diff options
Diffstat (limited to 'src/mesa/drivers/dri')
68 files changed, 1872 insertions, 1000 deletions
diff --git a/src/mesa/drivers/dri/Makefile.template b/src/mesa/drivers/dri/Makefile.template index 5c01d233c1..bd38e3be47 100644 --- a/src/mesa/drivers/dri/Makefile.template +++ b/src/mesa/drivers/dri/Makefile.template @@ -92,7 +92,7 @@ clean: install: $(LIBNAME) $(INSTALL) -d $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR) - $(INSTALL) -m 755 $(LIBNAME) $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR) + $(MINSTALL) -m 755 $(LIBNAME) $(DESTDIR)$(DRI_DRIVER_INSTALL_DIR) -include depend diff --git a/src/mesa/drivers/dri/common/extension_helper.h b/src/mesa/drivers/dri/common/extension_helper.h index 8dcaaee307..f5e35e4161 100644 --- a/src/mesa/drivers/dri/common/extension_helper.h +++ b/src/mesa/drivers/dri/common/extension_helper.h @@ -1763,6 +1763,13 @@ static const char DeleteFencesNV_names[] = ""; #endif +#if defined(need_GL_SGIX_polynomial_ffd) +static const char DeformationMap3dSGIX_names[] = + "iddiiddiiddiip\0" /* Parameter signature */ + "glDeformationMap3dSGIX\0" + ""; +#endif + #if defined(need_GL_VERSION_2_0) static const char IsShader_names[] = "i\0" /* Parameter signature */ @@ -4396,6 +4403,13 @@ static const char WindowPos3ivMESA_names[] = ""; #endif +#if defined(need_GL_ARB_copy_buffer) +static const char CopyBufferSubData_names[] = + "iiiii\0" /* Parameter signature */ + "glCopyBufferSubData\0" + ""; +#endif + #if defined(need_GL_VERSION_1_5) || defined(need_GL_ARB_vertex_buffer_object) static const char IsBufferARB_names[] = "i\0" /* Parameter signature */ @@ -4560,13 +4574,6 @@ static const char Minmax_names[] = ""; #endif -#if defined(need_GL_SGIX_polynomial_ffd) -static const char DeformationMap3dSGIX_names[] = - "iddiiddiiddiip\0" /* Parameter signature */ - "glDeformationMap3dSGIX\0" - ""; -#endif - #if defined(need_GL_VERSION_1_4) || defined(need_GL_EXT_fog_coord) static const char FogCoorddvEXT_names[] = "p\0" /* Parameter signature */ @@ -4939,6 +4946,13 @@ static const struct dri_extension_function GL_APPLE_vertex_array_object_function }; #endif +#if defined(need_GL_ARB_copy_buffer) +static const struct dri_extension_function GL_ARB_copy_buffer_functions[] = { + { CopyBufferSubData_names, CopyBufferSubData_remap_index, -1 }, + { NULL, 0, 0 } +}; +#endif + #if defined(need_GL_ARB_draw_buffers) static const struct dri_extension_function GL_ARB_draw_buffers_functions[] = { { DrawBuffersARB_names, DrawBuffersARB_remap_index, -1 }, @@ -6055,9 +6069,9 @@ static const struct dri_extension_function GL_SGIX_pixel_texture_functions[] = { #if defined(need_GL_SGIX_polynomial_ffd) static const struct dri_extension_function GL_SGIX_polynomial_ffd_functions[] = { { LoadIdentityDeformationMapSGIX_names, LoadIdentityDeformationMapSGIX_remap_index, -1 }, + { DeformationMap3dSGIX_names, DeformationMap3dSGIX_remap_index, -1 }, { DeformSGIX_names, DeformSGIX_remap_index, -1 }, { DeformationMap3fSGIX_names, DeformationMap3fSGIX_remap_index, -1 }, - { DeformationMap3dSGIX_names, DeformationMap3dSGIX_remap_index, -1 }, { NULL, 0, 0 } }; #endif diff --git a/src/mesa/drivers/dri/i915/Makefile b/src/mesa/drivers/dri/i915/Makefile index 9f4bd1699f..beaf9a4b12 100644 --- a/src/mesa/drivers/dri/i915/Makefile +++ b/src/mesa/drivers/dri/i915/Makefile @@ -19,6 +19,7 @@ DRIVER_SOURCES = \ intel_batchbuffer.c \ intel_clear.c \ intel_extensions.c \ + intel_generatemipmap.c \ intel_mipmap_tree.c \ intel_tex_layout.c \ intel_tex_image.c \ diff --git a/src/mesa/drivers/dri/i915/i830_reg.h b/src/mesa/drivers/dri/i915/i830_reg.h index d210c2d08e..db16871001 100644 --- a/src/mesa/drivers/dri/i915/i830_reg.h +++ b/src/mesa/drivers/dri/i915/i830_reg.h @@ -48,19 +48,6 @@ #define AA_LINE_ENABLE ((1<<1) | 1) #define AA_LINE_DISABLE (1<<1) -#define _3DSTATE_BUF_INFO_CMD (CMD_3D | (0x1d<<24) | (0x8e<<16) | 1) -/* Dword 1 */ -#define BUF_3D_ID_COLOR_BACK (0x3<<24) -#define BUF_3D_ID_DEPTH (0x7<<24) -#define BUF_3D_USE_FENCE (1<<23) -#define BUF_3D_TILED_SURFACE (1<<22) -#define BUF_3D_TILE_WALK_X 0 -#define BUF_3D_TILE_WALK_Y (1<<21) -#define BUF_3D_PITCH(x) (((x)/4)<<2) -/* Dword 2 */ -#define BUF_3D_ADDR(x) ((x) & ~0x3) - - #define _3DSTATE_COLOR_FACTOR_CMD (CMD_3D | (0x1d<<24) | (0x1<<16)) #define _3DSTATE_COLOR_FACTOR_N_CMD(stage) (CMD_3D | (0x1d<<24) | \ diff --git a/src/mesa/drivers/dri/i915/i830_state.c b/src/mesa/drivers/dri/i915/i830_state.c index d9cad0c4bf..8ef6c9144f 100644 --- a/src/mesa/drivers/dri/i915/i830_state.c +++ b/src/mesa/drivers/dri/i915/i830_state.c @@ -39,6 +39,7 @@ #include "intel_screen.h" #include "intel_batchbuffer.h" #include "intel_fbo.h" +#include "intel_buffers.h" #include "i830_context.h" #include "i830_reg.h" @@ -446,6 +447,24 @@ i830DepthMask(GLcontext * ctx, GLboolean flag) i830->state.Ctx[I830_CTXREG_ENABLES_2] |= DISABLE_DEPTH_WRITE; } +/** Called from ctx->Driver.Viewport() */ +static void +i830Viewport(GLcontext * ctx, + GLint x, GLint y, GLsizei width, GLsizei height) +{ + intelCalcViewport(ctx); + + intel_viewport(ctx, x, y, width, height); +} + + +/** Called from ctx->Driver.DepthRange() */ +static void +i830DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval) +{ + intelCalcViewport(ctx); +} + /* ============================================================= * Polygon stipple * @@ -1064,6 +1083,8 @@ i830InitStateFuncs(struct dd_function_table *functions) functions->StencilFuncSeparate = i830StencilFuncSeparate; functions->StencilMaskSeparate = i830StencilMaskSeparate; functions->StencilOpSeparate = i830StencilOpSeparate; + functions->DepthRange = i830DepthRange; + functions->Viewport = i830Viewport; } void diff --git a/src/mesa/drivers/dri/i915/i830_texstate.c b/src/mesa/drivers/dri/i915/i830_texstate.c index 753c25b57e..6f998fa6f7 100644 --- a/src/mesa/drivers/dri/i915/i830_texstate.c +++ b/src/mesa/drivers/dri/i915/i830_texstate.c @@ -174,14 +174,16 @@ i830_update_tex_unit(struct intel_context *intel, GLuint unit, GLuint ss3) state[I830_TEXREG_TM0LI] = (_3DSTATE_LOAD_STATE_IMMEDIATE_2 | (LOAD_TEXTURE_MAP0 << unit) | 4); -/* state[I830_TEXREG_TM0S0] = (TM0S0_USE_FENCE | */ -/* t->intel.TextureOffset); */ - - state[I830_TEXREG_TM0S1] = (((firstImage->Height - 1) << TM0S1_HEIGHT_SHIFT) | ((firstImage->Width - 1) << TM0S1_WIDTH_SHIFT) | format); + if (intelObj->mt->region->tiling != I915_TILING_NONE) { + state[I830_TEXREG_TM0S1] |= TM0S1_TILED_SURFACE; + if (intelObj->mt->region->tiling == I915_TILING_Y) + state[I830_TEXREG_TM0S1] |= TM0S1_TILE_WALK; + } + state[I830_TEXREG_TM0S2] = ((((pitch / 4) - 1) << TM0S2_PITCH_SHIFT) | TM0S2_CUBE_FACE_ENA_MASK); diff --git a/src/mesa/drivers/dri/i915/i830_vtbl.c b/src/mesa/drivers/dri/i915/i830_vtbl.c index 3bf02de61f..cbee9f9efe 100644 --- a/src/mesa/drivers/dri/i915/i830_vtbl.c +++ b/src/mesa/drivers/dri/i915/i830_vtbl.c @@ -552,7 +552,7 @@ i830_emit_state(struct intel_context *intel) if (state->tex_buffer[i]) { OUT_RELOC(state->tex_buffer[i], I915_GEM_DOMAIN_SAMPLER, 0, - state->tex_offset[i] | TM0S0_USE_FENCE); + state->tex_offset[i]); } else if (state == &i830->meta) { assert(i == 0); @@ -634,21 +634,11 @@ i830_state_draw_region(struct intel_context *intel, /* * Set stride/cpp values */ - if (color_region) { - state->Buffer[I830_DESTREG_CBUFADDR0] = _3DSTATE_BUF_INFO_CMD; - state->Buffer[I830_DESTREG_CBUFADDR1] = - (BUF_3D_ID_COLOR_BACK | - BUF_3D_PITCH(color_region->pitch * color_region->cpp) | - BUF_3D_USE_FENCE); - } + i915_set_buf_info_for_region(&state->Buffer[I830_DESTREG_CBUFADDR0], + color_region, BUF_3D_ID_COLOR_BACK); - if (depth_region) { - state->Buffer[I830_DESTREG_DBUFADDR0] = _3DSTATE_BUF_INFO_CMD; - state->Buffer[I830_DESTREG_DBUFADDR1] = - (BUF_3D_ID_DEPTH | - BUF_3D_PITCH(depth_region->pitch * depth_region->cpp) | - BUF_3D_USE_FENCE); - } + i915_set_buf_info_for_region(&state->Buffer[I830_DESTREG_DBUFADDR0], + depth_region, BUF_3D_ID_DEPTH); /* * Compute/set I830_DESTREG_DV1 value @@ -718,26 +708,6 @@ i830_set_draw_region(struct intel_context *intel, i830_state_draw_region(intel, &i830->state, color_regions[0], depth_region); } -#if 0 -static void -i830_update_color_z_regions(intelContextPtr intel, - const intelRegion * colorRegion, - const intelRegion * depthRegion) -{ - i830ContextPtr i830 = I830_CONTEXT(intel); - - i830->state.Buffer[I830_DESTREG_CBUFADDR1] = - (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(colorRegion->pitch) | - BUF_3D_USE_FENCE); - i830->state.Buffer[I830_DESTREG_CBUFADDR2] = colorRegion->offset; - - i830->state.Buffer[I830_DESTREG_DBUFADDR1] = - (BUF_3D_ID_DEPTH | BUF_3D_PITCH(depthRegion->pitch) | BUF_3D_USE_FENCE); - i830->state.Buffer[I830_DESTREG_DBUFADDR2] = depthRegion->offset; -} -#endif - - /* This isn't really handled at the moment. */ static void diff --git a/src/mesa/drivers/dri/i915/i915_context.c b/src/mesa/drivers/dri/i915/i915_context.c index 1f9f363df9..367d2a3b64 100644 --- a/src/mesa/drivers/dri/i915/i915_context.c +++ b/src/mesa/drivers/dri/i915/i915_context.c @@ -73,7 +73,7 @@ i915InvalidateState(GLcontext * ctx, GLuint new_state) p->params_uptodate = 0; } - if (new_state & (_NEW_FOG | _NEW_HINT | _NEW_PROGRAM)) + if (new_state & (_NEW_FOG | _NEW_HINT | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS)) i915_update_fog(ctx); } diff --git a/src/mesa/drivers/dri/i915/i915_reg.h b/src/mesa/drivers/dri/i915/i915_reg.h index 8891e11c6f..84db58ea95 100644 --- a/src/mesa/drivers/dri/i915/i915_reg.h +++ b/src/mesa/drivers/dri/i915/i915_reg.h @@ -93,20 +93,6 @@ /* 3DSTATE_BIN_CONTROL p141 */ -/* p143 */ -#define _3DSTATE_BUF_INFO_CMD (CMD_3D | (0x1d<<24) | (0x8e<<16) | 1) -/* Dword 1 */ -#define BUF_3D_ID_COLOR_BACK (0x3<<24) -#define BUF_3D_ID_DEPTH (0x7<<24) -#define BUF_3D_USE_FENCE (1<<23) -#define BUF_3D_TILED_SURFACE (1<<22) -#define BUF_3D_TILE_WALK_X 0 -#define BUF_3D_TILE_WALK_Y (1<<21) -#define BUF_3D_PITCH(x) (((x)/4)<<2) -/* Dword 2 */ -#define BUF_3D_ADDR(x) ((x) & ~0x3) - - /* 3DSTATE_CHROMA_KEY */ /* 3DSTATE_CLEAR_PARAMETERS, p150 */ @@ -155,6 +141,7 @@ /* p161 */ #define _3DSTATE_DST_BUF_VARS_CMD (CMD_3D | (0x1d<<24) | (0x85<<16)) /* Dword 1 */ +#define CLASSIC_EARLY_DEPTH (1<<31) #define TEX_DEFAULT_COLOR_OGL (0<<30) #define TEX_DEFAULT_COLOR_D3D (1<<30) #define ZR_EARLY_DEPTH (1<<29) diff --git a/src/mesa/drivers/dri/i915/i915_tex_layout.c b/src/mesa/drivers/dri/i915/i915_tex_layout.c index d44a2f47b3..d9588e5b56 100644 --- a/src/mesa/drivers/dri/i915/i915_tex_layout.c +++ b/src/mesa/drivers/dri/i915/i915_tex_layout.c @@ -55,6 +55,17 @@ static GLint step_offsets[6][2] = { [FACE_NEG_Z] = {-1, 1}, }; + +static GLint bottom_offsets[6] = { + [FACE_POS_X] = 16 + 0 * 8, + [FACE_POS_Y] = 16 + 1 * 8, + [FACE_POS_Z] = 16 + 2 * 8, + [FACE_NEG_X] = 16 + 3 * 8, + [FACE_NEG_Y] = 16 + 4 * 8, + [FACE_NEG_Z] = 16 + 5 * 8, +}; + + /** * Cube texture map layout for i830M-GM915. * @@ -101,7 +112,8 @@ static GLint step_offsets[6][2] = { */ static void i915_miptree_layout_cube(struct intel_context *intel, - struct intel_mipmap_tree * mt) + struct intel_mipmap_tree * mt, + uint32_t tiling) { const GLuint dim = mt->width0; GLuint face; @@ -111,7 +123,7 @@ i915_miptree_layout_cube(struct intel_context *intel, assert(lvlWidth == lvlHeight); /* cubemap images are square */ /* double pitch for cube layouts */ - mt->pitch = intel_miptree_pitch_align (intel, mt, dim * 2); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, dim * 2); mt->total_height = dim * 4; for (level = mt->first_level; level <= mt->last_level; level++) { @@ -145,7 +157,8 @@ i915_miptree_layout_cube(struct intel_context *intel, static void i915_miptree_layout_3d(struct intel_context *intel, - struct intel_mipmap_tree * mt) + struct intel_mipmap_tree * mt, + uint32_t tiling) { GLuint width = mt->width0; GLuint height = mt->height0; @@ -154,7 +167,7 @@ i915_miptree_layout_3d(struct intel_context *intel, GLint level; /* Calculate the size of a single slice. */ - mt->pitch = intel_miptree_pitch_align (intel, mt, mt->width0); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->width0); /* XXX: hardware expects/requires 9 levels at minimum. */ for (level = mt->first_level; level <= MAX2(8, mt->last_level); level++) { @@ -189,14 +202,15 @@ i915_miptree_layout_3d(struct intel_context *intel, static void i915_miptree_layout_2d(struct intel_context *intel, - struct intel_mipmap_tree * mt) + struct intel_mipmap_tree * mt, + uint32_t tiling) { GLuint width = mt->width0; GLuint height = mt->height0; GLuint img_height; GLint level; - mt->pitch = intel_miptree_pitch_align (intel, mt, mt->width0); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->width0); mt->total_height = 0; for (level = mt->first_level; level <= mt->last_level; level++) { @@ -217,19 +231,20 @@ i915_miptree_layout_2d(struct intel_context *intel, } GLboolean -i915_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree * mt) +i915_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree * mt, + uint32_t tiling) { switch (mt->target) { case GL_TEXTURE_CUBE_MAP: - i915_miptree_layout_cube(intel, mt); + i915_miptree_layout_cube(intel, mt, tiling); break; case GL_TEXTURE_3D: - i915_miptree_layout_3d(intel, mt); + i915_miptree_layout_3d(intel, mt, tiling); break; case GL_TEXTURE_1D: case GL_TEXTURE_2D: case GL_TEXTURE_RECTANGLE_ARB: - i915_miptree_layout_2d(intel, mt); + i915_miptree_layout_2d(intel, mt, tiling); break; default: _mesa_problem(NULL, "Unexpected tex target in i915_miptree_layout()"); @@ -297,7 +312,7 @@ i915_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree * mt) * +---+ +---+ +---+ +---+ +---+ +---+ * * The bottom row continues with the remaining 2x2 then the 1x1 mip contents - * in order, with each of them aligned to a 4x4 block boundary. Thus, for + * in order, with each of them aligned to a 8x8 block boundary. Thus, for * 32x32 cube maps and smaller, the bottom row layout is going to dictate the * pitch of the tree. For a tree with 4x4 images, the pitch is at least * 14 * 8 = 112 texels, for 2x2 it is at least 12 * 8 texels, and for 1x1 @@ -306,7 +321,8 @@ i915_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree * mt) static void i945_miptree_layout_cube(struct intel_context *intel, - struct intel_mipmap_tree * mt) + struct intel_mipmap_tree * mt, + uint32_t tiling) { const GLuint dim = mt->width0; GLuint face; @@ -320,9 +336,9 @@ i945_miptree_layout_cube(struct intel_context *intel, * or the final row of 4x4, 2x2 and 1x1 faces below this. */ if (dim > 32) - mt->pitch = intel_miptree_pitch_align (intel, mt, dim * 2); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, dim * 2); else - mt->pitch = intel_miptree_pitch_align (intel, mt, 14 * 8); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, 14 * 8); if (dim >= 4) mt->total_height = dim * 4 + 4; @@ -375,10 +391,11 @@ i945_miptree_layout_cube(struct intel_context *intel, x = (face - 4) * 8; break; } + break; case 2: y = mt->total_height - 4; - x = 16 + face * 8; + x = bottom_offsets[face]; break; case 1: @@ -396,7 +413,8 @@ i945_miptree_layout_cube(struct intel_context *intel, static void i945_miptree_layout_3d(struct intel_context *intel, - struct intel_mipmap_tree * mt) + struct intel_mipmap_tree * mt, + uint32_t tiling) { GLuint width = mt->width0; GLuint height = mt->height0; @@ -405,7 +423,7 @@ i945_miptree_layout_3d(struct intel_context *intel, GLuint pack_y_pitch; GLuint level; - mt->pitch = intel_miptree_pitch_align (intel, mt, mt->width0); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->width0); mt->total_height = 0; pack_y_pitch = MAX2(mt->height0, 2); @@ -450,19 +468,23 @@ i945_miptree_layout_3d(struct intel_context *intel, } GLboolean -i945_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree * mt) +i945_miptree_layout(struct intel_context *intel, struct intel_mipmap_tree * mt, + uint32_t tiling) { switch (mt->target) { case GL_TEXTURE_CUBE_MAP: - i945_miptree_layout_cube(intel, mt); + if (mt->compressed) + i945_miptree_layout_cube(intel, mt, tiling); + else + i915_miptree_layout_cube(intel, mt, tiling); break; case