1/* 2 * Copyright 2003 VMware, Inc. 3 * All Rights Reserved. 4 * 5 * Permission is hereby granted, free of charge, to any person obtaining a 6 * copy of this software and associated documentation files (the 7 * "Software"), to deal in the Software without restriction, including 8 * without limitation the rights to use, copy, modify, merge, publish, 9 * distribute, sublicense, and/or sell copies of the Software, and to 10 * permit persons to whom the Software is furnished to do so, subject to 11 * the following conditions: 12 * 13 * The above copyright notice and this permission notice (including the 14 * next paragraph) shall be included in all copies or substantial portions 15 * of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 18 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 19 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 20 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR 21 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, 22 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE 23 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 24 */ 25 26#include "main/bufferobj.h" 27#include "main/image.h" 28#include "main/macros.h" 29#include "main/mtypes.h" 30#include "main/pbo.h" 31#include "main/texobj.h" 32#include "main/texstore.h" 33#include "main/texcompress.h" 34#include "main/enums.h" 35#include "drivers/common/meta.h" 36 37#include "brw_context.h" 38#include "intel_batchbuffer.h" 39#include "intel_tex.h" 40#include "intel_mipmap_tree.h" 41#include "intel_blit.h" 42#include "intel_tiled_memcpy.h" 43 44#define FILE_DEBUG_FLAG DEBUG_TEXTURE 45 46/** 47 * \brief A fast path for glTexImage and glTexSubImage. 48 * 49 * \param for_glTexImage Was this called from glTexImage or glTexSubImage? 50 * 51 * This fast path is taken when the texture format is BGRA, RGBA, 52 * A or L and when the texture memory is X- or Y-tiled. It uploads 53 * the texture data by mapping the texture memory without a GTT fence, thus 54 * acquiring a tiled view of the memory, and then copying sucessive 55 * spans within each tile. 56 * 57 * This is a performance win over the conventional texture upload path because 58 * it avoids the performance penalty of writing through the write-combine 59 * buffer. In the conventional texture upload path, 60 * texstore.c:store_texsubimage(), the texture memory is mapped through a GTT 61 * fence, thus acquiring a linear view of the memory, then each row in the 62 * image is memcpy'd. In this fast path, we replace each row's copy with 63 * a sequence of copies over each linear span in tile. 64 * 65 * One use case is Google Chrome's paint rectangles. Chrome (as 66 * of version 21) renders each page as a tiling of 256x256 GL_BGRA textures. 67 * Each page's content is initially uploaded with glTexImage2D and damaged 68 * regions are updated with glTexSubImage2D. On some workloads, the 69 * performance gain of this fastpath on Sandybridge is over 5x. 70 */ 71bool 72intel_texsubimage_tiled_memcpy(struct gl_context * ctx, 73 GLuint dims, 74 struct gl_texture_image *texImage, 75 GLint xoffset, GLint yoffset, GLint zoffset, 76 GLsizei width, GLsizei height, GLsizei depth, 77 GLenum format, GLenum type, 78 const GLvoid *pixels, 79 const struct gl_pixelstore_attrib *packing, 80 bool for_glTexImage) 81{ 82 struct brw_context *brw = brw_context(ctx); 83 struct intel_texture_image *image = intel_texture_image(texImage); 84 int src_pitch; 85 86 /* The miptree's buffer. */ 87 drm_intel_bo *bo; 88 89 int error = 0; 90 91 uint32_t cpp; 92 mem_copy_fn mem_copy = NULL; 93 94 /* This fastpath is restricted to specific texture types: 95 * a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support 96 * more types. 97 * 98 * FINISHME: The restrictions below on packing alignment and packing row 99 * length are likely unneeded now because we calculate the source stride 100 * with _mesa_image_row_stride. However, before removing the restrictions 101 * we need tests. 102 */ 103 if (!brw->has_llc || 104 !(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) || 105 !