_TYPE_F)); 1473665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry break; 1474665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry case SAMPLER_MESSAGE_ARG_U_INT: 1475665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry expand_to_32_bits(X, mrf); 1476665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry break; 1477665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry case SAMPLER_MESSAGE_ARG_V_INT: 1478665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry expand_to_32_bits(Y, mrf); 1479665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry break; 1480233c207e9e477b6b0a5c6705e727129b92989073Paul Berry case SAMPLER_MESSAGE_ARG_SI_INT: 1481233c207e9e477b6b0a5c6705e727129b92989073Paul Berry /* Note: on Gen7, this code may be reached with s_is_zero==true 1482233c207e9e477b6b0a5c6705e727129b92989073Paul Berry * because in Gen7's ld2dss message, the sample index is the first 1483233c207e9e477b6b0a5c6705e727129b92989073Paul Berry * argument. When this happens, we need to move a 0 into the 1484233c207e9e477b6b0a5c6705e727129b92989073Paul Berry * appropriate message register. 1485233c207e9e477b6b0a5c6705e727129b92989073Paul Berry */ 1486233c207e9e477b6b0a5c6705e727129b92989073Paul Berry if (s_is_zero) 1487233c207e9e477b6b0a5c6705e727129b92989073Paul Berry brw_MOV(&func, mrf, brw_imm_ud(0)); 1488233c207e9e477b6b0a5c6705e727129b92989073Paul Berry else 1489233c207e9e477b6b0a5c6705e727129b92989073Paul Berry expand_to_32_bits(S, mrf); 1490233c207e9e477b6b0a5c6705e727129b92989073Paul Berry break; 14914ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry case SAMPLER_MESSAGE_ARG_MCS_INT: 149260c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry switch (key->tex_layout) { 149360c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry case INTEL_MSAA_LAYOUT_CMS: 149460c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry brw_MOV(&func, mrf, mcs_data); 149560c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry break; 149660c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry case INTEL_MSAA_LAYOUT_IMS: 149760c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry /* When sampling from an IMS surface, MCS data is not relevant, 149860c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * and the hardware ignores it. So don't bother populating it. 149960c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry */ 150060c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry break; 150160c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry default: 150260c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry /* We shouldn't be trying to send MCS data with any other 150360c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * layouts. 150460c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry */ 150560c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry assert (!"Unsupported layout for MCS data"); 150660c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry break; 150760c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry } 15084ebbc766210190cb1f03fa4fc762bf7ecc0c7f90Paul Berry break; 1509233c207e9e477b6b0a5c6705e727129b92989073Paul Berry case SAMPLER_MESSAGE_ARG_ZERO_INT: 1510233c207e9e477b6b0a5c6705e727129b92989073Paul Berry brw_MOV(&func, mrf, brw_imm_ud(0)); 1511233c207e9e477b6b0a5c6705e727129b92989073Paul Berry break; 1512665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry } 1513665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry mrf.nr += 2; 1514665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry } 1515506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1516506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_SAMPLE(&func, 15174725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry retype(dst, BRW_REGISTER_TYPE_UW) /* dest */, 1518506d70be21cd3469118de89297cba0c0f709c1aePaul Berry base_mrf /* msg_reg_nr */, 1519665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry brw_message_reg(base_mrf) /* src0 */, 1520506d70be21cd3469118de89297cba0c0f709c1aePaul Berry BRW_BLORP_TEXTURE_BINDING_TABLE_INDEX, 1521665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry 0 /* sampler */, 1522506d70be21cd3469118de89297cba0c0f709c1aePaul Berry WRITEMASK_XYZW, 1523506d70be21cd3469118de89297cba0c0f709c1aePaul Berry msg_type, 1524506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 8 /* response_length. TODO: should be smaller for non-RGBA formats? */, 1525665dc82bdc0e83854dd0f700ec264021bfb5cb39Paul Berry mrf.nr - base_mrf /* msg_length */, 1526506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 0 /* header_present */, 1527506d70be21cd3469118de89297cba0c0f709c1aePaul Berry BRW_SAMPLER_SIMD_MODE_SIMD16, 1528506d70be21cd3469118de89297cba0c0f709c1aePaul Berry BRW_SAMPLER_RETURN_FORMAT_FLOAT32); 1529506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1530506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1531506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef X 1532506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef Y 1533506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef U 1534506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef V 1535506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef S 1536506d70be21cd3469118de89297cba0c0f709c1aePaul Berry#undef SWAP_XY_AND_XPYP 1537506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1538506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 1539506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_program::render_target_write() 1540506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1541e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry struct brw_reg mrf_rt_write = 1542e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry retype(vec16(brw_message_reg(base_mrf)), key->texture_data_type); 1543506d70be21cd3469118de89297cba0c0f709c1aePaul Berry int mrf_offset = 0; 1544506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1545506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* If we may have killed pixels, then we need to send R0 and R1 in a header 1546506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * so that the render target knows which pixels we killed. 