GL_TEXTURE_3D: - i945_miptree_layout_3d(intel, mt); + i945_miptree_layout_3d(intel, mt, tiling); break; case GL_TEXTURE_1D: case GL_TEXTURE_2D: case GL_TEXTURE_RECTANGLE_ARB: - i945_miptree_layout_2d(intel, mt); + i945_miptree_layout_2d(intel, mt, tiling); break; default: _mesa_problem(NULL, "Unexpected tex target in i945_miptree_layout()"); diff --git a/src/mesa/drivers/dri/i915/i915_texstate.c b/src/mesa/drivers/dri/i915/i915_texstate.c index a37dd7f4fb..32d4b30cf9 100644 --- a/src/mesa/drivers/dri/i915/i915_texstate.c +++ b/src/mesa/drivers/dri/i915/i915_texstate.c @@ -185,8 +185,13 @@ i915_update_tex_unit(struct intel_context *intel, GLuint unit, GLuint ss3) state[I915_TEXREG_MS3] = (((firstImage->Height - 1) << MS3_HEIGHT_SHIFT) | - ((firstImage->Width - 1) << MS3_WIDTH_SHIFT) | format | - MS3_USE_FENCE_REGS); + ((firstImage->Width - 1) << MS3_WIDTH_SHIFT) | format); + + if (intelObj->mt->region->tiling != I915_TILING_NONE) { + state[I915_TEXREG_MS3] |= MS3_TILED_SURFACE; + if (intelObj->mt->region->tiling == I915_TILING_Y) + state[I915_TEXREG_MS3] |= MS3_TILE_WALK; + } state[I915_TEXREG_MS4] = ((((pitch / 4) - 1) << MS4_PITCH_SHIFT) | MS4_CUBE_FACE_ENA_MASK | diff --git a/src/mesa/drivers/dri/i915/i915_vtbl.c b/src/mesa/drivers/dri/i915/i915_vtbl.c index 115004616f..2fca247af1 100644 --- a/src/mesa/drivers/dri/i915/i915_vtbl.c +++ b/src/mesa/drivers/dri/i915/i915_vtbl.c @@ -42,6 +42,7 @@ #include "intel_regions.h" #include "intel_tris.h" #include "intel_fbo.h" +#include "intel_chipset.h" #include "i915_reg.h" #include "i915_context.h" @@ -529,6 +530,23 @@ i915_destroy_context(struct intel_context *intel) _tnl_free_vertices(&intel->ctx); } +void +i915_set_buf_info_for_region(uint32_t *state, struct intel_region *region, + uint32_t buffer_id) +{ + state[0] = _3DSTATE_BUF_INFO_CMD; + state[1] = buffer_id; + + if (region != NULL) { + state[1] |= BUF_3D_PITCH(region->pitch * region->cpp); + + if (region->tiling != I915_TILING_NONE) { + state[1] |= BUF_3D_TILED_SURFACE; + if (region->tiling == I915_TILING_Y) + state[1] |= BUF_3D_TILE_WALK_Y; + } + } +} /** * Set the drawing regions for the color and depth/stencil buffers. @@ -562,21 +580,11 @@ i915_state_draw_region(struct intel_context *intel, /* * Set stride/cpp values */ - if (color_region) { - state->Buffer[I915_DESTREG_CBUFADDR0] = _3DSTATE_BUF_INFO_CMD; - state->Buffer[I915_DESTREG_CBUFADDR1] = - (BUF_3D_ID_COLOR_BACK | - BUF_3D_PITCH(color_region->pitch * color_region->cpp) | - BUF_3D_USE_FENCE); - } + i915_set_buf_info_for_region(&state->Buffer[I915_DESTREG_CBUFADDR0], + color_region, BUF_3D_ID_COLOR_BACK); - if (depth_region) { - state->Buffer[I915_DESTREG_DBUFADDR0] = _3DSTATE_BUF_INFO_CMD; - state->Buffer[I915_DESTREG_DBUFADDR1] = - (BUF_3D_ID_DEPTH | - BUF_3D_PITCH(depth_region->pitch * depth_region->cpp) | - BUF_3D_USE_FENCE); - } + i915_set_buf_info_for_region(&state->Buffer[I915_DESTREG_DBUFADDR0], + depth_region, BUF_3D_ID_DEPTH); /* * Compute/set I915_DESTREG_DV1 value @@ -604,6 +612,14 @@ i915_state_draw_region(struct intel_context *intel, } } + /* This isn't quite safe, thus being hidden behind an option. When changing + * the value of this bit, the pipeline needs to be MI_FLUSHed. And it + * can only be set when a depth buffer is already defined. + */ + if (IS_945(intel->intelScreen->deviceID) && intel->use_early_z && + depth_region->tiling != I915_TILING_NONE) + value |= CLASSIC_EARLY_DEPTH; + if (depth_region && depth_region->cpp == 4) { value |= DEPTH_FRMT_24_FIXED_8_OTHER; } diff --git a/src/mesa/drivers/dri/i915/intel_generatemipmap.c b/src/mesa/drivers/dri/i915/intel_generatemipmap.c new file mode 120000 index 0000000000..4c6b37ada0 --- /dev/null +++ b/src/mesa/drivers/dri/i915/intel_generatemipmap.c @@ -0,0 +1 @@ +../intel/intel_generatemipmap.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index 2934414d99..9712c38725 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -14,6 +14,7 @@ DRIVER_SOURCES = \ intel_decode.c \ intel_extensions.c \ intel_fbo.c \ + intel_generatemipmap.c \ intel_mipmap_tree.c \ intel_regions.c \ intel_screen.c \ @@ -69,6 +70,7 @@ DRIVER_SOURCES = \ brw_vs_constval.c \ brw_vs_emit.c \ brw_vs_state.c \ + brw_vs_surface_state.c \ brw_vtbl.c \ brw_wm.c \ brw_wm_debug.c \ diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index aef2ff5f86..873fc8ffff 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -131,6 +131,7 @@ struct brw_context; #define BRW_NEW_WM_INPUT_DIMENSIONS 0x100 #define BRW_NEW_INPUT_VARYING 0x200 #define BRW_NEW_PSP 0x800 +#define BRW_NEW_WM_SURFACES 0x1000 #define BRW_NEW_FENCE 0x2000 #define BRW_NEW_INDICES 0x4000 #define BRW_NEW_VERTICES 0x8000 @@ -245,6 +246,9 @@ struct brw_vs_ouput_sizes { }; +/** Number of general purpose registers (VS, WM, etc) */ +#define BRW_MAX_GRF 128 + /** Number of texture sampler units */ #define BRW_MAX_TEX_UNIT 16 @@ -450,8 +454,6 @@ struct brw_context struct { struct brw_state_flags dirty; - struct brw_tracked_state **atoms; - GLuint nr_atoms; GLuint nr_color_regions; struct intel_region *color_regions[MAX_DRAW_BUFFERS]; @@ -471,7 +473,8 @@ struct brw_context int validated_bo_count; } state; - struct brw_cache cache; + struct brw_cache cache; /** non-surface items */ + struct brw_cache surface_cache; /* surface items */ struct brw_cached_batch_item *cached_batch_items; struct { @@ -555,11 +558,6 @@ struct brw_context GLuint vs_size; GLuint total_size; - /* Dynamic tracker which changes to reflect the state referenced - * by active fp and vp program parameters: - */ - struct brw_tracked_state tracked_state; - dri_bo *curbe_bo; /** Offset within curbe_bo of space for current curbe entry */ GLuint curbe_offset; diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c index 9197fede2d..a1a6c53d0e 100644 --- a/src/mesa/drivers/dri/i965/brw_curbe.c +++ b/src/mesa/drivers/dri/i965/brw_curbe.c @@ -36,6 +36,7 @@ #include "main/macros.h" #include "main/enums.h" #include "shader/prog_parameter.h" +#include "shader/prog_print.h" #include "shader/prog_statevars.h" #include "intel_batchbuffer.h" #include "intel_regions.h" @@ -188,13 +189,6 @@ static void prepare_constant_buffer(struct brw_context *brw) GLfloat *buf; GLuint i; - /* Update our own dependency flags. This works because this - * function will also be called whenever fp or vp changes. - */ - brw->curbe.tracked_state.dirty.mesa = (_NEW_TRANSFORM|_NEW_PROJECTION); - brw->curbe.tracked_state.dirty.mesa |= vp->program.Base.Parameters->StateFlags; - brw->curbe.tracked_state.dirty.mesa |= fp->program.Base.Parameters->StateFlags; - if (sz == 0) { if (brw->curbe.last_buf) { free(brw->curbe.last_buf); @@ -335,78 +329,11 @@ static void prepare_constant_buffer(struct brw_context *brw) */ } - -/** - * Copy Mesa program parameters into given constant buffer. - */ -static void -update_constant_buffer(struct brw_context *brw, - const struct gl_program_parameter_list *params, - dri_bo *const_buffer) -{ - struct intel_context *intel = &brw->intel; - const int size = params->NumParameters * 4 * sizeof(GLfloat); - - /* copy Mesa program constants into the buffer */ - if (const_buffer && size > 0) { - - assert(const_buffer); - assert(const_buffer->size >= size); - - if (intel->intelScreen->kernel_exec_fencing) { - drm_intel_gem_bo_map_gtt(const_buffer); - memcpy(const_buffer->virtual, params->ParameterValues, size); - drm_intel_gem_bo_unmap_gtt(const_buffer); - } - else { - dri_bo_subdata(const_buffer, 0, size, params->ParameterValues); - } - - if (0) { - int i; - for (i = 0; i < params->NumParameters; i++) { - float *p = params->ParameterValues[i]; - printf("%d: %f %f %f %f\n", i, p[0], p[1], p[2], p[3]); - } - } - } -} - - -/** Copy current vertex program's parameters into the constant buffer */ -static void -update_vertex_constant_buffer(struct brw_context *brw) -{ - struct brw_vertex_program *vp = - (struct brw_vertex_program *) brw->vertex_program; - if (0) { - printf("update VS constants in buffer %p\n", vp->const_buffer); - printf("program %u\n", vp->program.Base.Id); - } - if (vp->use_const_buffer) - update_constant_buffer(brw, vp->program.Base.Parameters, vp->const_buffer); -} - - -/** Copy current fragment program's parameters into the constant buffer */ -static void -update_fragment_constant_buffer(struct brw_context *brw) -{ - struct brw_fragment_program *fp = - (struct brw_fragment_program *) brw->fragment_program; - if (fp->use_const_buffer) - update_constant_buffer(brw, fp->program.Base.Parameters, fp->const_buffer); -} - - static void emit_constant_buffer(struct brw_context *brw) { struct intel_context *intel = &brw->intel; GLuint sz = brw->curbe.total_size; - update_vertex_constant_buffer(brw); - update_fragment_constant_buffer(brw); - BEGIN_BATCH(2, IGNORE_CLIPRECTS); if (sz == 0) { OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2)); @@ -428,7 +355,7 @@ static void emit_constant_buffer(struct brw_context *brw) */ const struct brw_tracked_state brw_constant_buffer = { .dirty = { - .mesa = (_NEW_TRANSFORM|_NEW_PROJECTION), /* plus fp and vp flags */ + .mesa = _NEW_PROGRAM_CONSTANTS, .brw = (BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_VERTEX_PROGRAM | BRW_NEW_URB_FENCE | /* Implicit - hardware requires this, not used above */ diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index b91b20bec6..1b8bcc14ec 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -343,7 +343,7 @@ static void brw_prepare_vertices(struct brw_context *brw) { GLcontext *ctx = &brw->intel.ctx; struct intel_context *intel = intel_context(ctx); - GLuint tmp = brw->vs.prog_data->inputs_read; + GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; GLuint i; const unsigned char *ptr = NULL; GLuint interleave = 0; @@ -362,11 +362,11 @@ static void brw_prepare_vertices(struct brw_context *brw) _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); /* Accumulate the list of enabled arrays. */ - while (tmp) { - GLuint i = _mesa_ffsll(tmp)-1; + while (vs_inputs) { + GLuint i = _mesa_ffsll(vs_inputs) - 1; struct brw_vertex_element *input = &brw->vb.inputs[i]; - tmp &= ~(1<<i); + vs_inputs &= ~(1 << i); enabled[nr_enabled++] = input; } @@ -477,17 +477,17 @@ static void brw_emit_vertices(struct brw_context *brw) { GLcontext *ctx = &brw->intel.ctx; struct intel_context *intel = intel_context(ctx); - GLuint tmp = brw->vs.prog_data->inputs_read; + GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; GLuint i; GLuint nr_enabled = 0; /* Accumulate the list of enabled arrays. */ - while (tmp) { - i = _mesa_ffsll(tmp)-1; + while (vs_inputs) { + i = _mesa_ffsll(vs_inputs) - 1; struct brw_vertex_element *input = &brw->vb.inputs[i]; - tmp &= ~(1<<i); + vs_inputs &= ~(1 << i); enabled[nr_enabled++] = input; } diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h index 62c98bd8bb..bc7756ceab 100644 --- a/src/mesa/drivers/dri/i965/brw_eu.h +++ b/src/mesa/drivers/dri/i965/brw_eu.h @@ -97,7 +97,7 @@ struct brw_glsl_call; #define BRW_EU_MAX_INSN_STACK 5 -#define BRW_EU_MAX_INSN 4000 +#define BRW_EU_MAX_INSN 10000 struct brw_compile { struct brw_instruction store[BRW_EU_MAX_INSN]; diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c index 9bc5c35139..4784254bc7 100644 --- a/src/mesa/drivers/dri/i965/brw_misc_state.c +++ b/src/mesa/drivers/dri/i965/brw_misc_state.c @@ -118,7 +118,10 @@ static void upload_binding_table_pointers(struct brw_context *brw) BEGIN_BATCH(6, IGNORE_CLIPRECTS); OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2)); - OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */ + if (brw->vs.bind_bo != NULL) + OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */ + else + OUT_BATCH(0); OUT_BATCH(0); /* gs */ OUT_BATCH(0); /* clip */ OUT_BATCH(0); /* sf */ diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c index 68fa9820b6..c99918724b 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_state.c +++ b/src/mesa/drivers/dri/i965/brw_sf_state.c @@ -66,7 +66,9 @@ static void upload_sf_vp(struct brw_context *brw) sfv.viewport.m31 = v[MAT_TY] * y_scale + y_bias; sfv.viewport.m32 = v[MAT_TZ] * depth_scale; - /* _NEW_SCISSOR */ + /* _NEW_SCISSOR | _NEW_BUFFERS | _NEW_VIEWPORT + * for DrawBuffer->_[XY]{min,max} + */ /* The scissor only needs to handle the intersection of drawable and * scissor rect. Clipping to the boundaries of static shared buffers @@ -97,7 +99,8 @@ static void upload_sf_vp(struct brw_context *brw) const struct brw_tracked_state brw_sf_vp = { .dirty = { .mesa = (_NEW_VIEWPORT | - _NEW_SCISSOR), + _NEW_SCISSOR | + _NEW_BUFFERS), .brw = 0, .cache = 0 }, diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index 81b0a45998..bf9f6cae55 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -72,11 +72,13 @@ const struct brw_tracked_state brw_sf_vp; const struct brw_tracked_state brw_state_base_address; const struct brw_tracked_state brw_urb_fence; const struct brw_tracked_state brw_vertex_state; +const struct brw_tracked_state brw_vs_surfaces; const struct brw_tracked_state brw_vs_prog; const struct brw_tracked_state brw_vs_unit; const struct brw_tracked_state brw_wm_input_sizes; const struct brw_tracked_state brw_wm_prog; const struct brw_tracked_state brw_wm_samplers; +const struct brw_tracked_state brw_wm_constant_surface; const struct brw_tracked_state brw_wm_surfaces; const struct brw_tracked_state brw_wm_unit; @@ -91,6 +93,20 @@ const struct brw_tracked_state brw_drawing_rect; const struct brw_tracked_state brw_indices; const struct brw_tracked_state brw_vertices; +/** + * Use same key for WM and VS surfaces. + */ +struct brw_surface_key { + GLenum target, depthmode; + dri_bo *bo; + GLint format, internal_format; + GLint first_level, last_level; + GLint width, height, depth; + GLint pitch, cpp; + uint32_t tiling; + GLuint offset; +}; + /*********************************************************************** * brw_state.c */ @@ -135,8 +151,8 @@ dri_bo *brw_search_cache( struct brw_cache *cache, void *aux_return); void brw_state_cache_check_size( struct brw_context *brw ); -void brw_init_cache( struct brw_context *brw ); -void brw_destroy_cache( struct brw_context *brw ); +void brw_init_caches( struct brw_context *brw ); +void brw_destroy_caches( struct brw_context *brw ); /*********************************************************************** * brw_state_batch.c @@ -150,4 +166,9 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, void brw_destroy_batch_cache( struct brw_context *brw ); void brw_clear_batch_cache_flush( struct brw_context *brw ); +/* brw_wm_surface_state.c */ +dri_bo * +brw_create_constant_surface( struct brw_context *brw, + struct brw_surface_key *key ); + #endif diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c index d5b5166406..e40d7a0416 100644 --- a/src/mesa/drivers/dri/i965/brw_state_cache.c +++ b/src/mesa/drivers/dri/i965/brw_state_cache.c @@ -56,9 +56,9 @@ * incorrect program is run for the other instance. */ +#include "main/imports.h" #include "brw_state.h" #include "intel_batchbuffer.h" -#include "main/imports.h" /* XXX: Fixme - have to include these to get the sizes of the prog_key * structs: @@ -69,8 +69,10 @@ #include "brw_sf.h" #include "brw_gs.h" -static GLuint hash_key( const void *key, GLuint key_size, - dri_bo **reloc_bufs, GLuint nr_reloc_bufs) + +static GLuint +hash_key(const void *key, GLuint key_size, + dri_bo **reloc_bufs, GLuint nr_reloc_bufs) { GLuint *ikey = (GLuint *)key; GLuint hash = 0, i; @@ -95,6 +97,7 @@ static GLuint hash_key( const void *key, GLuint key_size, return hash; } + /** * Marks a new buffer as being chosen for the given cache id. */ @@ -111,6 +114,7 @@ update_cache_last(struct brw_cache *cache, enum brw_cache_id cache_id, cache->brw->state.dirty.cache |= 1 << cache_id; } + static struct brw_cache_item * search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, GLuint hash, const void *key, GLuint key_size, @@ -143,7 +147,8 @@ search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, } -static void rehash( struct brw_cache *cache ) +static void +rehash(struct brw_cache *cache) { struct brw_cache_item **items; struct brw_cache_item *c, *next; @@ -164,15 +169,17 @@ static void rehash( struct brw_cache *cache ) cache->size = size; } + /** * Returns the buffer object matching cache_id and key, or NULL. */ -dri_bo *brw_search_cache( struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *key, - GLuint key_size, - dri_bo **reloc_bufs, GLuint nr_reloc_bufs, - void *aux_return ) +dri_bo * +brw_search_cache(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_size, + dri_bo **reloc_bufs, GLuint nr_reloc_bufs, + void *aux_return) { struct brw_cache_item *item; GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs); @@ -192,6 +199,7 @@ dri_bo *brw_search_cache( struct brw_cache *cache, return item->bo; } + dri_bo * brw_upload_cache( struct brw_cache *cache, enum brw_cache_id cache_id, @@ -265,7 +273,9 @@ brw_upload_cache( struct brw_cache *cache, return bo; } -/* This doesn't really work with aux data. Use search/upload instead + +/** + * This doesn't really work with aux data. Use search/upload instead */ dri_bo * brw_cache_data_sz(struct brw_cache *cache, @@ -296,6 +306,7 @@ brw_cache_data_sz(struct brw_cache *cache, return bo; } + /** * Wrapper around brw_cache_data_sz using the cache_id's canonical key size. * @@ -319,21 +330,22 @@ enum pool_type { DW_GENERAL_STATE }; + static void -brw_init_cache_id( struct brw_context *brw, - const char *name, - enum brw_cache_id id, - GLuint key_size, - GLuint aux_size) +brw_init_cache_id(struct brw_cache *cache, + const char *name, + enum brw_cache_id id, + GLuint key_size, + GLuint aux_size) { - struct brw_cache *cache = &brw->cache; - cache->name[id] = strdup(name); cache->key_size[id] = key_size; cache->aux_size[id] = aux_size; } -void brw_init_cache( struct brw_context *brw ) + +static void +brw_init_non_surface_cache(struct brw_context *brw) { struct brw_cache *cache = &brw->cache; @@ -342,114 +354,136 @@ void brw_init_cache( struct brw_context *brw ) cache->size = 7; cache->n_items = 0; cache->items = (struct brw_cache_item **) - _mesa_calloc(cache->size * - sizeof(struct brw_cache_item)); + _mesa_calloc(cache->size * sizeof(struct brw_cache_item)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CC_VP", BRW_CC_VP, sizeof(struct brw_cc_viewport), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CC_UNIT", BRW_CC_UNIT, sizeof(struct brw_cc_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "WM_PROG", BRW_WM_PROG, sizeof(struct brw_wm_prog_key), sizeof(struct brw_wm_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SAMPLER_DEFAULT_COLOR", BRW_SAMPLER_DEFAULT_COLOR, sizeof(struct brw_sampler_default_color), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SAMPLER", BRW_SAMPLER, 0, /* variable key/data size */ 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "WM_UNIT", BRW_WM_UNIT, sizeof(struct brw_wm_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SF_PROG", BRW_SF_PROG, sizeof(struct brw_sf_prog_key), sizeof(struct brw_sf_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SF_VP", BRW_SF_VP, sizeof(struct brw_sf_viewport), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SF_UNIT", BRW_SF_UNIT, sizeof(struct brw_sf_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "VS_UNIT", BRW_VS_UNIT, sizeof(struct brw_vs_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "VS_PROG", BRW_VS_PROG, sizeof(struct brw_vs_prog_key), sizeof(struct brw_vs_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CLIP_UNIT", BRW_CLIP_UNIT, sizeof(struct brw_clip_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CLIP_PROG", BRW_CLIP_PROG, sizeof(struct brw_clip_prog_key), sizeof(struct brw_clip_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "GS_UNIT", BRW_GS_UNIT, sizeof(struct brw_gs_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "GS_PROG", BRW_GS_PROG, sizeof(struct brw_gs_prog_key), sizeof(struct brw_gs_prog_data)); +} + + +static void +brw_init_surface_cache(struct brw_context *brw) +{ + struct brw_cache *cache = &brw->surface_cache; + + cache->brw = brw; - brw_init_cache_id(brw, + cache->size = 7; + cache->n_items = 0; + cache->items = (struct brw_cache_item **) + _mesa_calloc(cache->size * sizeof(struct brw_cache_item)); + + brw_init_cache_id(cache, "SS_SURFACE", BRW_SS_SURFACE, sizeof(struct brw_surface_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SS_SURF_BIND", BRW_SS_SURF_BIND, 0, 0); } + +void +brw_init_caches(struct brw_context *brw) +{ + brw_init_non_surface_cache(brw); + brw_init_surface_cache(brw); +} + + static void -brw_clear_cache( struct brw_context *brw ) +brw_clear_cache(struct brw_context *brw, struct brw_cache *cache) { struct brw_cache_item *c, *next; GLuint i; @@ -457,8 +491,8 @@ brw_clear_cache( struct brw_context *brw ) if (INTEL_DEBUG & DEBUG_STATE) _mesa_printf("%s\n", __FUNCTION__); - for (i = 0; i < brw->cache.size; i++) { - for (c = brw->cache.items[i]; c; c = next) { + for (i = 0; i < cache->size; i++) { + for (c = cache->items[i]; c; c = next) { int j; next = c->next; @@ -468,10 +502,10 @@ brw_clear_cache( struct brw_context *brw ) free((void *)c->key); free(c); } - brw->cache.items[i] = NULL; + cache->items[i] = NULL; } - brw->cache.n_items = 0; + cache->n_items = 0; if (brw->curbe.last_buf) { _mesa_free(brw->curbe.last_buf); @@ -483,25 +517,46 @@ brw_clear_cache( struct brw_context *brw ) brw->state.dirty.cache |= ~0; } -void brw_state_cache_check_size( struct brw_context *brw ) + +void +brw_state_cache_check_size(struct brw_context *brw) { + if (INTEL_DEBUG & DEBUG_STATE) + _mesa_printf("%s (n_items=%d)\n", __FUNCTION__, brw->cache.n_items); + /* un-tuned guess. We've got around 20 state objects for a total of around * 32k, so 1000 of them is around 1.5MB. */ if (brw->cache.n_items > 1000) - brw_clear_cache(brw); + brw_clear_cache(brw, &brw->cache); + + if (brw->surface_cache.n_items > 1000) + brw_clear_cache(brw, &brw->surface_cache); } -void brw_destroy_cache( struct brw_context *brw ) + +static void +brw_destroy_cache(struct brw_context *brw, struct brw_cache *cache) { GLuint i; - brw_clear_cache(brw); + if (INTEL_DEBUG & DEBUG_STATE) + _mesa_printf("%s\n", __FUNCTION__); + + brw_clear_cache(brw, cache); for (i = 0; i < BRW_MAX_CACHE; i++) { - dri_bo_unreference(brw->cache.last_bo[i]); - free(brw->cache.name[i]); + dri_bo_unreference(cache->last_bo[i]); + free(cache->name[i]); } - free(brw->cache.items); - brw->cache.items = NULL; - brw->cache.size = 0; + free(cache->items); + cache->items = NULL; + cache->size = 0; +} + + +void +brw_destroy_caches(struct brw_context *brw) +{ + brw_destroy_cache(brw, &brw->cache); + brw_destroy_cache(brw, &brw->surface_cache); } diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index 5de1450e61..c6dfea4743 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -59,11 +59,12 @@ const struct brw_tracked_state *atoms[] = &brw_curbe_offsets, &brw_recalculate_urb_fence, - &brw_cc_vp, &brw_cc_unit, - &brw_wm_surfaces, /* must do before samplers */ + &brw_vs_surfaces, /* must do before unit */ + &brw_wm_constant_surface, /* must do before wm surfaces/bind bo */ + &brw_wm_surfaces, /* must do before samplers and unit */ &brw_wm_samplers, &brw_wm_unit, @@ -88,54 +89,26 @@ const struct brw_tracked_state *atoms[] = &brw_line_stipple, &brw_aa_line_parameters, - /* Ordering of the commands below is documented as fixed. - */ -#if 0 - &brw_pipelined_state_pointers, - &brw_urb_fence, - &brw_constant_buffer_state, -#else + &brw_psp_urb_cbs, -#endif &brw_drawing_rect, &brw_indices, &brw_vertices, - NULL, /* brw_constant_buffer */ + &brw_constant_buffer }; void brw_init_state( struct brw_context *brw ) { - GLuint i; - - brw_init_cache(brw); - - brw->state.atoms = _mesa_malloc(sizeof(atoms)); - brw->state.nr_atoms = sizeof(atoms)/sizeof(*atoms); - _mesa_memcpy(brw->state.atoms, atoms, sizeof(atoms)); - - /* Patch in a pointer to the dynamic state atom: - */ - for (i = 0; i < brw->state.nr_atoms; i++) - if (brw->state.atoms[i] == NULL) - brw->state.atoms[i] = &brw->curbe.tracked_state; - - _mesa_memcpy(&brw->curbe.tracked_state, - &brw_constant_buffer, - sizeof(brw_constant_buffer)); + brw_init_caches(brw); } void brw_destroy_state( struct brw_context *brw ) { - if (brw->state.atoms) { - _mesa_free(brw->state.atoms); - brw->state.atoms = NULL; - } - - brw_destroy_cache(brw); + brw_destroy_caches(brw); brw_destroy_batch_cache(brw); } @@ -218,6 +191,7 @@ static struct dirty_bit_map mesa_bits[] = { DEFINE_BIT(_NEW_MULTISAMPLE), DEFINE_BIT(_NEW_TRACK_MATRIX), DEFINE_BIT(_NEW_PROGRAM), + DEFINE_BIT(_NEW_PROGRAM_CONSTANTS), {0, 0, 0} }; @@ -336,7 +310,7 @@ void brw_validate_state( struct brw_context *brw ) /* do prepare stage for all atoms */ for (i = 0; i < Elements(atoms); i++) { - const struct brw_tracked_state *atom = brw->state.atoms[i]; + const struct brw_tracked_state *atom = atoms[i]; if (brw->intel.Fallback) break; @@ -367,8 +341,8 @@ void brw_upload_state(struct brw_context *brw) _mesa_memset(&examined, 0, sizeof(examined)); prev = *state; - for (i = 0; i < brw->state.nr_atoms; i++) { - const struct brw_tracked_state *atom = brw->state.atoms[i]; + for (i = 0; i < Elements(atoms); i++) { + const struct brw_tracked_state *atom = atoms[i]; struct brw_state_flags generated; assert(atom->dirty.mesa || @@ -397,7 +371,7 @@ void brw_upload_state(struct brw_context *brw) } else { for (i = 0; i < Elements(atoms); i++) { - const struct brw_tracked_state *atom = brw->state.atoms[i]; + const struct brw_tracked_state *atom = atoms[i]; if (brw->intel.Fallback) break; diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c index 51a617fcb4..5c5455813a 100644 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c @@ -28,7 +28,6 @@ * Authors: * Keith Whitwell <keith@tungstengraphics.com> */ - /* Code to layout images in a mipmap tree for i965. */ @@ -40,14 +39,15 @@ #define FILE_DEBUG_FLAG DEBUG_MIPTREE -GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_tree *mt ) +GLboolean brw_miptree_layout(struct intel_context *intel, + struct intel_mipmap_tree *mt, + uint32_t tiling) { - /* XXX: these vary depending on image format: - */ -/* GLint align_w = 4; */ + /* XXX: these vary depending on image format: */ + /* GLint align_w = 4; */ switch (mt->target) { - case GL_TEXTURE_CUBE_MAP: + case GL_TEXTURE_CUBE_MAP: case GL_TEXTURE_3D: { GLuint width = mt->width0; GLuint height = mt->height0; @@ -59,25 +59,25 @@ GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_t GLuint align_w = 4; mt->total_height = 0; - + if (mt->compressed) { align_w = intel_compressed_alignment(mt->internal_format); mt->pitch = ALIGN(width, align_w); pack_y_pitch = (height + 3) / 4; } else { - mt->pitch = intel_miptree_pitch_align (intel, mt, mt->width0); - pack_y_pitch = ALIGN(mt->height0, align_h); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->width0); + pack_y_pitch = ALIGN(mt->height0, align_h); } pack_x_pitch = mt->pitch; pack_x_nr = 1; - for ( level = mt->first_level ; level <= mt->last_level ; level++ ) { + for (level = mt->first_level ; level <= mt->last_level ; level++) { GLuint nr_images = mt->target == GL_TEXTURE_3D ? depth : 6; GLint x = 0; GLint y = 0; GLint q, j; - + intel_miptree_set_level_info(mt, level, nr_images, 0, mt->total_height, width, height, depth); @@ -89,7 +89,7 @@ GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_t } x = 0; - y += pack_y_pitch; + y += pack_y_pitch; } @@ -98,40 +98,40 @@ GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_t height = minify(height); depth = minify(depth); - if (mt->compressed) { - pack_y_pitch = (height + 3) / 4; - - if (pack_x_pitch > ALIGN(width, align_w)) { - pack_x_pitch = ALIGN(width, align_w); - pack_x_nr <<= 1; - } - } else { - if (pack_x_pitch > 4) { - pack_x_pitch >>= 1; - pack_x_nr <<= 1; - assert(pack_x_pitch * pack_x_nr <= mt->pitch); - } - - if (pack_y_pitch > 2) { - pack_y_pitch >>= 1; - pack_y_pitch = ALIGN(pack_y_pitch, align_h); - } - } + if (mt->compressed) { + pack_y_pitch = (height + 3) / 4; + + if (pack_x_pitch > ALIGN(width, align_w)) { + pack_x_pitch = ALIGN(width, align_w); + pack_x_nr <<= 1; + } + } else { + if (pack_x_pitch > 4) { + pack_x_pitch >>= 1; + pack_x_nr <<= 1; + assert(pack_x_pitch * pack_x_nr <= mt->pitch); + } + + if (pack_y_pitch > 2) { + pack_y_pitch >>= 1; + pack_y_pitch = ALIGN(pack_y_pitch, align_h); + } + } } break; } default: - i945_miptree_layout_2d(intel, mt); + i945_miptree_layout_2d(intel, mt, tiling); break; } - DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, - mt->pitch, + DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, + mt->pitch, mt->total_height, mt->cpp, mt->pitch * mt->total_height * mt->cpp ); - + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c index b69616d6e5..d7f75e3685 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c @@ -69,13 +69,18 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) { GLuint i, reg = 0, mrf; -#if 0 - if (c->vp->program.Base.Parameters->NumParameters >= 6) - c->vp->use_const_buffer = 1; + /* Determine whether to use a real constant buffer or use a block + * of GRF registers for constants. The later is faster but only + * works if everything fits in the GRF. + * XXX this heuristic/check may need some fine tuning... + */ + if (c->vp->program.Base.Parameters->NumParameters + + c->vp->program.Base.NumTemporaries + 20 > BRW_MAX_GRF) + c->vp->use_const_buffer = GL_TRUE; else -#endif c->vp->use_const_buffer = GL_FALSE; - /*printf("use_const_buffer = %d\n", c->use_const_buffer);*/ + + /*printf("use_const_buffer = %d\n", c->vp->use_const_buffer);*/ /* r0 -- reserved as usual */ @@ -709,10 +714,11 @@ get_constant(struct brw_vs_compile *c, struct brw_compile *p = &c->func; struct brw_reg const_reg; struct brw_reg const2_reg; + const GLboolean relAddr = src->RelAddr; assert(argIndex < 3); - if (c->current_const[argIndex].index != src->Index || src->RelAddr) { + if (c->current_const[argIndex].index != src->Index || relAddr) { struct brw_reg addrReg = c->regs[PROGRAM_ADDRESS][0]; c->current_const[argIndex].index = src->Index; @@ -725,13 +731,13 @@ get_constant(struct brw_vs_compile *c, brw_dp_READ_4_vs(p, c->current_const[argIndex].reg,/* writeback dest */ 0, /* oword */ - src->RelAddr, /* relative indexing? */ + relAddr, /* relative indexing? */ addrReg, /* address register */ 16 * src->Index, /* byte offset */ SURF_INDEX_VERT_CONST_BUFFER /* binding table index */ ); - if (src->RelAddr) { + if (relAddr) { /* second read */ const2_reg = get_tmp(c); @@ -742,7 +748,7 @@ get_constant(struct brw_vs_compile *c, brw_dp_READ_4_vs(p, const2_reg, /* writeback dest */ 1, /* oword */ - src->RelAddr, /* relative indexing? */ + relAddr, /* relative indexing? */ addrReg, /* address register */ 16 * src->Index, /* byte offset */ SURF_INDEX_VERT_CONST_BUFFER @@ -752,7 +758,7 @@ get_constant(struct brw_vs_compile *c, const_reg = c->current_const[argIndex].reg; - if (src->RelAddr) { + if (relAddr) { /* merge the two Owords into the constant register */ /* const_reg[7..4] = const2_reg[7..4] */ brw_MOV(p, @@ -1219,7 +1225,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) for (insn = 0; insn < nr_insns; insn++) { - struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn]; + const struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn]; struct brw_reg args[3], dst; GLuint i; @@ -1232,7 +1238,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) */ if (inst->Opcode != OPCODE_SWZ) for (i = 0; i < 3; i++) { - struct prog_src_register *src = &inst->SrcReg[i]; + const struct prog_src_register *src = &inst->SrcReg[i]; index = src->Index; file = src->File; if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src) diff --git a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c new file mode 100644 index 0000000000..89f47522a1 --- /dev/null +++ b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c @@ -0,0 +1,226 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "main/mtypes.h" +#include "main/texformat.h" +#include "main/texstore.h" +#include "shader/prog_parameter.h" + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" + +/* Creates