(texImage->TexObject->Target == GL_TEXTURE_2D || 106 texImage->TexObject->Target == GL_TEXTURE_RECTANGLE) || 107 pixels == NULL || 108 _mesa_is_bufferobj(packing->BufferObj) || 109 packing->Alignment > 4 || 110 packing->SkipPixels > 0 || 111 packing->SkipRows > 0 || 112 (packing->RowLength != 0 && packing->RowLength != width) || 113 packing->SwapBytes || 114 packing->LsbFirst || 115 packing->Invert) 116 return false; 117 118 /* Only a simple blit, no scale, bias or other mapping. */ 119 if (ctx->_ImageTransferState) 120 return false; 121 122 if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp)) 123 return false; 124 125 /* If this is a nontrivial texture view, let another path handle it instead. */ 126 if (texImage->TexObject->MinLayer) 127 return false; 128 129 if (for_glTexImage) 130 ctx->Driver.AllocTextureImageBuffer(ctx, texImage); 131 132 if (!image->mt || 133 (image->mt->tiling != I915_TILING_X && 134 image->mt->tiling != I915_TILING_Y)) { 135 /* The algorithm is written only for X- or Y-tiled memory. */ 136 return false; 137 } 138 139 /* Since we are going to write raw data to the miptree, we need to resolve 140 * any pending fast color clears before we start. 141 */ 142 intel_miptree_all_slices_resolve_color(brw, image->mt, 0); 143 144 bo = image->mt->bo; 145 146 if (drm_intel_bo_references(brw->batch.bo, bo)) { 147 perf_debug("Flushing before mapping a referenced bo.\n"); 148 intel_batchbuffer_flush(brw); 149 } 150 151 error = brw_bo_map(brw, bo, true /* write enable */, "miptree"); 152 if (error || bo->virtual == NULL) { 153 DBG("%s: failed to map bo\n", __func__); 154 return false; 155 } 156 157 src_pitch = _mesa_image_row_stride(packing, width, format, type); 158 159 /* We postponed printing this message until having committed to executing 160 * the function. 161 */ 162 DBG("%s: level=%d offset=(%d,%d) (w,h)=(%d,%d) format=0x%x type=0x%x " 163 "mesa_format=0x%x tiling=%d " 164 "packing=(alignment=%d row_length=%d skip_pixels=%d skip_rows=%d) " 165 "for_glTexImage=%d\n", 166 __func__, texImage->Level, xoffset, yoffset, width, height, 167 format, type, texImage->TexFormat, image->mt->tiling, 168 packing->Alignment, packing->RowLength, packing->SkipPixels, 169 packing->SkipRows, for_glTexImage); 170 171 int level = texImage->Level + texImage->TexObject->MinLevel; 172 173 /* Adjust x and y offset based on miplevel */ 174 xoffset += image->mt->level[level].level_x; 175 yoffset += image->mt->level[level].level_y; 176 177 linear_to_tiled( 178 xoffset * cpp, (xoffset + width) * cpp, 179 yoffset, yoffset + height, 180 bo->virtual, 181 pixels - (ptrdiff_t) yoffset * src_pitch - (ptrdiff_t) xoffset * cpp, 182 image->mt->pitch, src_pitch, 183 brw->has_swizzling, 184 image->mt->tiling, 185 mem_copy 186 ); 187 188 drm_intel_bo_unmap(bo); 189 return true; 190} 191 192static void 193intelTexSubImage(struct gl_context * ctx, 194 GLuint dims, 195 struct gl_texture_image *texImage, 196 GLint xoffset, GLint yoffset, GLint zoffset, 197 GLsizei width, GLsizei height, GLsizei depth, 198 GLenum format, GLenum type, 199 const GLvoid * pixels, 200 const struct gl_pixelstore_attrib *packing) 201{ 202 struct intel_mipmap_tree *mt = intel_texture_image(texImage)->mt; 203 bool ok; 204 205 bool tex_busy = mt && drm_intel_bo_busy(mt->bo); 206 207 if (mt && mt->format == MESA_FORMAT_S_UINT8) 208 mt->r8stencil_needs_update = true; 209 210 DBG("%s mesa_format %s target %s format %s type %s level %d %dx%dx%d\n", 211 __func__, _mesa_get_format_name(texImage->TexFormat), 212 _mesa_enum_to_string(texImage->TexObject->Target), 213 _mesa_enum_to_string(format), _mesa_enum_to_string(type), 214 texImage->Level, texImage->Width, texImage->Height, texImage->Depth); 215 216 ok = _mesa_meta_pbo_TexSubImage(ctx, dims, texImage, 217 xoffset, yoffset, zoffset, 218 width, height, depth, format, type, 219 pixels, tex_busy, packing); 220 if (ok) 221 return; 222 223 ok = intel_texsubimage_tiled_memcpy(ctx, dims, texImage, 224 xoffset, yoffset, zoffset, 225 width, height, depth, 226 format, type, pixels, packing, 227 false /*for_glTexImage*/); 228 if (ok) 229 return; 230 231 _mesa_store_texsubimage(ctx, dims, texImage, 232 xoffset, yoffset, zoffset, 233 width, height, depth, 234 format, type, pixels, packing); 235} 236 237void 238intelInitTextureSubImageFuncs(struct dd_function_table *functions) 239{ 240 functions->TexSubImage = intelTexSubImage; 241} 242