1547506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 1548506d70be21cd3469118de89297cba0c0f709c1aePaul Berry bool use_header = key->use_kill; 1549506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (use_header) { 1550506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Copy R0/1 to MRF */ 1551506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_MOV(&func, retype(mrf_rt_write, BRW_REGISTER_TYPE_UD), 1552506d70be21cd3469118de89297cba0c0f709c1aePaul Berry retype(R0, BRW_REGISTER_TYPE_UD)); 1553506d70be21cd3469118de89297cba0c0f709c1aePaul Berry mrf_offset += 2; 1554506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 1555506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1556506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Copy texture data to MRFs */ 1557506d70be21cd3469118de89297cba0c0f709c1aePaul Berry for (int i = 0; i < 4; ++i) { 1558506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* E.g. mov(16) m2.0<1>:f r2.0<8;8,1>:f { Align1, H1 } */ 15594725ba03cae87ddbf1fa10feaca3d42f24115f91Paul Berry brw_MOV(&func, offset(mrf_rt_write, mrf_offset), 1560b961d37e613b8b14927c42e09d16d09d70ebcb77Paul Berry offset(vec8(texture_data[0]), 2*i)); 1561506d70be21cd3469118de89297cba0c0f709c1aePaul Berry mrf_offset += 2; 1562506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 1563506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1564506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* Now write to the render target and terminate the thread */ 1565506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_fb_WRITE(&func, 1566506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 16 /* dispatch_width */, 1567506d70be21cd3469118de89297cba0c0f709c1aePaul Berry base_mrf /* msg_reg_nr */, 1568506d70be21cd3469118de89297cba0c0f709c1aePaul Berry mrf_rt_write /* src0 */, 156929362875f2613ad87abe7725ce3c56c36d16cf9bEric Anholt BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE, 1570506d70be21cd3469118de89297cba0c0f709c1aePaul Berry BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX, 1571506d70be21cd3469118de89297cba0c0f709c1aePaul Berry mrf_offset /* msg_length. TODO: Should be smaller for non-RGBA formats. */, 1572506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 0 /* response_length */, 1573506d70be21cd3469118de89297cba0c0f709c1aePaul Berry true /* eot */, 1574506d70be21cd3469118de89297cba0c0f709c1aePaul Berry use_header); 1575506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1576506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1577506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1578506d70be21cd3469118de89297cba0c0f709c1aePaul Berryvoid 1579506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_coord_transform_params::setup(GLuint src0, GLuint dst0, GLuint dst1, 1580506d70be21cd3469118de89297cba0c0f709c1aePaul Berry bool mirror) 1581506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1582506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (!mirror) { 1583506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* When not mirroring a coordinate (say, X), we need: 1584506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * x' - src_x0 = x - dst_x0 1585506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Therefore: 1586506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * x' = 1*x + (src_x0 - dst_x0) 1587506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 1588506d70be21cd3469118de89297cba0c0f709c1aePaul Berry multiplier = 1; 1589506d70be21cd3469118de89297cba0c0f709c1aePaul Berry offset = src0 - dst0; 1590506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } else { 1591506d70be21cd3469118de89297cba0c0f709c1aePaul Berry /* When mirroring X we need: 1592506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * x' - src_x0 = dst_x1 - x - 1 1593506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * Therefore: 1594506d70be21cd3469118de89297cba0c0f709c1aePaul Berry * x' = -1*x + (src_x0 + dst_x1 - 1) 1595506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 1596506d70be21cd3469118de89297cba0c0f709c1aePaul Berry multiplier = -1; 1597506d70be21cd3469118de89297cba0c0f709c1aePaul Berry offset = src0 + dst1 - 1; 1598506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 1599506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1600506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1601506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 16021bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry/** 16031bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * Determine which MSAA layout the GPU pipeline should be configured for, 16041bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * based on the chip generation, the number of samples, and the true layout of 16051bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * the image in memory. 