a new VS constant buffer reflecting the current VS program's + * constants, if needed by the VS program. + * + * Otherwise, constants go through the CURBEs using the brw_constant_buffer + * state atom. + */ +static drm_intel_bo * +brw_vs_update_constant_buffer(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + struct brw_vertex_program *vp = + (struct brw_vertex_program *) brw->vertex_program; + const struct gl_program_parameter_list *params = vp->program.Base.Parameters; + const int size = params->NumParameters * 4 * sizeof(GLfloat); + drm_intel_bo *const_buffer; + + /* BRW_NEW_VERTEX_PROGRAM */ + if (!vp->use_const_buffer) + return NULL; + + const_buffer = drm_intel_bo_alloc(intel->bufmgr, "vp_const_buffer", + size, 64); + + /* _NEW_PROGRAM_CONSTANTS */ + dri_bo_subdata(const_buffer, 0, size, params->ParameterValues); + + return const_buffer; +} + +/** + * Update the surface state for a VS constant buffer. + * + * Sets brw->vs.surf_bo[surf] and brw->vp->const_buffer. + */ +static void +brw_update_vs_constant_surface( GLcontext *ctx, + GLuint surf) +{ + struct brw_context *brw = brw_context(ctx); + struct brw_surface_key key; + struct brw_vertex_program *vp = + (struct brw_vertex_program *) brw->vertex_program; + const struct gl_program_parameter_list *params = vp->program.Base.Parameters; + + assert(surf == 0); + + /* If we're in this state update atom, we need to update VS constants, so + * free the old buffer and create a new one for the new contents. + */ + dri_bo_unreference(vp->const_buffer); + vp->const_buffer = brw_vs_update_constant_buffer(brw); + + /* If there's no constant buffer, then no surface BO is needed to point at + * it. + */ + if (vp->const_buffer == 0) { + drm_intel_bo_unreference(brw->vs.surf_bo[surf]); + brw->vs.surf_bo[surf] = NULL; + return; + } + + memset(&key, 0, sizeof(key)); + + key.format = MESA_FORMAT_RGBA_FLOAT32; + key.internal_format = GL_RGBA; + key.bo = vp->const_buffer; + key.depthmode = GL_NONE; + key.pitch = params->NumParameters; + key.width = params->NumParameters; + key.height = 1; + key.depth = 1; + key.cpp = 16; + + /* + printf("%s:\n", __FUNCTION__); + printf(" width %d height %d depth %d cpp %d pitch %d\n", + key.width, key.height, key.depth, key.cpp, key.pitch); + */ + + drm_intel_bo_unreference(brw->vs.surf_bo[surf]); + brw->vs.surf_bo[surf] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, key.bo ? 1 : 0, + NULL); + if (brw->vs.surf_bo[surf] == NULL) { + brw->vs.surf_bo[surf] = brw_create_constant_surface(brw, &key); + } +} + + +/** + * Constructs the binding table for the VS surface state. + */ +static dri_bo * +brw_vs_get_binding_table(struct brw_context *brw) +{ + dri_bo *bind_bo; + + bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->vs.surf_bo, BRW_VS_MAX_SURF, + NULL); + + if (bind_bo == NULL) { + GLuint data_size = BRW_VS_MAX_SURF * sizeof(GLuint); + uint32_t *data = malloc(data_size); + int i; + + for (i = 0; i < BRW_VS_MAX_SURF; i++) + if (brw->vs.surf_bo[i]) + data[i] = brw->vs.surf_bo[i]->offset; + else + data[i] = 0; + + bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->vs.surf_bo, BRW_VS_MAX_SURF, + data, data_size, + NULL, NULL); + + /* Emit binding table relocations to surface state */ + for (i = 0; i < BRW_VS_MAX_SURF; i++) { + if (brw->vs.surf_bo[i] != NULL) { + /* The presumed offsets were set in the data values for + * brw_upload_cache. + */ + drm_intel_bo_emit_reloc(bind_bo, i * 4, + brw->vs.surf_bo[i], 0, + I915_GEM_DOMAIN_INSTRUCTION, 0); + } + } + + free(data); + } + + return bind_bo; +} + +/** + * Vertex shader surfaces (constant buffer). + * + * This consumes the state updates for the constant buffer needing + * to be updated, and produces BRW_NEW_NR_VS_SURFACES for the VS unit and + * CACHE_NEW_SURF_BIND for the binding table upload. + */ +static void prepare_vs_surfaces(struct brw_context *brw ) +{ + GLcontext *ctx = &brw->intel.ctx; + int i; + int nr_surfaces = 0; + + brw_update_vs_constant_surface(ctx, SURF_INDEX_VERT_CONST_BUFFER); + + for (i = 0; i < BRW_VS_MAX_SURF; i++) { + if (brw->vs.surf_bo[i] != NULL) { + nr_surfaces = i + 1; + } + } + + if (brw->vs.nr_surfaces != nr_surfaces) { + brw->state.dirty.brw |= BRW_NEW_NR_VS_SURFACES; + brw->vs.nr_surfaces = nr_surfaces; + } + + /* Note that we don't end up updating the bind_bo if we don't have a + * surface to be pointing at. This should be relatively harmless, as it + * just slightly increases our working set size. + */ + if (brw->vs.nr_surfaces != 0) { + dri_bo_unreference(brw->vs.bind_bo); + brw->vs.bind_bo = brw_vs_get_binding_table(brw); + } +} + +const struct brw_tracked_state brw_vs_surfaces = { + .dirty = { + .mesa = (_NEW_PROGRAM_CONSTANTS), + .brw = (BRW_NEW_VERTEX_PROGRAM), + .cache = 0 + }, + .prepare = prepare_vs_surfaces, +}; + + + diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 90d74c2885..3e476fd3be 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -146,6 +146,13 @@ static void do_wm_prog( struct brw_context *brw, if (c == NULL) { brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data)); c = brw->wm.compile_data; + if (c == NULL) { + /* Ouch - big out of memory problem. Can't continue + * without triggering a segfault, no way to signal, + * so just return. + */ + return; + } } else { memset(c, 0, sizeof(*brw->wm.compile_data)); } @@ -312,6 +319,9 @@ static void brw_wm_populate_key( struct brw_context *brw, key->drawable_height = brw->intel.driDrawable->h; } + /* CACHE_NEW_VS_PROG */ + key->vp_outputs_written = brw->vs.prog_data->outputs_written & DO_SETUP_BITS; + /* The unique fragment program ID */ key->program_string_id = fp->id; } @@ -350,7 +360,7 @@ const struct brw_tracked_state brw_wm_prog = { .brw = (BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_WM_INPUT_DIMENSIONS | BRW_NEW_REDUCED_PRIMITIVE), - .cache = 0 + .cache = CACHE_NEW_VS_PROG, }, .prepare = brw_prepare_wm_prog }; diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index f0d31fc1dd..fb15c03e83 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -75,6 +75,7 @@ struct brw_wm_prog_key { GLuint program_string_id:32; GLuint origin_x, origin_y; GLuint drawable_height; + GLuint vp_outputs_written; }; @@ -240,15 +241,20 @@ struct brw_wm_compile { GLuint max_wm_grf; GLuint last_scratch; + GLuint cur_inst; /**< index of current instruction */ + + GLboolean out_of_regs; /**< ran out of GRF registers? */ + /** Mapping from Mesa registers to hardware registers */ struct { GLboolean inited; struct brw_reg reg; } wm_regs[PROGRAM_PAYLOAD+1][256][4]; + GLboolean used_grf[BRW_WM_MAX_GRF]; + GLuint first_free_grf; struct brw_reg stack; struct brw_reg emit_mask_reg; - GLuint reg_index; /**< Index of next free GRF register */ GLuint tmp_regs[BRW_WM_MAX_GRF]; GLuint tmp_index; GLuint tmp_max; diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index 72fc21d2eb..14ab9042de 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -353,6 +353,19 @@ static void emit_mad( struct brw_compile *p, } } +static void emit_trunc( struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) +{ + GLuint i; + + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + brw_RNDZ(p, dst[i], arg0[i]); + } + } +} static void emit_lrp( struct brw_compile *p, const struct brw_reg *dst, @@ -1224,6 +1237,10 @@ void brw_wm_emit( struct brw_wm_compile *c ) emit_dph(p, dst, dst_flags, args[0], args[1]); break; + case OPCODE_TRUNC: + emit_trunc(p, dst, dst_flags, args[0]); + break; + case OPCODE_LRP: emit_lrp(p, dst, dst_flags, args[0], args[1], args[2]); break; diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 117460842a..0e6a2f8ef0 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -1,5 +1,7 @@ #include "main/macros.h" #include "shader/prog_parameter.h" +#include "shader/prog_print.h" +#include "shader/prog_optimize.h" #include "brw_context.h" #include "brw_eu.h" #include "brw_wm.h" @@ -21,7 +23,6 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) const struct prog_instruction *inst = &fp->Base.Instructions[i]; switch (inst->Opcode) { case OPCODE_IF: - case OPCODE_TRUNC: case OPCODE_ENDIF: case OPCODE_CAL: case OPCODE_BRK: @@ -42,6 +43,83 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) } + +static void +reclaim_temps(struct brw_wm_compile *c); + + +/** Mark GRF register as used. */ +static void +prealloc_grf(struct brw_wm_compile *c, int r) +{ + c->used_grf[r] = GL_TRUE; +} + + +/** Mark given GRF register as not in use. */ +static void +release_grf(struct brw_wm_compile *c, int r) +{ + /*assert(c->used_grf[r]);*/ + c->used_grf[r] = GL_FALSE; + c->first_free_grf = MIN2(c->first_free_grf, r); +} + + +/** Return index of a free GRF, mark it as used. */ +static int +alloc_grf(struct brw_wm_compile *c) +{ + GLuint r; + for (r = c->first_free_grf; r < BRW_WM_MAX_GRF; r++) { + if (!c->used_grf[r]) { + c->used_grf[r] = GL_TRUE; + c->first_free_grf = r + 1; /* a guess */ + return r; + } + } + + /* no free temps, try to reclaim some */ + reclaim_temps(c); + c->first_free_grf = 0; + + /* try alloc again */ + for (r = c->first_free_grf; r < BRW_WM_MAX_GRF; r++) { + if (!c->used_grf[r]) { + c->used_grf[r] = GL_TRUE; + c->first_free_grf = r + 1; /* a guess */ + return r; + } + } + + for (r = 0; r < BRW_WM_MAX_GRF; r++) { + assert(c->used_grf[r]); + } + + /* really, no free GRF regs found */ + if (!c->out_of_regs) { + /* print warning once per compilation */ + _mesa_warning(NULL, "i965: ran out of registers for fragment program"); + c->out_of_regs = GL_TRUE; + } + + return -1; +} + + +/** Return number of GRF registers used */ +static int +num_grf_used(const struct brw_wm_compile *c) +{ + int r; + for (r = BRW_WM_MAX_GRF - 1; r >= 0; r--) + if (c->used_grf[r]) + return r + 1; + return 0; +} + + + /** * Record the mapping of a Mesa register to a hardware register. */ @@ -68,11 +146,23 @@ static int get_scalar_dst_index(const struct prog_instruction *inst) static struct brw_reg alloc_tmp(struct brw_wm_compile *c) { struct brw_reg reg; - if(c->tmp_index == c->tmp_max) - c->tmp_regs[ c->tmp_max++ ] = c->reg_index++; - + + /* if we need to allocate another temp, grow the tmp_regs[] array */ + if (c->tmp_index == c->tmp_max) { + int r = alloc_grf(c); + if (r < 0) { + /*printf("Out of temps in %s\n", __FUNCTION__);*/ + r = 50; /* XXX random register! */ + } + c->tmp_regs[ c->tmp_max++ ] = r; + } + + /* form the GRF register */ reg = brw_vec8_grf(c->tmp_regs[ c->tmp_index++ ], 0); + /*printf("alloc_temp %d\n", reg.nr);*/ + assert(reg.nr < BRW_WM_MAX_GRF); return reg; + } /** @@ -130,35 +220,29 @@ get_reg(struct brw_wm_compile *c, int file, int index, int component, return brw_null_reg(); } + assert(index < 256); + assert(component < 4); + /* see if we've already allocated a HW register for this Mesa register */ if (c->wm_regs[file][index][component].inited) { - /* yes, re-use */ - reg = c->wm_regs[file][index][component].reg; + /* yes, re-use */ + reg = c->wm_regs[file][index][component].reg; } else { /* no, allocate new register */ - reg = brw_vec8_grf(c->reg_index, 0); - } + int grf = alloc_grf(c); + /*printf("alloc grf %d for reg %d:%d.%d\n", grf, file, index, component);*/ + if (grf < 0) { + /* totally out of temps */ + grf = 51; /* XXX random register! */ + } - /* if this is a new register allocation, record it in the table */ - if (!c->wm_regs[file][index][component].inited) { - set_reg(c, file, index, component, reg); - c->reg_index++; - } + reg = brw_vec8_grf(grf, 0); + /*printf("Alloc new grf %d for %d.%d\n", reg.nr, index, component);*/ - if (c->reg_index >= BRW_WM_MAX_GRF - 12) { - /* ran out of temporary registers! */ -#if 1 - /* This is a big hack for now. - * Return bad register index, just don't hang the GPU. - */ - _mesa_fprintf(stderr, "out of regs %d\n", c->reg_index); - c->reg_index = BRW_WM_MAX_GRF - 13; -#else - return brw_null_reg(); -#endif + set_reg(c, file, index, component, reg); } - + if (neg & (1 << component)) { reg = negate(reg); } @@ -168,6 +252,46 @@ get_reg(struct brw_wm_compile *c, int file, int index, int component, } + +/** + * This is called if we run out of GRF registers. Examine the live intervals + * of temp regs in the program and free those which won't be used again. + */ +static void +reclaim_temps(struct brw_wm_compile *c) +{ + GLint intBegin[MAX_PROGRAM_TEMPS]; + GLint intEnd[MAX_PROGRAM_TEMPS]; + int index; + + /*printf("Reclaim temps:\n");*/ + + _mesa_find_temp_intervals(c->prog_instructions, c->nr_fp_insns, + intBegin, intEnd); + + for (index = 0; index < MAX_PROGRAM_TEMPS; index++) { + if (intEnd[index] != -1 && intEnd[index] < c->cur_inst) { + /* program temp[i] can be freed */ + int component; + /*printf(" temp[%d] is dead\n", index);*/ + for (component = 0; component < 4; component++) { + if (c->wm_regs[PROGRAM_TEMPORARY][index][component].inited) { + int r = c->wm_regs[PROGRAM_TEMPORARY][index][component].reg.nr; + release_grf(c, r); + /* + printf(" Reclaim temp %d, reg %d at inst %d\n", + index, r, c->cur_inst); + */ + c->wm_regs[PROGRAM_TEMPORARY][index][component].inited = GL_FALSE; + } + } + } + } +} + + + + /** * Preallocate registers. This sets up the Mesa to hardware register * mapping for certain registers, such as constants (uniforms/state vars) @@ -177,8 +301,12 @@ static void prealloc_reg(struct brw_wm_compile *c) { int i, j; struct brw_reg reg; - int nr_interp_regs = 0; + int urb_read_length = 0; GLuint inputs = FRAG_BIT_WPOS | c->fp_interp_emitted | c->fp_deriv_emitted; + GLuint reg_index = 0; + + memset(c->used_grf, GL_FALSE, sizeof(c->used_grf)); + c->first_free_grf = 0; for (i = 0; i < 4; i++) { if (i < c->key.nr_depth_regs) @@ -187,14 +315,20 @@ static void prealloc_reg(struct brw_wm_compile *c) reg = brw_vec8_grf(0, 0); set_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, i, reg); } - c->reg_index += 2 * c->key.nr_depth_regs; + reg_index += 2 * c->key.nr_depth_regs; /* constants */ { - const int nr_params = c->fp->program.Base.Parameters->NumParameters; + const GLuint nr_params = c->fp->program.Base.Parameters->NumParameters; + const GLuint nr_temps = c->fp->program.Base.NumTemporaries; /* use a real constant buffer, or just use a section of the GRF? */ - c->fp->use_const_buffer = GL_FALSE; /* (nr_params > 8);*/ + /* XXX this heuristic may need adjustment... */ + if ((nr_params + nr_temps) * 4 + reg_index > 80) + c->fp->use_const_buffer = GL_TRUE; + else + c->fp->use_const_buffer = GL_FALSE; + /*printf("WM use_const_buffer = %d\n", c->fp->use_const_buffer);*/ if (c->fp->use_const_buffer) { /* We'll use a real constant buffer and fetch constants from @@ -216,7 +350,7 @@ static void prealloc_reg(struct brw_wm_compile *c) for (i = 0; i < nr_params; i++) { /* loop over XYZW channels */ for (j = 0; j < 4; j++, index++) { - reg = brw_vec1_grf(c->reg_index + index / 8, index % 8); + reg = brw_vec1_grf(reg_index + index / 8, index % 8); /* Save pointer to parameter/constant value. * Constants will be copied in prepare_constant_buffer() */ @@ -226,28 +360,47 @@ static void prealloc_reg(struct brw_wm_compile *c) } /* number of constant regs used (each reg is float[8]) */ c->nr_creg = 2 * ((4 * nr_params + 15) / 16); - c->reg_index += c->nr_creg; + reg_index += c->nr_creg; } } /* fragment shader inputs */ - for (i = 0; i < FRAG_ATTRIB_MAX; i++) { - if (inputs & (1<<i)) { - nr_interp_regs++; - reg = brw_vec8_grf(c->reg_index, 0); - for (j = 0; j < 4; j++) - set_reg(c, PROGRAM_PAYLOAD, i, j, reg); - c->reg_index += 2; - } + for (i = 0; i < VERT_RESULT_MAX; i++) { + int fp_input; + + if (i >= VERT_RESULT_VAR0) + fp_input = i - VERT_RESULT_VAR0 + FRAG_ATTRIB_VAR0; + else if (i <= VERT_RESULT_TEX7) + fp_input = i; + else + fp_input = -1; + + if (fp_input >= 0 && inputs & (1 << fp_input)) { + urb_read_length = reg_index; + reg = brw_vec8_grf(reg_index, 0); + for (j = 0; j < 4; j++) + set_reg(c, PROGRAM_PAYLOAD, fp_input, j, reg); + } + if (c->key.vp_outputs_written & (1 << i)) { + reg_index += 2; + } } c->prog_data.first_curbe_grf = c->key.nr_depth_regs * 2; - c->prog_data.urb_read_length = nr_interp_regs * 2; + c->prog_data.urb_read_length = urb_read_length; c->prog_data.curb_read_length = c->nr_creg; - c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0); - c->reg_index++; - c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0); - c->reg_index += 2; + c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, reg_index, 0); + reg_index++; + c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg_index, 0); + reg_index += 2; + + /* mark GRF regs [0..reg_index-1] as in-use */ + for (i = 0; i < reg_index; i++) + prealloc_grf(c, i); + + /* Don't use GRF 126, 127. Using them seems to lead to GPU lock-ups */ + prealloc_grf(c, 126); + prealloc_grf(c, 127); /* An instruction may reference up to three constants. * They'll be found in these registers. @@ -256,12 +409,12 @@ static void prealloc_reg(struct brw_wm_compile *c) if (c->fp->use_const_buffer) { for (i = 0; i < 3; i++) { c->current_const[i].index = -1; - c->current_const[i].reg = alloc_tmp(c); + c->current_const[i].reg = brw_vec8_grf(alloc_grf(c), 0); } } #if 0 printf("USE CONST BUFFER? %d\n", c->fp->use_const_buffer); - printf("AFTER PRE_ALLOC, reg_index = %d\n", c->reg_index); + printf("AFTER PRE_ALLOC, reg_index = %d\n", reg_index); #endif } @@ -283,23 +436,21 @@ static void fetch_constants(struct brw_wm_compile *c, if (src->File == PROGRAM_STATE_VAR || src->File == PROGRAM_CONSTANT || src->File == PROGRAM_UNIFORM) { - if (c->current_const[i].index != src->Index) { - c->current_const[i].index = src->Index; + c->current_const[i].index = src->Index; #if 0 - printf(" fetch const[%d] for arg %d into reg %d\n", - src->Index, i, c->current_const[i].reg.nr); + printf(" fetch const[%d] for arg %d into reg %d\n", + src->Index, i, c->current_const[i].reg.nr); #endif - /* need to fetch the constant now */ - brw_dp_READ_4(p, - c->current_const[i].reg, /* writeback dest */ - 1, /* msg_reg */ - src->RelAddr, /* relative indexing? */ - 16 * src->Index, /* byte offset */ - SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */ - ); - } + /* need to fetch the constant now */ + brw_dp_READ_4(p, + c->current_const[i].reg, /* writeback dest */ + 1, /* msg_reg */ + src->RelAddr, /* relative indexing? */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */ + ); } } } @@ -368,6 +519,14 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c, const GLuint nr = 1; const GLuint component = GET_SWZ(src->Swizzle, channel); + /* Extended swizzle terms */ + if (component == SWIZZLE_ZERO) { + return brw_imm_f(0.0F); + } + else if (component == SWIZZLE_ONE) { + return brw_imm_f(1.0F); + } + if (c->fp->use_const_buffer && (src->File == PROGRAM_STATE_VAR || src->File == PROGRAM_CONSTANT || @@ -665,27 +824,26 @@ static void emit_fb_write(struct brw_wm_compile *c, } if (c->key.dest_depth_reg) { - GLuint comp = c->key.dest_depth_reg / 2; - GLuint off = c->key.dest_depth_reg % 2; + const GLuint comp = c->key.dest_depth_reg / 2; + const GLuint off = c->key.dest_depth_reg % 2; - assert(comp == 1); - assert(off == 0); -#if 0 - /* XXX do we need this code? comp always 1, off always 0, it seems */ if (off != 0) { + /* XXX this code needs review/testing */ + struct brw_reg arg1_0 = get_src_reg(c, inst, 1, comp); + struct brw_reg arg1_1 = get_src_reg(c, inst, 1, comp+1); + brw_push_insn_state(p); brw_set_compression_control(p, BRW_COMPRESSION_NONE); - brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1)); + brw_MOV(p, brw_message_reg(nr), offset(arg1_0, 1)); /* 2nd half? */ - brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]); + brw_MOV(p, brw_message_reg(nr+1), arg1_1); brw_pop_insn_state(p); } else -#endif { - struct brw_reg src = get_src_reg(c, inst, 1, 1); - brw_MOV(p, brw_message_reg(nr), src); + struct brw_reg src = get_src_reg(c, inst, 1, 1); + brw_MOV(p, brw_message_reg(nr), src); } nr += 2; } @@ -2595,7 +2753,8 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) struct brw_compile *p = &c->func; struct brw_indirect stack_index = brw_indirect(0, 0); - c->reg_index = 0; + c->out_of_regs = GL_FALSE; + prealloc_reg(c); brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack)); @@ -2603,6 +2762,8 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) for (i = 0; i < c->nr_fp_insns; i++) { const struct prog_instruction *inst = &c->prog_instructions[i]; + c->cur_inst = i; + #if 0 _mesa_printf("Inst %d: ", i); _mesa_print_instruction(inst); @@ -2833,17 +2994,13 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) _mesa_printf("unsupported IR in fragment shader %d\n", inst->Opcode); } + if (inst->CondUpdate) brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); else brw_set_predicate_control(p, BRW_PREDICATE_NONE); } post_wm_emit(c); - - if (c->reg_index >= BRW_WM_MAX_GRF) { - _mesa_problem(NULL, "Ran out of registers in brw_wm_emit_glsl()"); - /* XXX we need to do some proper error recovery here */ - } } @@ -2867,6 +3024,6 @@ void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c) brw_wm_print_program(c, "brw_wm_glsl_emit done"); } - c->prog_data.total_grf = c->reg_index; + c->prog_data.total_grf = num_grf_used(c); c->prog_data.total_scratch = 0; } diff --git a/src/mesa/drivers/dri/i965/brw_wm_iz.c b/src/mesa/drivers/dri/i965/brw_wm_iz.c index bd60ac9b31..8fd067abe7 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_iz.c +++ b/src/mesa/drivers/dri/i965/brw_wm_iz.c @@ -116,6 +116,10 @@ const struct { { C, 0, 1, 1, 1 } }; +/** + * \param line_aa AA_NEVER, AA_ALWAYS or AA_SOMETIMES + * \param lookup bitmask of IZ_* flags + */ void brw_wm_lookup_iz( GLuint line_aa, GLuint lookup, struct brw_wm_prog_key *key ) diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass1.c b/src/mesa/drivers/dri/i965/brw_wm_pass1.c index ab9aa2f10d..3436a24717 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass1.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass1.c @@ -159,6 +159,7 @@ void brw_wm_pass1( struct brw_wm_compile *c ) case OPCODE_FRC: case OPCODE_MOV: case OPCODE_SWZ: + case OPCODE_TRUNC: read0 = writemask; break; diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass2.c b/src/mesa/drivers/dri/i965/brw_wm_pass2.c index 780edbc42e..6faea018fb 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass2.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass2.c @@ -69,8 +69,6 @@ static void prealloc_reg(struct brw_wm_compile *c, */ static void init_registers( struct brw_wm_compile *c ) { - struct brw_context *brw = c->func.brw; - GLuint inputs = (brw->vs.prog_data->outputs_written & DO_SETUP_BITS); GLuint nr_interp_regs = 0; GLuint i = 0; GLuint j; @@ -85,16 +83,19 @@ static void init_registers( struct brw_wm_compile *c ) prealloc_reg(c, &c->creg[j], i++); for (j = 0; j < FRAG_ATTRIB_MAX; j++) { - if (inputs & (1<<j)) { - /* index for vs output and ps input are not the same - in shader varying */ - GLuint index; - if (j > FRAG_ATTRIB_VAR0) - index = j - (VERT_RESULT_VAR0 - FRAG_ATTRIB_VAR0); + if (c->key.vp_outputs_written & (1<<j)) { + int fp_index; + + if (j >= VERT_RESULT_VAR0) + fp_index = j - (VERT_RESULT_VAR0 - FRAG_ATTRIB_VAR0); + else if (j <= VERT_RESULT_TEX7) + fp_index = j; else - index = j; + fp_index = -1; + nr_interp_regs++; - prealloc_reg(c, &c->payload.input_interp[index], i++); + if (fp_index >= 0) + prealloc_reg(c, &c->payload.input_interp[fp_index], i++); } } diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index c604ef0162..3fc18ff1f3 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -178,6 +178,16 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key, sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; } + else if (key->tex_target == GL_TEXTURE_1D) { + /* There's a bug in 1D texture sampling - it actually pays + * attention to the wrap_t value, though it should not. + * Override the wrap_t value here to GL_REPEAT to keep + * any nonexistent border pixels from floating in. + */ + sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); + sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP; + } else { sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index 805df8a4af..c49a5f6b4e 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -176,22 +176,6 @@ static GLuint translate_tex_format( GLuint mesa_format, GLenum internal_format, } } - -/** - * Use same key for WM and VS surfaces. - */ -struct brw_surface_key { - GLenum target, depthmode; - dri_bo *bo; - GLint format, internal_format; - GLint first_level, last_level; - GLint width, height, depth; - GLint pitch, cpp; - uint32_t tiling; - GLuint offset; -}; - - static void brw_set_surface_tiling(struct brw_surface_state *surf, uint32_t tiling) { @@ -268,7 +252,7 @@ brw_create_texture_surface( struct brw_context *brw, surf.ss0.cube_neg_z = 1; } - bo = brw_upload_cache(&brw->cache, BRW_SS_SURFACE, + bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, key, sizeof(*key), &key->bo, key->bo ? 1 : 0, &surf, sizeof(surf), @@ -321,10 +305,11 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit ) key.tiling = intelObj->mt->region->tiling; dri_bo_unreference(brw->wm.surf_bo[surf]); - brw->wm.surf_bo[surf] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), - &key.bo, key.bo ? 1 : 0, - NULL); + brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, key.bo ? 1 : 0, + NULL); if (brw->wm.surf_bo[surf] == NULL) { brw->wm.surf_bo[surf] = brw_create_texture_surface(brw, &key); } @@ -336,7 +321,7 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit ) * Create the constant buffer surface. Vertex/fragment shader constants will be * read from this buffer with Data Port Read instructions/messages. */ -static dri_bo * +dri_bo * brw_create_constant_surface( struct brw_context *brw, struct brw_surface_key *key ) { @@ -362,7 +347,7 @@ brw_create_constant_surface( struct brw_context *brw, surf.ss3.pitch = (key->pitch * key->cpp) - 1; /* ignored?? */ brw_set_surface_tiling(&surf, key->tiling); /* tiling now allowed */ - bo = brw_upload_cache(&brw->cache, BRW_SS_SURFACE, + bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, key, sizeof(*key), &key->bo, key->bo ? 1 : 0, &surf, sizeof(surf), @@ -380,39 +365,70 @@ brw_create_constant_surface( struct brw_context *brw, return bo; } +/* Creates a new WM constant buffer reflecting the current fragment program's + * constants, if needed by the fragment program. + * + * Otherwise, constants go through the CURBEs using the brw_constant_buffer + * state atom. + */ +static drm_intel_bo * +brw_wm_update_constant_buffer(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + struct brw_fragment_program *fp = + (struct brw_fragment_program *) brw->fragment_program; + const struct gl_program_parameter_list *params = fp->program.Base.Parameters; + const int size = params->NumParameters * 4 * sizeof(GLfloat); + drm_intel_bo *const_buffer; + + /* BRW_NEW_FRAGMENT_PROGRAM */ + if (!fp->use_const_buffer) + return NULL; + + const_buffer = drm_intel_bo_alloc(intel->bufmgr, "fp_const_buffer", + size, 64); + + /* _NEW_PROGRAM_CONSTANTS */ + dri_bo_subdata(const_buffer, 0, size, params->ParameterValues); + + return const_buffer; +} /** * Update the surface state for a WM constant buffer. * The constant buffer will be (re)allocated here if needed. */ -static dri_bo * +static void brw_update_wm_constant_surface( GLcontext *ctx, - GLuint surf, - dri_bo *const_buffer, - const struct gl_program_parameter_list *params) + GLuint surf) { struct brw_context *brw = brw_context(ctx); struct brw_surface_key key; - struct intel_context *intel = &brw->intel; - const int size = params->NumParameters * 4 * sizeof(GLfloat); + struct brw_fragment_program *fp = + (struct brw_fragment_program *) brw->fragment_program; + const struct gl_program_parameter_list *params = + fp->program.Base.Parameters; - /* free old const buffer if too small */ - if (const_buffer && const_buffer->size < size) { - dri_bo_unreference(const_buffer); - const_buffer = NULL; - } + /* If we're in this state update atom, we need to update WM constants, so + * free the old buffer and create a new one for the new contents. + */ + dri_bo_unreference(fp->const_buffer); + fp->const_buffer = brw_wm_update_constant_buffer(brw); - /* alloc new buffer if needed */ - if (!const_buffer) { - const_buffer = - drm_intel_bo_alloc(intel->bufmgr, "fp_const_buffer", size, 64); + /* If there's no constant buffer, then no surface BO is needed to point at + * it. + */ + if (fp->const_buffer == 0) { + drm_intel_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = NULL; + return; } memset(&key, 0, sizeof(key)); key.format = MESA_FORMAT_RGBA_FLOAT32; key.internal_format = GL_RGBA; - key.bo = const_buffer; + key.bo = fp->const_buffer; key.depthmode = GL_NONE; key.pitch = params->NumParameters; key.width = params->NumParameters; @@ -427,77 +443,59 @@ brw_update_wm_constant_surface( GLcontext *ctx, */ dri_bo_unreference(brw->wm.surf_bo[surf]); - brw->wm.surf_bo[surf] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, + brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, &key, sizeof(key), &key.bo, key.bo ? 1 : 0, NULL); if (brw->wm.surf_bo[surf] == NULL) { brw->wm.surf_bo[surf] = brw_create_constant_surface(brw, &key); } - - return const_buffer; + brw->state.dirty.brw |= BRW_NEW_WM_SURFACES; } - /** - * Update the surface state for a VS constant buffer. - * The constant buffer will be (re)allocated here if needed. + * Updates surface / buffer for fragment shader constant buffer, if + * one is required. + * + * This consumes the state updates for the constant buffer, and produces + * BRW_NEW_WM_SURFACES to get picked up by brw_prepare_wm_surfaces for + * inclusion in the binding table. */ -static dri_bo * -brw_update_vs_constant_surface( GLcontext *ctx, - GLuint surf, - dri_bo *const_buffer, - const struct gl_program_parameter_list *params) +static void prepare_wm_constant_surface(struct brw_context *brw ) { - struct brw_context *brw = brw_context(ctx); - struct brw_surface_key key; - struct intel_context *intel = &brw->intel; - const int size = params->NumParameters * 4 * sizeof(GLfloat); - - assert(surf == 0); - - /* free old const buffer if too small */ - if (const_buffer && const_buffer->size < size) { - dri_bo_unreference(const_buffer); - const_buffer = NULL; - } - - /* alloc new buffer if needed */ - if (!const_buffer) { - const_buffer = - drm_intel_bo_alloc(intel->bufmgr, "vp_const_buffer", size, 64); - } - - memset(&key, 0, sizeof(key)); - - key.format = MESA_FORMAT_RGBA_FLOAT32; - key.internal_format = GL_RGBA; - key.bo = const_buffer; - key.depthmode = GL_NONE; - key.pitch = params->NumParameters; - key.width = params->NumParameters; - key.height = 1; - key.depth = 1; - key.cpp = 16; + GLcontext *ctx = &brw->intel.ctx; + struct brw_fragment_program *fp = + (struct brw_fragment_program *) brw->fragment_program; + GLuint surf = SURF_INDEX_FRAG_CONST_BUFFER; - /* - printf("%s:\n", __FUNCTION__); - printf(" width %d height %d depth %d cpp %d pitch %d\n", - key.width, key.height, key.depth, key.cpp, key.pitch); - */ + drm_intel_bo_unreference(fp->const_buffer); + fp->const_buffer = brw_wm_update_constant_buffer(brw); - dri_bo_unreference(brw->vs.surf_bo[surf]); - brw->vs.surf_bo[surf] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), - &key.bo, key.bo ? 