16061bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry */ 16071bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berryinline intel_msaa_layout 16081bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berrycompute_msaa_layout_for_pipeline(struct brw_context *brw, unsigned num_samples, 16091bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry intel_msaa_layout true_layout) 16101bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry{ 161197fc89c6cbaa3b5ef7f678d2dc2c7d5bbba05315Paul Berry if (num_samples <= 1) { 16121bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* When configuring the GPU for non-MSAA, we can still accommodate IMS 16131bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * format buffers, by transforming coordinates appropriately. 16141bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry */ 16151bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert(true_layout == INTEL_MSAA_LAYOUT_NONE || 16161bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry true_layout == INTEL_MSAA_LAYOUT_IMS); 16171bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry return INTEL_MSAA_LAYOUT_NONE; 16180dd5e98aa5c146ef21ab44b34fb7714206d5ec08Paul Berry } else { 16190dd5e98aa5c146ef21ab44b34fb7714206d5ec08Paul Berry assert(true_layout != INTEL_MSAA_LAYOUT_NONE); 16201bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry } 16211bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry 16221bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* Prior to Gen7, all MSAA surfaces use IMS layout. */ 16231bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry if (brw->intel.gen == 6) { 16241bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert(true_layout == INTEL_MSAA_LAYOUT_IMS); 16251bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry } 16261bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry 16271bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry return true_layout; 16281bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry} 16291bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry 16301bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry 16318b1f467cce34340637e9baca4847fc5273cf7541Paul Berrybrw_blorp_blit_params::brw_blorp_blit_params(struct brw_context *brw, 16328b1f467cce34340637e9baca4847fc5273cf7541Paul Berry struct intel_mipmap_tree *src_mt, 1633e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry unsigned src_level, unsigned src_layer, 1634506d70be21cd3469118de89297cba0c0f709c1aePaul Berry struct intel_mipmap_tree *dst_mt, 1635e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry unsigned dst_level, unsigned dst_layer, 1636506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLuint src_x0, GLuint src_y0, 1637506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLuint dst_x0, GLuint dst_y0, 1638506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLuint dst_x1, GLuint dst_y1, 1639506d70be21cd3469118de89297cba0c0f709c1aePaul Berry bool mirror_x, bool mirror_y) 1640506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1641e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry src.set(brw, src_mt, src_level, src_layer); 1642e87174cf4b499c8e9558438e70b0da5f0f38f54aPaul Berry dst.set(brw, dst_mt, dst_level, dst_layer); 1643506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 16448c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke /* If we are blitting from sRGB to linear or vice versa, we still want the 16458c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * blit to be a direct copy, so we need source and destination to use the 16468c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * same format. However, we want the destination sRGB/linear state to be 16478c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * correct (so that sRGB blending is used when doing an MSAA resolve to an 16488c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * sRGB surface, and linear blending is used when doing an MSAA resolve to 16498c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * a linear surface). Since blorp blits don't support any format 16508c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * conversion (except between sRGB and linear), we can accomplish this by 16518c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke * simply setting up the source to use the same format as the destination. 16528c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke */ 16538c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke assert(_mesa_get_srgb_format_linear(src_mt->format) == 16548c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke _mesa_get_srgb_format_linear(dst_mt->format)); 16558c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke src.brw_surfaceformat = dst.brw_surfaceformat; 16568c1c18769ef4838b11065b353f6f62bfd1de1cd2Kenneth Graunke 1657506d70be21cd3469118de89297cba0c0f709c1aePaul Berry use_wm_prog = true; 1658506d70be21cd3469118de89297cba0c0f709c1aePaul Berry memset(&wm_prog_key, 0, sizeof(wm_prog_key)); 1659506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1660e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry /* texture_data_type indicates the register type that should be used to 1661e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry * manipulate texture data. 1662e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry */ 1663e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry switch (_mesa_get_format_datatype(src_mt->format)) { 1664e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry case GL_UNSIGNED_NORMALIZED: 1665e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry case GL_SIGNED_NORMALIZED: 1666e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry case GL_FLOAT: 1667e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_F; 1668e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry break; 1669e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry case GL_UNSIGNED_INT: 1670e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry if (src_mt->format == MESA_FORMAT_S8) { 1671e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry /* We process stencil as though it's an unsigned normalized color */ 1672e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_F; 1673e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry } else { 1674e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_UD; 1675e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry } 1676e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry break; 1677e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry case GL_INT: 1678e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry wm_prog_key.texture_data_type = BRW_REGISTER_TYPE_D; 1679e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry break; 1680e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry default: 1681e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry assert(!"Unrecognized blorp format"); 1682e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry break; 1683e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry } 1684e5d983267a98bf9f73f0ea981eaca339b975a8dbPaul Berry 16858b1f467cce34340637e9baca4847fc5273cf7541Paul Berry if (brw->intel.gen > 6) { 168660c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry /* Gen7's rendering hardware only supports the IMS layout for depth and 168760c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * stencil render targets. Blorp always maps its destination surface as 168860c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * a color render target (even if it's actually a depth or stencil 168960c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * buffer). So if the destination is IMS, we'll have to map it as a 169060c3e69dbf297426c42e4b8f94c5f0493bd9be5fPaul Berry * single-sampled texture and interleave the samples ourselves. 16918b1f467cce34340637e9baca4847fc5273cf7541Paul Berry */ 16921bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry if (dst_mt->msaa_layout == INTEL_MSAA_LAYOUT_IMS) 16938b1f467cce34340637e9baca4847fc5273cf7541Paul Berry dst.num_samples = 0; 16948b1f467cce34340637e9baca4847fc5273cf7541Paul Berry } 16958b1f467cce34340637e9baca4847fc5273cf7541Paul Berry 169697fc89c6cbaa3b5ef7f678d2dc2c7d5bbba05315Paul Berry if (dst.map_stencil_as_y_tiled && dst.num_samples > 1) { 169734a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry /* If the destination surface is a W-tiled multisampled stencil buffer 169834a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * that we're mapping as Y tiled, then we need to arrange for the WM 169934a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * program to run once per sample rather than once per pixel, because 170034a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * the memory layout of related samples doesn't match between W and Y 170134a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * tiling. 1702233c207e9e477b6b0a5c6705e727129b92989073Paul Berry */ 170334a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry wm_prog_key.persample_msaa_dispatch = true; 1704233c207e9e477b6b0a5c6705e727129b92989073Paul Berry } 1705233c207e9e477b6b0a5c6705e727129b92989073Paul Berry 170697fc89c6cbaa3b5ef7f678d2dc2c7d5bbba05315Paul Berry if (src.num_samples > 0 && dst.num_samples > 1) { 170719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* We are blitting from a multisample buffer to a multisample buffer, so 170819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * we must preserve samples within a pixel. This means we have to 170934a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * arrange for the WM program to run once per sample rather than once 171034a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * per pixel. 171119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 171234a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry wm_prog_key.persample_msaa_dispatch = true; 171319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 171419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 171519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* The render path must be configured to use the same number of samples as 171619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * the destination buffer. 