1 : 0, - NULL); - if (brw->vs.surf_bo[surf] == NULL) { - brw->vs.surf_bo[surf] = brw_create_constant_surface(brw, &key); + /* If there's no constant buffer, then no surface BO is needed to point at + * it. + */ + if (fp->const_buffer == 0) { + if (brw->wm.surf_bo[surf] != NULL) { + drm_intel_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = NULL; + brw->state.dirty.brw |= BRW_NEW_WM_SURFACES; + } + return; } - return const_buffer; + brw_update_wm_constant_surface(ctx, surf); } +const struct brw_tracked_state brw_wm_constant_surface = { + .dirty = { + .mesa = (_NEW_PROGRAM_CONSTANTS), + .brw = (BRW_NEW_FRAGMENT_PROGRAM), + .cache = 0 + }, + .prepare = prepare_wm_constant_surface, +}; + /** * Sets up a surface state structure to point at the given region. @@ -507,7 +505,7 @@ brw_update_vs_constant_surface( GLcontext *ctx, static void brw_update_renderbuffer_surface(struct brw_context *brw, struct gl_renderbuffer *rb, - unsigned int unit, GLboolean cached) + unsigned int unit) { GLcontext *ctx = &brw->intel.ctx; dri_bo *region_bo = NULL; @@ -567,12 +565,11 @@ brw_update_renderbuffer_surface(struct brw_context *brw, ctx->Color.BlendEnabled); dri_bo_unreference(brw->wm.surf_bo[unit]); - brw->wm.surf_bo[unit] = NULL; - if (cached) - brw->wm.surf_bo[unit] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), - ®ion_bo, 1, - NULL); + brw->wm.surf_bo[unit] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + ®ion_bo, 1, + NULL); if (brw->wm.surf_bo[unit] == NULL) { struct brw_surface_state surf; @@ -598,7 +595,8 @@ brw_update_renderbuffer_surface(struct brw_context *brw, surf.ss0.writedisable_alpha = !key.color_mask[3]; /* Key size will never match key size for textures, so we're safe. */ - brw->wm.surf_bo[unit] = brw_upload_cache(&brw->cache, BRW_SS_SURFACE, + brw->wm.surf_bo[unit] = brw_upload_cache(&brw->surface_cache, + BRW_SS_SURFACE, &key, sizeof(key), ®ion_bo, 1, &surf, sizeof(surf), @@ -630,7 +628,7 @@ brw_wm_get_binding_table(struct brw_context *brw) assert(brw->wm.nr_surfaces <= BRW_WM_MAX_SURF); - bind_bo = brw_search_cache(&brw->cache, BRW_SS_SURF_BIND, + bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, NULL, 0, brw->wm.surf_bo, brw->wm.nr_surfaces, NULL); @@ -646,7 +644,7 @@ brw_wm_get_binding_table(struct brw_context *brw) else data[i] = 0; - bind_bo = brw_upload_cache( &brw->cache, BRW_SS_SURF_BIND, + bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, NULL, 0, brw->wm.surf_bo, brw->wm.nr_surfaces, data, data_size, @@ -682,27 +680,17 @@ static void prepare_wm_surfaces(struct brw_context *brw ) for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { brw_update_renderbuffer_surface(brw, ctx->DrawBuffer->_ColorDrawBuffers[i], - i, - GL_FALSE); + i); } } else { - brw_update_renderbuffer_surface(brw, NULL, 0, GL_TRUE); + brw_update_renderbuffer_surface(brw, NULL, 0); } old_nr_surfaces = brw->wm.nr_surfaces; brw->wm.nr_surfaces = MAX_DRAW_BUFFERS; - /* Update surface / buffer for fragment shader constant buffer */ - { - const GLuint surf = SURF_INDEX_FRAG_CONST_BUFFER; - struct brw_fragment_program *fp = - (struct brw_fragment_program *) brw->fragment_program; - fp->const_buffer = - brw_update_wm_constant_surface(ctx, surf, fp->const_buffer, - fp->program.Base.Parameters); - - brw->wm.nr_surfaces = surf + 1; - } + if (brw->wm.surf_bo[SURF_INDEX_FRAG_CONST_BUFFER] != NULL) + brw->wm.nr_surfaces = SURF_INDEX_FRAG_CONST_BUFFER + 1; /* Update surfaces for textures */ for (i = 0; i < BRW_MAX_TEX_UNIT; i++) { @@ -735,100 +723,16 @@ static void prepare_wm_surfaces(struct brw_context *brw ) brw->state.dirty.brw |= BRW_NEW_NR_WM_SURFACES; } - -/** - * Constructs the binding table for the VS surface state. - */ -static dri_bo * -brw_vs_get_binding_table(struct brw_context *brw) -{ - dri_bo *bind_bo; - - assert(brw->vs.nr_surfaces <= BRW_VS_MAX_SURF); - - bind_bo = brw_search_cache(&brw->cache, BRW_SS_SURF_BIND, - NULL, 0, - brw->vs.surf_bo, brw->vs.nr_surfaces, - NULL); - - if (bind_bo == NULL) { - GLuint data_size = brw->vs.nr_surfaces * sizeof(GLuint); - uint32_t *data = malloc(data_size); - int i; - - for (i = 0; i < brw->vs.nr_surfaces; i++) - if (brw->vs.surf_bo[i]) - data[i] = brw->vs.surf_bo[i]->offset; - else - data[i] = 0; - - bind_bo = brw_upload_cache( &brw->cache, BRW_SS_SURF_BIND, - NULL, 0, - brw->vs.surf_bo, brw->vs.nr_surfaces, - data, data_size, - NULL, NULL); - - /* Emit binding table relocations to surface state */ - for (i = 0; i < BRW_VS_MAX_SURF; i++) { - if (brw->vs.surf_bo[i] != NULL) { - dri_bo_emit_reloc(bind_bo, - I915_GEM_DOMAIN_INSTRUCTION, 0, - 0, - i * sizeof(GLuint), - brw->vs.surf_bo[i]); - } - } - - free(data); - } - - return bind_bo; -} - - -/** - * Vertex shader surfaces. Just constant buffer for now. Could add vertex - * shader textures in the future. - */ -static void prepare_vs_surfaces(struct brw_context *brw ) -{ - GLcontext *ctx = &brw->intel.ctx; - - /* Update surface / buffer for vertex shader constant buffer */ - { - const GLuint surf = SURF_INDEX_VERT_CONST_BUFFER; - struct brw_vertex_program *vp = - (struct brw_vertex_program *) brw->vertex_program; - vp->const_buffer = - brw_update_vs_constant_surface(ctx, surf, vp->const_buffer, - vp->program.Base.Parameters); - - brw->vs.nr_surfaces = 1; - } - - dri_bo_unreference(brw->vs.bind_bo); - brw->vs.bind_bo = brw_vs_get_binding_table(brw); - - if (1) - brw->state.dirty.brw |= BRW_NEW_NR_VS_SURFACES; -} - - -static void -prepare_surfaces(struct brw_context *brw) -{ - prepare_wm_surfaces(brw); - prepare_vs_surfaces(brw); -} - - const struct brw_tracked_state brw_wm_surfaces = { .dirty = { - .mesa = _NEW_COLOR | _NEW_TEXTURE | _NEW_BUFFERS | _NEW_PROGRAM, - .brw = BRW_NEW_CONTEXT, + .mesa = (_NEW_COLOR | + _NEW_TEXTURE | + _NEW_BUFFERS), + .brw = (BRW_NEW_CONTEXT | + BRW_NEW_WM_SURFACES), .cache = 0 }, - .prepare = prepare_surfaces, + .prepare = prepare_wm_surfaces, }; diff --git a/src/mesa/drivers/dri/i965/intel_generatemipmap.c b/src/mesa/drivers/dri/i965/intel_generatemipmap.c new file mode 120000 index 0000000000..4c6b37ada0 --- /dev/null +++ b/src/mesa/drivers/dri/i965/intel_generatemipmap.c @@ -0,0 +1 @@ +../intel/intel_generatemipmap.c
\ No newline at end of file diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c index c849e4869e..23ba3b9ef6 100644 --- a/src/mesa/drivers/dri/intel/intel_buffer_objects.c +++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c @@ -35,6 +35,9 @@ #include "intel_batchbuffer.h" #include "intel_regions.h" +static GLboolean +intel_bufferobj_unmap(GLcontext * ctx, + GLenum target, struct gl_buffer_object *obj); /** Allocates a new dri_bo to store the data for the buffer object. */ static void @@ -100,8 +103,15 @@ intel_bufferobj_free(GLcontext * ctx, struct gl_buffer_object *obj) struct intel_buffer_object *intel_obj = intel_buffer_object(obj); assert(intel_obj); - assert(!obj->Pointer); /* Mesa should have unmapped it */ + /* Buffer objects are automatically unmapped when deleting according + * to the spec, but Mesa doesn't do UnmapBuffer for us at context destroy + * (though it does if you call glDeleteBuffers) + */ + if (obj->Pointer) + intel_bufferobj_unmap(ctx, 0, obj); + + _mesa_free(intel_obj->sys_buffer); if (intel_obj->region) { intel_bufferobj_release_region(intel, intel_obj); } @@ -142,7 +152,23 @@ intel_bufferobj_data(GLcontext * ctx, dri_bo_unreference(intel_obj->buffer); intel_obj->buffer = NULL; } + _mesa_free(intel_obj->sys_buffer); + intel_obj->sys_buffer = NULL; + if (size != 0) { +#ifdef I915 + /* On pre-965, stick VBOs in system memory, as we're always doing swtnl + * with their contents anyway. + */ + if (target == GL_ARRAY_BUFFER || target == GL_ELEMENT_ARRAY_BUFFER) { + intel_obj->sys_buffer = _mesa_malloc(size); + if (intel_obj->sys_buffer != NULL) { + if (data != NULL) + memcpy(intel_obj->sys_buffer, data, size); + return; + } + } +#endif intel_bufferobj_alloc_buffer(intel, intel_obj); if (data != NULL) @@ -172,7 +198,10 @@ intel_bufferobj_subdata(GLcontext * ctx, if (intel_obj->region) intel_bufferobj_cow(intel, intel_obj); - dri_bo_subdata(intel_obj->buffer, offset, size, data); + if (intel_obj->sys_buffer) + memcpy((char *)intel_obj->sys_buffer + offset, data, size); + else + dri_bo_subdata(intel_obj->buffer, offset, size, data); } @@ -205,9 +234,15 @@ intel_bufferobj_map(GLcontext * ctx, struct intel_context *intel = intel_context(ctx); struct intel_buffer_object *intel_obj = intel_buffer_object(obj); GLboolean read_only = (access == GL_READ_ONLY_ARB); + GLboolean write_only = (access == GL_WRITE_ONLY_ARB); assert(intel_obj); + if (intel_obj->sys_buffer) { + obj->Pointer = intel_obj->sys_buffer; + return obj->Pointer; + } + if (intel_obj->region) intel_bufferobj_cow(intel, intel_obj); @@ -216,7 +251,14 @@ intel_bufferobj_map(GLcontext * ctx, return NULL; } - dri_bo_map(intel_obj->buffer, !read_only); + if (write_only && intel->intelScreen->kernel_exec_fencing) { + drm_intel_gem_bo_map_gtt(intel_obj->buffer); + intel_obj->mapped_gtt = GL_TRUE; + } else { + drm_intel_bo_map(intel_obj->buffer, !read_only); + intel_obj->mapped_gtt = GL_FALSE; + } + obj->Pointer = intel_obj->buffer->virtual; return obj->Pointer; } @@ -234,7 +276,11 @@ intel_bufferobj_unmap(GLcontext * ctx, assert(intel_obj); if (intel_obj->buffer != NULL) { assert(obj->Pointer); - dri_bo_unmap(intel_obj->buffer); + if (intel_obj->mapped_gtt) { + drm_intel_gem_bo_unmap_gtt(intel_obj->buffer); + } else { + drm_intel_bo_unmap(intel_obj->buffer); + } obj->Pointer = NULL; } return GL_TRUE; @@ -253,6 +299,18 @@ intel_bufferobj_buffer(struct intel_context *intel, } } + if (intel_obj->buffer == NULL) { + intel_bufferobj_alloc_buffer(intel, intel_obj); + intel_bufferobj_subdata(&intel->ctx, + GL_ARRAY_BUFFER_ARB, + 0, + intel_obj->Base.Size, + intel_obj->sys_buffer, + &intel_obj->Base); + _mesa_free(intel_obj->sys_buffer); + intel_obj->sys_buffer = NULL; + } + return intel_obj->buffer; } diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.h b/src/mesa/drivers/dri/intel/intel_buffer_objects.h index bf6dbd58f2..0431015631 100644 --- a/src/mesa/drivers/dri/intel/intel_buffer_objects.h +++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.h @@ -42,10 +42,13 @@ struct intel_buffer_object { struct gl_buffer_object Base; dri_bo *buffer; /* the low-level buffer manager's buffer handle */ + /** System memory buffer data, if not using a BO to store the data. */ + void *sys_buffer; struct intel_region *region; /* Is there a zero-copy texture associated with this (pixel) buffer object? */ + GLboolean mapped_gtt; }; diff --git a/src/mesa/drivers/dri/intel/intel_buffers.c b/src/mesa/drivers/dri/intel/intel_buffers.c index b86cafea24..df5c3fc176 100644 --- a/src/mesa/drivers/dri/intel/intel_buffers.c +++ b/src/mesa/drivers/dri/intel/intel_buffers.c @@ -157,7 +157,7 @@ intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb) /* Do this here, not core Mesa, since this function is called from * many places within the driver. */ - if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { + if (ctx->NewState & _NEW_BUFFERS) { /* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */ _mesa_update_framebuffer(ctx); /* this updates the DrawBuffer's Width/Height if it's a FBO */ @@ -276,7 +276,7 @@ intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb) ctx->Driver.Enable(ctx, GL_DEPTH_TEST, (ctx->Depth.Test && fb->Visual.depthBits > 0)); ctx->Driver.Enable(ctx, GL_STENCIL_TEST, - (ctx->Stencil._Enabled && fb->Visual.stencilBits > 0)); + (ctx->Stencil.Enabled && fb->Visual.stencilBits > 0)); } else { /* Mesa's Stencil._Enabled field is updated when diff --git a/src/mesa/drivers/dri/intel/intel_clear.c b/src/mesa/drivers/dri/intel/intel_clear.c index eb0d890f47..309ac1923b 100644 --- a/src/mesa/drivers/dri/intel/intel_clear.c +++ b/src/mesa/drivers/dri/intel/intel_clear.c @@ -53,6 +53,7 @@ #include "intel_clear.h" #include "intel_fbo.h" #include "intel_pixel.h" +#include "intel_regions.h" #define FILE_DEBUG_FLAG DEBUG_BLIT @@ -255,8 +256,10 @@ intel_clear_tris(GLcontext *ctx, GLbitfield mask) /* Control writing of the stencil clear value to stencil. */ if (this_mask & BUFFER_BIT_STENCIL) { _mesa_Enable(GL_STENCIL_TEST); - _mesa_StencilOp(GL_REPLACE, GL_REPLACE, GL_REPLACE); - _mesa_StencilFuncSeparate(GL_FRONT, GL_ALWAYS, ctx->Stencil.Clear, + _mesa_StencilOpSeparate(GL_FRONT_AND_BACK, + GL_REPLACE, GL_REPLACE, GL_REPLACE); + _mesa_StencilFuncSeparate(GL_FRONT_AND_BACK, GL_ALWAYS, + ctx->Stencil.Clear, ctx->Stencil.WriteMask[0]); } else { _mesa_Disable(GL_STENCIL_TEST); @@ -310,7 +313,6 @@ static const char *buffer_names[] = { static void intelClear(GLcontext *ctx, GLbitfield mask) { - struct intel_context *intel = intel_context(ctx); const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask); GLbitfield tri_mask = 0; GLbitfield blit_mask = 0; @@ -338,7 +340,7 @@ intelClear(GLcontext *ctx, GLbitfield mask) = intel_get_rb_region(fb, BUFFER_STENCIL); if (stencilRegion) { /* have hw stencil */ - if (IS_965(intel->intelScreen->deviceID) || + if (stencilRegion->tiling == I915_TILING_Y || (ctx->Stencil.WriteMask[0] & 0xff) != 0xff) { /* We have to use the 3D engine if we're clearing a partial mask * of the stencil buffer, or if we're on a 965 which has a tiled @@ -355,9 +357,10 @@ intelClear(GLcontext *ctx, GLbitfield mask) /* HW depth */ if (mask & BUFFER_BIT_DEPTH) { + const struct intel_region *irb = intel_get_rb_region(fb, BUFFER_DEPTH); + /* clear depth with whatever method is used for stencil (see above) */ - if (IS_965(intel->intelScreen->deviceID) || - tri_mask & BUFFER_BIT_STENCIL) + if (irb->tiling == I915_TILING_Y || tri_mask & BUFFER_BIT_STENCIL) tri_mask |= BUFFER_BIT_DEPTH; else blit_mask |= BUFFER_BIT_DEPTH; diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c index 5dc3df395d..f88b37d0f3 100644 --- a/src/mesa/drivers/dri/intel/intel_context.c +++ b/src/mesa/drivers/dri/intel/intel_context.c @@ -394,7 +394,7 @@ intel_viewport(GLcontext *ctx, GLint x, GLint y, GLsizei w, GLsizei h) if (!driContext->driScreenPriv->dri2.enabled) return; - if (!intel->internal_viewport_call) { + if (!intel->internal_viewport_call && ctx->DrawBuffer->Name == 0) { intel_update_renderbuffers(driContext, driContext->driDrawablePriv); if (driContext->driDrawablePriv != driContext->driReadablePriv) intel_update_renderbuffers(driContext, driContext->driReadablePriv); @@ -550,6 +550,9 @@ intelInitDriverFunctions(struct dd_function_table *functions) functions->CopyConvolutionFilter2D = _swrast_CopyConvolutionFilter2D; intelInitTextureFuncs(functions); + intelInitTextureImageFuncs(functions); + intelInitTextureSubImageFuncs(functions); + intelInitTextureCopyImageFuncs(functions); intelInitStateFuncs(functions); intelInitClearFuncs(functions); intelInitBufferFuncs(functions); @@ -650,6 +653,13 @@ intelInitContext(struct intel_context *intel, _mesa_init_point(ctx); ctx->Const.MaxColorAttachments = 4; /* XXX FBO: review this */ + if (IS_965(intelScreen->deviceID)) { + if (MAX_WIDTH > 8192) + ctx->Const.MaxRenderbufferSize = 8192; + } else { + if (MAX_WIDTH > 2048) + ctx->Const.MaxRenderbufferSize = 2048; + } /* Initialize the software rasterizer and helper modules. */ _swrast_CreateContext(ctx); @@ -717,6 +727,9 @@ intelInitContext(struct intel_context *intel, else if (driQueryOptionb(&intel->optionCache, "force_s3tc_enable")) { _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc"); } + intel->use_texture_tiling = driQueryOptionb(&intel->optionCache, + "texture_tiling"); + intel->use_early_z = driQueryOptionb(&intel->optionCache, "early_z"); intel->prim.primitive = ~0; @@ -778,13 +791,64 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) intel->prim.vb_bo = NULL; if (release_texture_heaps) { - /* This share group is about to go away, free our private - * texture object data. + /* Nothing is currently done here to free texture heaps; + * but we're not using the texture heap utilities, so I + * rather think we shouldn't. I've taken a look, and can't + * find any private texture data hanging around anywhere, but + * I'm not yet certain there isn't any at all... */ - if (INTEL_DEBUG & DEBUG_TEXTURE) + /* if (INTEL_DEBUG & DEBUG_TEXTURE) fprintf(stderr, "do something to free texture heaps\n"); + */ } + /* XXX In intelMakeCurrent() below, the context's static regions are + * referenced inside the frame buffer; it's listed as a hack, + * with a comment of "XXX FBO temporary fix-ups!", but + * as long as it's there, we should release the regions here. + * The do/while loop around the block is used to allow the + * "continue" statements inside the block to exit the block, + * to avoid many layers of "if" constructs. + */ + do { + __DRIdrawablePrivate * driDrawPriv = intel->driDrawable; + struct intel_framebuffer *intel_fb; + struct intel_renderbuffer *irbDepth, *irbStencil; + if (!driDrawPriv) { + /* We're