171719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 171819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry num_samples = dst.num_samples; 171919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 172019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry GLenum base_format = _mesa_get_format_base_format(src_mt->format); 172119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry if (base_format != GL_DEPTH_COMPONENT && /* TODO: what about depth/stencil? */ 172219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry base_format != GL_STENCIL_INDEX && 172397fc89c6cbaa3b5ef7f678d2dc2c7d5bbba05315Paul Berry src_mt->num_samples > 1 && dst_mt->num_samples <= 1) { 172419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* We are downsampling a color buffer, so blend. */ 172519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry wm_prog_key.blend = true; 172619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 172719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 172819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* src_samples and dst_samples are the true sample counts */ 172919e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry wm_prog_key.src_samples = src_mt->num_samples; 173019e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry wm_prog_key.dst_samples = dst_mt->num_samples; 173119e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 173219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* tex_samples and rt_samples are the sample counts that are set up in 173319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * SURFACE_STATE. 173419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 173519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry wm_prog_key.tex_samples = src.num_samples; 173619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry wm_prog_key.rt_samples = dst.num_samples; 173719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 17381bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* tex_layout and rt_layout indicate the MSAA layout the GPU pipeline will 17391bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * use to access the source and destination surfaces. 174067b0f7c7dddeb92ee4d24ed3977e20b70f5674f6Paul Berry */ 17411bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry wm_prog_key.tex_layout = 17421bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry compute_msaa_layout_for_pipeline(brw, src.num_samples, src.msaa_layout); 17431bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry wm_prog_key.rt_layout = 17441bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry compute_msaa_layout_for_pipeline(brw, dst.num_samples, dst.msaa_layout); 174567b0f7c7dddeb92ee4d24ed3977e20b70f5674f6Paul Berry 17461bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry /* src_layout and dst_layout indicate the true MSAA layout used by src and 17471bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * dst. 17488b1f467cce34340637e9baca4847fc5273cf7541Paul Berry */ 17491bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry wm_prog_key.src_layout = src_mt->msaa_layout; 17501bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry wm_prog_key.dst_layout = dst_mt->msaa_layout; 17518b1f467cce34340637e9baca4847fc5273cf7541Paul Berry 1752506d70be21cd3469118de89297cba0c0f709c1aePaul Berry wm_prog_key.src_tiled_w = src.map_stencil_as_y_tiled; 1753506d70be21cd3469118de89297cba0c0f709c1aePaul Berry wm_prog_key.dst_tiled_w = dst.map_stencil_as_y_tiled; 1754506d70be21cd3469118de89297cba0c0f709c1aePaul Berry x0 = wm_push_consts.dst_x0 = dst_x0; 1755506d70be21cd3469118de89297cba0c0f709c1aePaul Berry y0 = wm_push_consts.dst_y0 = dst_y0; 1756506d70be21cd3469118de89297cba0c0f709c1aePaul Berry x1 = wm_push_consts.dst_x1 = dst_x1; 1757506d70be21cd3469118de89297cba0c0f709c1aePaul Berry y1 = wm_push_consts.dst_y1 = dst_y1; 1758506d70be21cd3469118de89297cba0c0f709c1aePaul Berry wm_push_consts.x_transform.setup(src_x0, dst_x0, dst_x1, mirror_x); 1759506d70be21cd3469118de89297cba0c0f709c1aePaul Berry wm_push_consts.y_transform.setup(src_y0, dst_y0, dst_y1, mirror_y); 1760506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 176197fc89c6cbaa3b5ef7f678d2dc2c7d5bbba05315Paul Berry if (dst.num_samples <= 1 && dst_mt->num_samples > 1) { 176219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry /* We must expand the rectangle we send through the rendering pipeline, 176319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * to account for the fact that we are mapping the destination region as 176419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * single-sampled when it is in fact multisampled. We must also align 176519e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * it to a multiple of the multisampling pattern, because the 176619e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * differences between multisampled and single-sampled surface formats 176719e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * will mean that pixels are scrambled within the multisampling pattern. 176819e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry * TODO: what if this makes the coordinates too large? 17698b1f467cce34340637e9baca4847fc5273cf7541Paul Berry * 17701bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * Note: this only works if the destination surface uses the IMS layout. 17711bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * If it's UMS, then we have no choice but to set up the rendering 17721bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry * pipeline as multisampled. 