already detached from the drawable; exit this block. */ + continue; + } + intel_fb = (struct intel_framebuffer *) driDrawPriv->driverPrivate; + if (!intel_fb) { + /* The frame buffer is already gone; exit this block. */ + continue; + } + irbDepth = intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH); + irbStencil = intel_get_renderbuffer(&intel_fb->Base, BUFFER_STENCIL); + + /* If the regions of the frame buffer still match the regions + * of the context, release them. If they've changed somehow, + * leave them alone. + */ + if (intel_fb->color_rb[0] && intel_fb->color_rb[0]->region == intel->front_region) { + intel_renderbuffer_set_region(intel_fb->color_rb[0], NULL); + } + if (intel_fb->color_rb[1] && intel_fb->color_rb[1]->region == intel->back_region) { + intel_renderbuffer_set_region(intel_fb->color_rb[1], NULL); + } + + if (irbDepth && irbDepth->region == intel->depth_region) { + intel_renderbuffer_set_region(irbDepth, NULL); + } + /* Usually, the stencil buffer is the same as the depth buffer; + * but they're handled separately in MakeCurrent, so we'll + * handle them separately here. + */ + if (irbStencil && irbStencil->region == intel->depth_region) { + intel_renderbuffer_set_region(irbStencil, NULL); + } + } while (0); + intel_region_release(&intel->front_region); intel_region_release(&intel->back_region); intel_region_release(&intel->depth_region); @@ -793,6 +857,8 @@ intelDestroyContext(__DRIcontextPrivate * driContextPriv) /* free the Mesa context */ _mesa_free_context_data(&intel->ctx); + + } } @@ -821,7 +887,10 @@ intelMakeCurrent(__DRIcontextPrivate * driContextPriv, if (driDrawPriv != driReadPriv) intel_update_renderbuffers(driContextPriv, driReadPriv); } else { - /* XXX FBO temporary fix-ups! */ + /* XXX FBO temporary fix-ups! These are released in + * intelDextroyContext(), above. Changes here should be + * reflected there. + */ /* if the renderbuffers don't have regions, init them from the context */ struct intel_renderbuffer *irbDepth = intel_get_renderbuffer(&intel_fb->Base, BUFFER_DEPTH); diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h index f45e24ca3a..7d3c80bb21 100644 --- a/src/mesa/drivers/dri/intel/intel_context.h +++ b/src/mesa/drivers/dri/intel/intel_context.h @@ -161,12 +161,22 @@ struct intel_context struct { struct gl_fragment_program *bitmap_fp; struct gl_vertex_program *passthrough_vp; + struct gl_buffer_object *texcoord_vbo; struct gl_fragment_program *saved_fp; GLboolean saved_fp_enable; struct gl_vertex_program *saved_vp; GLboolean saved_vp_enable; + struct gl_fragment_program *tex2d_fp; + + GLboolean saved_texcoord_enable; + struct gl_buffer_object *saved_array_vbo, *saved_texcoord_vbo; + GLenum saved_texcoord_type; + GLsizei saved_texcoord_size, saved_texcoord_stride; + const void *saved_texcoord_ptr; + int saved_active_texture; + GLint saved_vp_x, saved_vp_y; GLsizei saved_vp_width, saved_vp_height; GLenum saved_matrix_mode; @@ -295,6 +305,9 @@ struct intel_context */ GLboolean is_front_buffer_rendering; + GLboolean use_texture_tiling; + GLboolean use_early_z; + drm_clip_rect_t fboRect; /**< cliprect for FBO rendering */ int perf_boxes; @@ -549,6 +562,9 @@ void intel_viewport(GLcontext * ctx, GLint x, GLint y, void intel_update_renderbuffers(__DRIcontext *context, __DRIdrawable *drawable); +void i915_set_buf_info_for_region(uint32_t *state, struct intel_region *region, + uint32_t buffer_id); + /*====================================================================== * Inline conversion functions. * These are better-typed than the macros used previously: @@ -559,4 +575,10 @@ intel_context(GLcontext * ctx) return (struct intel_context *) ctx; } +static INLINE GLboolean +is_power_of_two(uint32_t value) +{ + return (value & (value - 1)) == 0; +} + #endif diff --git a/src/mesa/drivers/dri/intel/intel_extensions.c b/src/mesa/drivers/dri/intel/intel_extensions.c index 9ec1b4ec2f..1e8b1878ab 100644 --- a/src/mesa/drivers/dri/intel/intel_extensions.c +++ b/src/mesa/drivers/dri/intel/intel_extensions.c @@ -48,6 +48,7 @@ #define need_GL_EXT_point_parameters #define need_GL_EXT_secondary_color #define need_GL_EXT_stencil_two_side +#define need_GL_APPLE_vertex_array_object #define need_GL_ATI_separate_stencil #define need_GL_ATI_envmap_bumpmap #define need_GL_NV_point_sprite @@ -95,6 +96,7 @@ static const struct dri_extension card_extensions[] = { { "GL_EXT_texture_lod_bias", NULL }, { "GL_3DFX_texture_compression_FXT1", NULL }, { "GL_APPLE_client_storage", NULL }, + { "GL_APPLE_vertex_array_object", GL_APPLE_vertex_array_object_functions}, { "GL_MESA_pack_invert", NULL }, { "GL_MESA_ycbcr_texture", NULL }, { "GL_NV_blend_square", NULL }, diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c index 52647ddf8b..0ea413aee1 100644 --- a/src/mesa/drivers/dri/intel/intel_fbo.c +++ b/src/mesa/drivers/dri/intel/intel_fbo.c @@ -217,7 +217,8 @@ intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb, DBG("Allocating %d x %d Intel RBO (pitch %d)\n", width, height, pitch); - irb->region = intel_region_alloc(intel, cpp, width, height, pitch, + irb->region = intel_region_alloc(intel, I915_TILING_NONE, + cpp, width, height, pitch, GL_TRUE); if (!irb->region) return GL_FALSE; /* out of memory? */ @@ -574,9 +575,10 @@ intel_render_texture(GLcontext * ctx, ASSERT(newImage); - if (newImage->Border != 0) { - /* Fallback on drawing to a texture with a border, which won't have a - * miptree. + intel_image = intel_texture_image(newImage); + if (!intel_image->mt) { + /* Fallback on drawing to a texture that doesn't have a miptree + * (has a border, width/height 0, etc.) */ _mesa_reference_renderbuffer(&att->Renderbuffer, NULL); _mesa_render_texture(ctx, fb, att); @@ -607,7 +609,6 @@ intel_render_texture(GLcontext * ctx, irb->Base.RefCount); /* point the renderbufer's region to the texture image region */ - intel_image = intel_texture_image(newImage); if (irb->region != intel_image->mt->region) { if (irb->region) intel_region_release(&irb->region); @@ -679,6 +680,11 @@ intel_validate_framebuffer(GLcontext *ctx, struct gl_framebuffer *fb) if (rb == NULL) continue; + if (irb == NULL) { + fb->_Status = GL_FRAMEBUFFER_UNSUPPORTED_EXT; + continue; + } + switch (irb->texformat->MesaFormat) { case MESA_FORMAT_ARGB8888: case MESA_FORMAT_RGB565: diff --git a/src/mesa/drivers/dri/intel/intel_generatemipmap.c b/src/mesa/drivers/dri/intel/intel_generatemipmap.c new file mode 100644 index 0000000000..1060fbd9e5 --- /dev/null +++ b/src/mesa/drivers/dri/intel/intel_generatemipmap.c @@ -0,0 +1,283 @@ +/* + * Copyright (C) 1999-2007 Brian Paul All Rights Reserved. + * Copyright © 2009 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#include "main/glheader.h" +#include "main/enums.h" +#include "main/image.h" +#include "main/mtypes.h" +#include "main/macros.h" +#include "main/bufferobj.h" +#include "main/teximage.h" +#include "main/texenv.h" +#include "main/texobj.h" +#include "main/texstate.h" +#include "main/texparam.h" +#include "main/varray.h" +#include "main/attrib.h" +#include "main/enable.h" +#include "main/buffers.h" +#include "main/fbobject.h" +#include "main/framebuffer.h" +#include "main/renderbuffer.h" +#include "main/depth.h" +#include "main/hash.h" +#include "main/mipmap.h" +#include "main/blend.h" +#include "glapi/dispatch.h" +#include "swrast/swrast.h" + +#include "intel_screen.h" +#include "intel_context.h" +#include "intel_batchbuffer.h" +#include "intel_pixel.h" +#include "intel_tex.h" +#include "intel_mipmap_tree.h" + +static const char *intel_fp_tex2d = + "!!ARBfp1.0\n" + "TEX result.color, fragment.texcoord[0], texture[0], 2D;\n" + "END\n"; + +static GLboolean +intel_generate_mipmap_level(GLcontext *ctx, GLuint tex_name, + int level, int width, int height) +{ + struct intel_context *intel = intel_context(ctx); + GLfloat vertices[4][2]; + GLint status; + + /* Set to source from the previous level */ + _mesa_TexParameterf(GL_TEXTURE_2D, GL_TEXTURE_BASE_LEVEL, level - 1); + _mesa_TexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, level - 1); + + /* Set to draw into the current level */ + _mesa_FramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, + GL_COLOR_ATTACHMENT0_EXT, + GL_TEXTURE_2D, + tex_name, + level); + /* Choose to render to the color attachment. */ + _mesa_DrawBuffer(GL_COLOR_ATTACHMENT0_EXT); + + status = _mesa_CheckFramebufferStatusEXT (GL_FRAMEBUFFER_EXT); + if (status != GL_FRAMEBUFFER_COMPLETE_EXT) + return GL_FALSE; + + intel_meta_set_passthrough_transform(intel); + + /* XXX: Doing it right would involve setting up the transformation to do + * 0-1 mapping or something, and not changing the vertex data. + */ + vertices[0][0] = 0; + vertices[0][1] = 0; + vertices[1][0] = width; + vertices[1][1] = 0; + vertices[2][0] = width; + vertices[2][1] = height; + vertices[3][0] = 0; + vertices[3][1] = height; + + _mesa_VertexPointer(2, GL_FLOAT, 2 * sizeof(GLfloat), &vertices); + _mesa_Enable(GL_VERTEX_ARRAY); + intel_meta_set_default_texrect(intel); + + CALL_DrawArrays(ctx->Exec, (GL_TRIANGLE_FAN, 0, 4)); + + intel_meta_restore_texcoords(intel); + intel_meta_restore_transform(intel); + + return GL_TRUE; +} + +static GLboolean +intel_generate_mipmap_2d(GLcontext *ctx, + GLenum target, + struct gl_texture_object *texObj) +{ + struct intel_context *intel = intel_context(ctx); + GLint old_active_texture; + int level, max_levels, start_level, end_level; + GLuint fb_name; + GLboolean success = GL_FALSE; + struct gl_framebuffer *saved_fbo = NULL; + + _mesa_PushAttrib(GL_ENABLE_BIT | GL_TEXTURE_BIT | + GL_CURRENT_BIT | GL_COLOR_BUFFER_BIT | + GL_DEPTH_BUFFER_BIT); + _mesa_PushClientAttrib(GL_CLIENT_VERTEX_ARRAY_BIT); + old_active_texture = ctx->Texture.CurrentUnit; + _mesa_reference_framebuffer(&saved_fbo, ctx->DrawBuffer); + + _mesa_Disable(GL_POLYGON_STIPPLE); + _mesa_Disable(GL_DEPTH_TEST); + _mesa_Disable(GL_STENCIL_TEST); + _mesa_ColorMask(GL_TRUE, GL_TRUE, GL_TRUE, GL_TRUE); + _mesa_DepthMask(GL_FALSE); + + /* Bind the given texture to GL_TEXTURE_2D with linear filtering for our + * minification. + */ + _mesa_ActiveTextureARB(GL_TEXTURE0_ARB); + _mesa_Enable(GL_TEXTURE_2D); + _mesa_BindTexture(GL_TEXTURE_2D, texObj->Name); + _mesa_TexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, + GL_LINEAR_MIPMAP_NEAREST); + _mesa_TexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); + + /* Bind the new renderbuffer to the color attachment point. */ + _mesa_GenFramebuffersEXT(1, &fb_name); + _mesa_BindFramebufferEXT(GL_FRAMEBUFFER_EXT, fb_name); + + intel_meta_set_fragment_program(intel, &intel->meta.tex2d_fp, + intel_fp_tex2d); + intel_meta_set_passthrough_vertex_program(intel); + + max_levels = _mesa_max_texture_levels(ctx, texObj->Target); + start_level = texObj->BaseLevel; + end_level = texObj->MaxLevel; + + /* Loop generating level+1 from level. */ + for (level = start_level; level < end_level && level < max_levels - 1; level++) { + const struct gl_texture_image *srcImage; + int width, height; + + srcImage = _mesa_select_tex_image(ctx, texObj, target, level); + if (srcImage->Border != 0) + goto fail; + + width = srcImage->Width / 2; + if (width < 1) + width = 1; + height = srcImage->Height / 2; + if (height < 1) + height = 1; + + if (width == srcImage->Width && + height == srcImage->Height) { + /* Neither _mesa_max_texture_levels nor texObj->MaxLevel are the + * maximum texture level for the object, so break out when we've gone + * over the edge. + */ + break; + } + + /* Make sure that there's space allocated for the target level. + * We could skip this if there's already space allocated and save some + * time. + */ + _mesa_TexImage2D(GL_TEXTURE_2D, level + 1, srcImage->InternalFormat, + width, height, 0, + GL_RGBA, GL_UNSIGNED_INT, NULL); + + if (!intel_generate_mipmap_level(ctx, texObj->Name, level + 1, + width, height)) + goto fail; + } + + success = GL_TRUE; + +fail: + intel_meta_restore_fragment_program(intel); + intel_meta_restore_vertex_program(intel); + + _mesa_DeleteFramebuffersEXT(1, &fb_name); + _mesa_ActiveTextureARB(GL_TEXTURE0_ARB + old_active_texture); + if (saved_fbo) + _mesa_BindFramebufferEXT(GL_FRAMEBUFFER_EXT, saved_fbo->Name); + _mesa_reference_framebuffer(&saved_fbo, NULL); + _mesa_PopClientAttrib(); + _mesa_PopAttrib(); + + return success; +} + + +/** + * Generate new mipmap data from BASE+1 to BASE+p (the minimally-sized mipmap + * level). + * + * The texture object's miptree must be mapped. + * + * It would be really nice if this was just called by Mesa whenever mipmaps + * needed to be regenerated, rather than us having to remember to do so in + * each texture image modification path. + * + * This function should also include an accelerated path. + */ +void +intel_generate_mipmap(GLcontext *ctx, GLenum target, + struct gl_texture_object *texObj) +{ + struct intel_context *intel = intel_context(ctx); + struct intel_texture_object *intelObj = intel_texture_object(texObj); + GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; + int face, i; + + /* HW path */ + if (target == GL_TEXTURE_2D && + ctx->Extensions.EXT_framebuffer_object && + ctx->Extensions.ARB_fragment_program && + ctx->Extensions.ARB_vertex_program) { + GLboolean success; + + /* We'll be accessing this texture using GL entrypoints, which should + * be resilient against other access to this texture. + */ + _mesa_unlock_texture(ctx, texObj); + success = intel_generate_mipmap_2d(ctx, target, texObj); + _mesa_lock_texture(ctx, texObj); + + if (success) + return; + } + + /* SW path */ + intel_tex_map_level_images(intel, intelObj, texObj->BaseLevel); + _mesa_generate_mipmap(ctx, target, texObj); + intel_tex_unmap_level_images(intel, intelObj, texObj->BaseLevel); + + /* Update the level information in our private data in the new images, since + * it didn't get set as part of a normal TexImage path. + */ + for (face = 0; face < nr_faces; face++) { + for (i = texObj->BaseLevel + 1; i < texObj->MaxLevel; i++) { + struct intel_texture_image *intelImage; + + intelImage = intel_texture_image(texObj->Image[face][i]); + if (intelImage == NULL) + break; + + intelImage->level = i; + intelImage->face = face; + /* Unreference the miptree to signal that the new Data is a bare + * pointer from mesa. + */ + intel_miptree_release(intel, &intelImage->mt); + } + } +} diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c index 6e1e034e53..0d34f28311 100644 --- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c +++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c @@ -57,14 +57,16 @@ intel_miptree_create_internal(struct intel_context *intel, GLuint last_level, GLuint width0, GLuint height0, - GLuint depth0, GLuint cpp, GLuint compress_byte) + GLuint depth0, GLuint cpp, GLuint compress_byte, + uint32_t tiling) { GLboolean ok; struct intel_mipmap_tree *mt = calloc(sizeof(*mt), 1); - DBG("%s target %s format %s level %d..%d\n", __FUNCTION__, + DBG("%s target %s format %s level %d..