177319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry */ 17741bd4d456cdecf7bea55f4e3dac574af54efad994Paul Berry assert(dst_mt->msaa_layout == INTEL_MSAA_LAYOUT_IMS); 1775082874e3891e588f674508be6578f600b35852c4Paul Berry switch (dst_mt->num_samples) { 1776082874e3891e588f674508be6578f600b35852c4Paul Berry case 4: 17774df2848786d4778a2ce7dbf2e046e191036ccb56Paul Berry x0 = ROUND_DOWN_TO(x0 * 2, 4); 17784df2848786d4778a2ce7dbf2e046e191036ccb56Paul Berry y0 = ROUND_DOWN_TO(y0 * 2, 4); 1779082874e3891e588f674508be6578f600b35852c4Paul Berry x1 = ALIGN(x1 * 2, 4); 1780082874e3891e588f674508be6578f600b35852c4Paul Berry y1 = ALIGN(y1 * 2, 4); 1781082874e3891e588f674508be6578f600b35852c4Paul Berry break; 1782082874e3891e588f674508be6578f600b35852c4Paul Berry case 8: 17834df2848786d4778a2ce7dbf2e046e191036ccb56Paul Berry x0 = ROUND_DOWN_TO(x0 * 4, 8); 17844df2848786d4778a2ce7dbf2e046e191036ccb56Paul Berry y0 = ROUND_DOWN_TO(y0 * 2, 4); 1785082874e3891e588f674508be6578f600b35852c4Paul Berry x1 = ALIGN(x1 * 4, 8); 1786082874e3891e588f674508be6578f600b35852c4Paul Berry y1 = ALIGN(y1 * 2, 4); 1787082874e3891e588f674508be6578f600b35852c4Paul Berry break; 1788082874e3891e588f674508be6578f600b35852c4Paul Berry default: 1789082874e3891e588f674508be6578f600b35852c4Paul Berry assert(!"Unrecognized sample count in brw_blorp_blit_params ctor"); 1790082874e3891e588f674508be6578f600b35852c4Paul Berry break; 1791082874e3891e588f674508be6578f600b35852c4Paul Berry } 179219e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry wm_prog_key.use_kill = true; 179319e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry } 179419e9b24626c2b9d7abef054d57bb2a52106c545bPaul Berry 1795506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (dst.map_stencil_as_y_tiled) { 1796602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry /* We must modify the rectangle we send through the rendering pipeline 179776c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * (and the size and x/y offset of the destination surface), to account 179876c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * for the fact that we are mapping it as Y-tiled when it is in fact 179976c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * W-tiled. 180021e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * 180121e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * Both Y tiling and W tiling can be understood as organizations of 180221e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * 32-byte sub-tiles; within each 32-byte sub-tile, the layout of pixels 180321e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * is different, but the layout of the 32-byte sub-tiles within the 4k 180421e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * tile is the same (8 sub-tiles across by 16 sub-tiles down, in 180521e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * column-major order). In Y tiling, the sub-tiles are 16 bytes wide 180621e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * and 2 rows high; in W tiling, they are 8 bytes wide and 4 rows high. 180721e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * 180821e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * Therefore, to account for the layout differences within the 32-byte 180921e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * sub-tiles, we must expand the rectangle so the X coordinates of its 181021e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * edges are multiples of 8 (the W sub-tile width), and its Y 181121e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * coordinates of its edges are multiples of 4 (the W sub-tile height). 181221e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * Then we need to scale the X and Y coordinates of the rectangle to 181321e9850d5369f9757b5005df4c8af38668a3053bPaul Berry * account for the differences in aspect ratio between the Y and W 181476c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * sub-tiles. We need to modify the layer width and height similarly. 181576c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * 181636bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry * A correction needs to be applied when MSAA is in use: since 181736bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry * INTEL_MSAA_LAYOUT_IMS uses an interleaving pattern whose height is 4, 181836bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry * we need to align the Y coordinates to multiples of 8, so that when 181936bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry * they are divided by two they are still multiples of 4. 182036bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry * 182176c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * Note: Since the x/y offset of the surface will be applied using the 182276c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * SURFACE_STATE command packet, it will be invisible to the swizzling 182376c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * code in the shader; therefore it needs to be in a multiple of the 182476c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * 32-byte sub-tile size. Fortunately it is, since the sub-tile is 8 182576c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * pixels wide and 4 pixels high (when viewed as a W-tiled stencil 182676c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * buffer), and the miplevel alignment used for stencil buffers is 8 182776c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * pixels horizontally and either 4 or 8 pixels vertically (see 182876c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * intel_horizontal_texture_alignment_unit() and 182976c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * intel_vertical_texture_alignment_unit()). 183076c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * 183176c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * Note: Also, since the SURFACE_STATE command packet can only apply 183276c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * offsets that are multiples of 4 pixels horizontally and 2 pixels 183376c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * vertically, it is important that the offsets will be multiples of 183476c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * these sizes after they are converted into Y-tiled coordinates. 183576c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * Fortunately they will be, since we know from above that the offsets 183676c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * are a multiple of the 32-byte sub-tile size, and in Y-tiled 183776c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * coordinates the sub-tile is 16 pixels wide and 2 pixels high. 183834a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry * 1839602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry * TODO: what if this makes the coordinates (or the texture size) too 1840602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry * large? 1841506d70be21cd3469118de89297cba0c0f709c1aePaul Berry */ 184236bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry const unsigned x_align = 8, y_align = dst.num_samples != 0 ? 8 : 4; 18434df2848786d4778a2ce7dbf2e046e191036ccb56Paul Berry x0 = ROUND_DOWN_TO(x0, x_align) * 2; 18444df2848786d4778a2ce7dbf2e046e191036ccb56Paul Berry y0 = ROUND_DOWN_TO(y0, y_align) / 2; 184534a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry x1 = ALIGN(x1, x_align) * 2; 184634a5f12e35dd4a5aff6683a8286d4582ba17df14Paul Berry y1 = ALIGN(y1, y_align) / 2; 184776c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry dst.width = ALIGN(dst.width, x_align) * 2; 184876c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry dst.height = ALIGN(dst.height, y_align) / 2; 184976c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry dst.x_offset *= 2; 185076c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry dst.y_offset /= 2; 1851506d70be21cd3469118de89297cba0c0f709c1aePaul Berry wm_prog_key.use_kill = true; 1852506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 1853602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry 1854602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry if (src.map_stencil_as_y_tiled) { 185576c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry /* We must modify the size and x/y offset of the source surface to 185676c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * account for the fact that we are mapping it as Y-tiled when it is in 185776c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * fact W tiled. 185876c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * 185976c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * See the comments above concerning x/y offset alignment for the 186076c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry * destination surface. 1861602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry * 1862602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry * TODO: what if this makes the texture size too large? 1863602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry */ 186436bc0fe4f2e90ea9efa19940f477472dad6fb18fPaul Berry const unsigned x_align = 8, y_align = src.num_samples != 0 ? 8 : 4; 186576c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry src.width = ALIGN(src.width, x_align) * 2; 186676c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry src.height = ALIGN(src.height, y_align) / 2; 186776c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry src.x_offset *= 2; 186876c1c34c4aa2fa48126aee8d16e943bf0e3ff750Paul Berry src.y_offset /= 2; 1869602e9a0f3727b036caf3a7b228fe90d36d832ea7Paul Berry } 1870506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1871506d70be21cd3469118de89297cba0c0f709c1aePaul Berry 1872506d70be21cd3469118de89297cba0c0f709c1aePaul Berryuint32_t 1873506d70be21cd3469118de89297cba0c0f709c1aePaul Berrybrw_blorp_blit_params::get_wm_prog(struct brw_context *brw, 1874506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_blorp_prog_data **prog_data) const 1875506d70be21cd3469118de89297cba0c0f709c1aePaul Berry{ 1876506d70be21cd3469118de89297cba0c0f709c1aePaul Berry uint32_t prog_offset; 1877506d70be21cd3469118de89297cba0c0f709c1aePaul Berry if (!brw_search_cache(&brw->cache, BRW_BLORP_BLIT_PROG, 1878506d70be21cd3469118de89297cba0c0f709c1aePaul Berry &this->wm_prog_key, sizeof(this->wm_prog_key), 1879506d70be21cd3469118de89297cba0c0f709c1aePaul Berry &prog_offset, prog_data)) { 1880506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_blorp_blit_program prog(brw, &this->wm_prog_key); 1881506d70be21cd3469118de89297cba0c0f709c1aePaul Berry GLuint program_size; 1882506d70be21cd3469118de89297cba0c0f709c1aePaul Berry const GLuint *program = prog.compile(brw, &program_size); 1883506d70be21cd3469118de89297cba0c0f709c1aePaul Berry brw_upload_cache(&brw->cache, BRW_BLORP_BLIT_PROG, 1884506d70be21cd3469118de89297cba0c0f709c1aePaul Berry &this->wm_prog_key, sizeof(this->wm_prog_key), 1885506d70be21cd3469118de89297cba0c0f709c1aePaul Berry program, program_size, 1886506d70be21cd3469118de89297cba0c0f709c1aePaul Berry &prog.prog_data, sizeof(prog.prog_data), 1887506d70be21cd3469118de89297cba0c0f709c1aePaul Berry &prog_offset, prog_data); 1888506d70be21cd3469118de89297cba0c0f709c1aePaul Berry } 1889506d70be21cd3469118de89297cba0c0f709c1aePaul Berry return prog_offset; 1890506d70be21cd3469118de89297cba0c0f709c1aePaul Berry} 1891