%d <-- %p\n", __FUNCTION__, _mesa_lookup_enum_by_nr(target), - _mesa_lookup_enum_by_nr(internal_format), first_level, last_level); + _mesa_lookup_enum_by_nr(internal_format), + first_level, last_level, mt); mt->target = target_to_target(target); mt->internal_format = internal_format; @@ -80,15 +82,16 @@ intel_miptree_create_internal(struct intel_context *intel, #ifdef I915 if (IS_945(intel->intelScreen->deviceID)) - ok = i945_miptree_layout(intel, mt); + ok = i945_miptree_layout(intel, mt, tiling); else - ok = i915_miptree_layout(intel, mt); + ok = i915_miptree_layout(intel, mt, tiling); #else - ok = brw_miptree_layout(intel, mt); + ok = brw_miptree_layout(intel, mt, tiling); #endif if (!ok) { free(mt); + DBG("%s not okay - returning NULL\n", __FUNCTION__); return NULL; } @@ -107,10 +110,18 @@ intel_miptree_create(struct intel_context *intel, GLboolean expect_accelerated_upload) { struct intel_mipmap_tree *mt; + uint32_t tiling; + + if (intel->use_texture_tiling && compress_byte == 0 && + intel->intelScreen->kernel_exec_fencing) + tiling = I915_TILING_X; + else + tiling = I915_TILING_NONE; mt = intel_miptree_create_internal(intel, target, internal_format, first_level, last_level, width0, - height0, depth0, cpp, compress_byte); + height0, depth0, cpp, compress_byte, + tiling); /* * pitch == 0 || height == 0 indicates the null texture */ @@ -118,6 +129,7 @@ intel_miptree_create(struct intel_context *intel, return NULL; mt->region = intel_region_alloc(intel, + tiling, mt->cpp, mt->pitch, mt->total_height, @@ -147,7 +159,8 @@ intel_miptree_create_for_region(struct intel_context *intel, mt = intel_miptree_create_internal(intel, target, internal_format, first_level, last_level, region->width, region->height, 1, - region->cpp, compress_byte); + region->cpp, compress_byte, + I915_TILING_NONE); if (!mt) return mt; #if 0 @@ -185,6 +198,7 @@ intel_miptree_create_for_region(struct intel_context *intel, int intel_miptree_pitch_align (struct intel_context *intel, struct intel_mipmap_tree *mt, + uint32_t tiling, int pitch) { #ifdef I915 @@ -205,6 +219,11 @@ int intel_miptree_pitch_align (struct intel_context *intel, pitch_align = 4; } + if (tiling == I915_TILING_X) + pitch_align = 512; + else if (tiling == I915_TILING_Y) + pitch_align = 128; + pitch = ALIGN(pitch * mt->cpp, pitch_align); #ifdef I915 diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h index 4060b9df78..3af9966827 100644 --- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h +++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h @@ -148,6 +148,7 @@ intel_miptree_create_for_region(struct intel_context *intel, int intel_miptree_pitch_align (struct intel_context *intel, struct intel_mipmap_tree *mt, + uint32_t tiling, int pitch); void intel_miptree_reference(struct intel_mipmap_tree **dst, @@ -218,10 +219,13 @@ void intel_miptree_image_copy(struct intel_context *intel, /* i915_mipmap_tree.c: */ GLboolean i915_miptree_layout(struct intel_context *intel, - struct intel_mipmap_tree *mt); + struct intel_mipmap_tree *mt, + uint32_t tiling); GLboolean i945_miptree_layout(struct intel_context *intel, - struct intel_mipmap_tree *mt); + struct intel_mipmap_tree *mt, + uint32_t tiling); GLboolean brw_miptree_layout(struct intel_context *intel, - struct intel_mipmap_tree *mt); + struct intel_mipmap_tree *mt, + uint32_t tiling); #endif diff --git a/src/mesa/drivers/dri/intel/intel_pixel.c b/src/mesa/drivers/dri/intel/intel_pixel.c index fc0ac0b79c..36a684b3b8 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel.c +++ b/src/mesa/drivers/dri/intel/intel_pixel.c @@ -27,9 +27,12 @@ #include "main/enums.h" #include "main/state.h" +#include "main/bufferobj.h" #include "main/context.h" #include "main/enable.h" #include "main/matrix.h" +#include "main/texstate.h" +#include "main/varray.h" #include "main/viewport.h" #include "swrast/swrast.h" #include "shader/arbprogram.h" @@ -334,6 +337,85 @@ intel_meta_restore_fragment_program(struct intel_context *intel) _mesa_Disable(GL_FRAGMENT_PROGRAM_ARB); } +static const float default_texcoords[4][2] = { { 0.0, 0.0 }, + { 1.0, 0.0 }, + { 1.0, 1.0 }, + { 0.0, 1.0 } }; + +void +intel_meta_set_default_texrect(struct intel_context *intel) +{ + GLcontext *ctx = &intel->ctx; + struct gl_client_array *old_texcoord_array; + + intel->meta.saved_active_texture = ctx->Texture.CurrentUnit; + if (intel->meta.saved_array_vbo == NULL) { + _mesa_reference_buffer_object(ctx, &intel->meta.saved_array_vbo, + ctx->Array.ArrayBufferObj); + } + + old_texcoord_array = &ctx->Array.ArrayObj->TexCoord[0]; + intel->meta.saved_texcoord_type = old_texcoord_array->Type; + intel->meta.saved_texcoord_size = old_texcoord_array->Size; + intel->meta.saved_texcoord_stride = old_texcoord_array->Stride; + intel->meta.saved_texcoord_enable = old_texcoord_array->Enabled; + intel->meta.saved_texcoord_ptr = old_texcoord_array->Ptr; + _mesa_reference_buffer_object(ctx, &intel->meta.saved_texcoord_vbo, + old_texcoord_array->BufferObj); + + _mesa_ClientActiveTextureARB(GL_TEXTURE0); + + if (intel->meta.texcoord_vbo == NULL) { + GLuint vbo_name; + + _mesa_GenBuffersARB(1, &vbo_name); + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, vbo_name); + _mesa_BufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(default_texcoords), + default_texcoords, GL_STATIC_DRAW_ARB); + _mesa_reference_buffer_object(ctx, &intel->meta.texcoord_vbo, + ctx->Array.ArrayBufferObj); + } else { + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, + intel->meta.texcoord_vbo->Name); + } + _mesa_TexCoordPointer(2, GL_FLOAT, 2 * sizeof(GLfloat), NULL); + + _mesa_Enable(GL_TEXTURE_COORD_ARRAY); +} + +void +intel_meta_restore_texcoords(struct intel_context *intel) +{ + GLcontext *ctx = &intel->ctx; + + /* Restore the old TexCoordPointer */ + if (intel->meta.saved_texcoord_vbo) { + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, + intel->meta.saved_texcoord_vbo->Name); + _mesa_reference_buffer_object(ctx, &intel->meta.saved_texcoord_vbo, NULL); + } else { + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, 0); + } + + _mesa_TexCoordPointer(intel->meta.saved_texcoord_size, + intel->meta.saved_texcoord_type, + intel->meta.saved_texcoord_stride, + intel->meta.saved_texcoord_ptr); + if (!intel->meta.saved_texcoord_enable) + _mesa_Disable(GL_TEXTURE_COORD_ARRAY); + + _mesa_ClientActiveTextureARB(GL_TEXTURE0 + + intel->meta.saved_active_texture); + + if (intel->meta.saved_array_vbo) { + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, + intel->meta.saved_array_vbo->Name); + _mesa_reference_buffer_object(ctx, &intel->meta.saved_array_vbo, NULL); + } else { + _mesa_BindBufferARB(GL_ARRAY_BUFFER_ARB, 0); + } +} + void intelInitPixelFuncs(struct dd_function_table *functions) { @@ -355,5 +437,7 @@ intel_free_pixel_state(struct intel_context *intel) _mesa_reference_vertprog(ctx, &intel->meta.passthrough_vp, NULL); _mesa_reference_fragprog(ctx, &intel->meta.bitmap_fp, NULL); + _mesa_reference_fragprog(ctx, &intel->meta.tex2d_fp, NULL); + _mesa_reference_buffer_object(ctx, &intel->meta.texcoord_vbo, NULL); } diff --git a/src/mesa/drivers/dri/intel/intel_pixel.h b/src/mesa/drivers/dri/intel/intel_pixel.h index cb41fa182c..6acf0813c8 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel.h +++ b/src/mesa/drivers/dri/intel/intel_pixel.h @@ -40,6 +40,9 @@ void intel_meta_set_fragment_program(struct intel_context *intel, const char *prog_string); void intel_meta_restore_fragment_program(struct intel_context *intel); void intel_free_pixel_state(struct intel_context *intel); +void intel_meta_set_default_texrect(struct intel_context *intel); +void intel_meta_set_default_texrect(struct intel_context *intel); +void intel_meta_restore_texcoords(struct intel_context *intel); GLboolean intel_check_blit_fragment_ops(GLcontext * ctx, GLboolean src_alpha_is_one); diff --git a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c index 1db7f5594e..80d3239189 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c +++ b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c @@ -360,7 +360,6 @@ intel_texture_bitmap(GLcontext * ctx, "END\n"; GLuint texname; GLfloat vertices[4][4]; - GLfloat texcoords[4][2]; GLint old_active_texture; GLubyte *unpacked_bitmap; GLubyte *a8_bitmap; @@ -401,6 +400,14 @@ intel_texture_bitmap(GLcontext * ctx, return GL_FALSE; } + if (!ctx->Extensions.ARB_texture_non_power_of_two && + (!is_power_of_two(width) || !is_power_of_two(height))) { + if (INTEL_DEBUG & DEBUG_FALLBACKS) + fprintf(stderr, + "glBitmap() fallback: NPOT texture\n"); + return GL_FALSE; + } + /* Check that we can load in a texture this big. */ if (width > (1 << (ctx->Const.MaxTextureLevels - 1)) || height > (1 << (ctx->Const.MaxTextureLevels - 1))) { @@ -485,22 +492,12 @@ intel_texture_bitmap(GLcontext * ctx, vertices[3][2] = dst_z; vertices[3][3] = 1.0; - texcoords[0][0] = 0.0; - texcoords[0][1] = 0.0; - texcoords[1][0] = 1.0; - texcoords[1][1] = 0.0; - texcoords[2][0] = 1.0; - texcoords[2][1] = 1.0; - texcoords[3][0] = 0.0; - texcoords[3][1] = 1.0; - _mesa_VertexPointer(4, GL_FLOAT, 4 * sizeof(GLfloat), &vertices); - _mesa_ClientActiveTextureARB(GL_TEXTURE0); - _mesa_TexCoordPointer(2, GL_FLOAT, 2 * sizeof(GLfloat), &texcoords); _mesa_Enable(GL_VERTEX_ARRAY); - _mesa_Enable(GL_TEXTURE_COORD_ARRAY); + intel_meta_set_default_texrect(intel); CALL_DrawArrays(ctx->Exec, (GL_TRIANGLE_FAN, 0, 4)); + intel_meta_restore_texcoords(intel); intel_meta_restore_transform(intel); intel_meta_restore_fragment_program(intel); intel_meta_restore_vertex_program(intel); diff --git a/src/mesa/drivers/dri/intel/intel_pixel_draw.c b/src/mesa/drivers/dri/intel/intel_pixel_draw.c index e8d5ac8569..7cda6adb32 100644 --- a/src/mesa/drivers/dri/intel/intel_pixel_draw.c +++ b/src/mesa/drivers/dri/intel/intel_pixel_draw.c @@ -70,7 +70,6 @@ intel_texture_drawpixels(GLcontext * ctx, struct intel_context *intel = intel_context(ctx); GLuint texname; GLfloat vertices[4][4]; - GLfloat texcoords[4][2]; GLfloat z; GLint old_active_texture; GLenum internalFormat; @@ -120,6 +119,14 @@ intel_texture_drawpixels(GLcontext * ctx, return GL_FALSE; } + if (!ctx->Extensions.ARB_texture_non_power_of_two && + (!is_power_of_two(width) || !is_power_of_two(height))) { + if (INTEL_DEBUG & DEBUG_FALLBACKS) + fprintf(stderr, + "glDrawPixels() fallback: NPOT texture\n"); + return GL_FALSE; + } + _mesa_PushAttrib(GL_ENABLE_BIT | GL_TEXTURE_BIT | GL_CURRENT_BIT); _mesa_PushClientAttrib(GL_CLIENT_VERTEX_ARRAY_BIT); @@ -169,22 +176,13 @@ intel_texture_drawpixels(GLcontext * ctx, vertices[3][2] = z; vertices[3][3] = 1.0; - texcoords[0][0] = 0.0; - texcoords[0][1] = 0.0; - texcoords[1][0] = 1.0; - texcoords[1][1] = 0.0; - texcoords[2][0] = 1.0; - texcoords[2][1] = 1.0; - texcoords[3][0] = 0.0; - texcoords[3][1] = 1.0; - _mesa_VertexPointer(4, GL_FLOAT, 4 * sizeof(GLfloat), &vertices); - _mesa_ClientActiveTextureARB(GL_TEXTURE0); - _mesa_TexCoordPointer(2, GL_FLOAT, 2 * sizeof(GLfloat), &texcoords); _mesa_Enable(GL_VERTEX_ARRAY); - _mesa_Enable(GL_TEXTURE_COORD_ARRAY); + intel_meta_set_default_texrect(intel); + CALL_DrawArrays(ctx->Exec, (GL_TRIANGLE_FAN, 0, 4)); + intel_meta_restore_texcoords(intel); intel_meta_restore_transform(intel); _mesa_ActiveTextureARB(GL_TEXTURE0_ARB + old_active_texture); @@ -208,7 +206,6 @@ intel_stencil_drawpixels(GLcontext * ctx, struct intel_context *intel = intel_context(ctx); GLuint texname, rb_name, fb_name, old_fb_name; GLfloat vertices[4][2]; - GLfloat texcoords[4][2]; struct intel_renderbuffer *irb; struct intel_renderbuffer *depth_irb; struct gl_renderbuffer *rb; @@ -273,6 +270,14 @@ intel_stencil_drawpixels(GLcontext * ctx, return GL_FALSE; } + if (!ctx->Extensions.ARB_texture_non_power_of_two && + (!is_power_of_two(width) || !is_power_of_two(height))) { + if (INTEL_DEBUG & DEBUG_FALLBACKS) + fprintf(stderr, + "glDrawPixels(GL_STENCIL_INDEX) fallback: NPOT texture\n"); + return GL_FALSE; + } + _mesa_PushAttrib(GL_ENABLE_BIT | GL_TEXTURE_BIT | GL_CURRENT_BIT | GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT); _mesa_PushClientAttrib(GL_CLIENT_VERTEX_ARRAY_BIT); @@ -343,7 +348,6 @@ intel_stencil_drawpixels(GLcontext * ctx, _mesa_free(stencil_pixels); intel_meta_set_passthrough_transform(intel); - vertices[0][0] = x; vertices[0][1] = y; vertices[1][0] = x + width * ctx->Pixel.ZoomX; @@ -353,22 +357,13 @@ intel_stencil_drawpixels(GLcontext * ctx, vertices[3][0] = x; vertices[3][1] = y + height * ctx->Pixel.ZoomY; - texcoords[0][0] = 0.0; - texcoords[0][1] = 0.0; - texcoords[1][0] = 1.0; - texcoords[1][1] = 0.0; - texcoords[2][0] = 1.0; - texcoords[2][1] = 1.0; - texcoords[3][0] = 0.0; - texcoords[3][1] = 1.0; - _mesa_VertexPointer(2, GL_FLOAT, 2 * sizeof(GLfloat), &vertices); - _mesa_ClientActiveTextureARB(GL_TEXTURE0); - _mesa_TexCoordPointer(2, GL_FLOAT, 2 * sizeof(GLfloat), &texcoords); _mesa_Enable(GL_VERTEX_ARRAY); - _mesa_Enable(GL_TEXTURE_COORD_ARRAY); + intel_meta_set_default_texrect(intel); + CALL_DrawArrays(ctx->Exec, (GL_TRIANGLE_FAN, 0, 4)); + intel_meta_restore_texcoords(intel); intel_meta_restore_transform(intel); _mesa_ActiveTextureARB(GL_TEXTURE0_ARB + old_active_texture); diff --git a/src/mesa/drivers/dri/intel/intel_reg.h b/src/mesa/drivers/dri/intel/intel_reg.h index 57ac8f0cc1..d19f1bae34 100644 --- a/src/mesa/drivers/dri/intel/intel_reg.h +++ b/src/mesa/drivers/dri/intel/intel_reg.h @@ -189,6 +189,19 @@ #define S7_DEPTH_OFFSET_CONST_MASK ~0 +/* p143 */ +#define _3DSTATE_BUF_INFO_CMD (CMD_3D | (0x1d<<24) | (0x8e<<16) | 1) +/* Dword 1 */ +#define BUF_3D_ID_COLOR_BACK (0x3<<24) +#define BUF_3D_ID_DEPTH (0x7<<24) +#define BUF_3D_USE_FENCE (1<<23) +#define BUF_3D_TILED_SURFACE (1<<22) +#define BUF_3D_TILE_WALK_X 0 +#define BUF_3D_TILE_WALK_Y (1<<21) +#define BUF_3D_PITCH(x) (((x)/4)<<2) +/* Dword 2 */ +#define BUF_3D_ADDR(x) ((x) & ~0x3) + /* Primitive dispatch on 830-945 */ #define _3DPRIMITIVE (CMD_3D | (0x1f << 24)) #define PRIM_INDIRECT (1<<23) diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c index 0aa5b8c02c..49bcb3c1dd 100644 --- a/src/mesa/drivers/dri/intel/intel_regions.c +++ b/src/mesa/drivers/dri/intel/intel_regions.c @@ -52,17 +52,74 @@ #define FILE_DEBUG_FLAG DEBUG_REGION +/* This should be set to the maximum backtrace size desired. + * Set it to 0 to disable backtrace debugging. + */ +#define DEBUG_BACKTRACE_SIZE 0 + +#if DEBUG_BACKTRACE_SIZE == 0 +/* Use the standard debug output */ +#define _DBG(...) DBG(__VA_ARGS__) +#else +/* Use backtracing debug output */ +#define _DBG(...) {debug_backtrace(); DBG(__VA_ARGS__);} + +/* Backtracing debug support */ +#include <execinfo.h> + +static void +debug_backtrace(void) +{ + void *trace[DEBUG_BACKTRACE_SIZE]; + char **strings = NULL; + int traceSize; + register int i; + + traceSize = backtrace(trace, DEBUG_BACKTRACE_SIZE); + strings = backtrace_symbols(trace, traceSize); + if (strings == NULL) { + DBG("no backtrace:"); + return; + } + + /* Spit out all the strings with a colon separator. Ignore + * the first, since we don't really care about the call + * to debug_backtrace() itself. Skip until the final "/" in + * the trace to avoid really long lines. + */ + for (i = 1; i < traceSize; i++) { + char *p = strings[i], *slash = strings[i]; + while (*p) { + if (*p++ == '/') { + slash = p; + } + } + + DBG("%s:", slash); + } + + /* Free up the memory, and we're done */ + free(strings); +} + +#endif + + + /* XXX: Thread safety? */ GLubyte * intel_region_map(struct intel_context *intel, struct intel_region *region) { - DBG("%s\n", __FUNCTION__); + _DBG("%s %p\n", __FUNCTION__, region); if (!region->map_refcount++) { if (region->pbo) intel_region_cow(intel, region); - dri_bo_map(region->buffer, GL_TRUE); + if (intel->intelScreen->kernel_exec_fencing) + drm_intel_gem_bo_map_gtt(region->buffer); + else + dri_bo_map(region->buffer, GL_TRUE); region->map = region->buffer->virtual; } @@ -72,9 +129,12 @@ intel_region_map(struct intel_context *intel, struct intel_region *region) void intel_region_unmap(struct intel_context *intel, struct intel_region *region) { - DBG("%s\n", __FUNCTION__); + _DBG("%s %p\n", __FUNCTION__, region); if (!--region->map_refcount) { - dri_bo_unmap(region->buffer); + if (intel->intelScreen->kernel_exec_fencing) + drm_intel_gem_bo_unmap_gtt(region->buffer); + else + dri_bo_unmap(region->buffer); region->map = NULL; } } @@ -87,10 +147,10 @@ intel_region_alloc_internal(struct intel_context *intel, { struct intel_region *region; - DBG("%s\n", __FUNCTION__); - - if (buffer == NULL) + if (buffer == NULL) { + _DBG("%s <-- NULL\n", __FUNCTION__); return NULL; + } region = calloc(sizeof(*region), 1); region->cpp = cpp; @@ -104,15 +164,18 @@ intel_region_alloc_internal(struct intel_context *intel, region->tiling = I915_TILING_NONE; region->bit_6_swizzle = I915_BIT_6_SWIZZLE_NONE; + _DBG("%s <-- %p\n", __FUNCTION__, region); return region; } struct intel_region * intel_region_alloc(struct intel_context *intel, + uint32_t tiling, GLuint cpp, GLuint width, GLuint height, GLuint pitch, GLboolean expect_accelerated_upload) { dri_bo *buffer; + struct intel_region *region; if (expect_accelerated_upload) { buffer = drm_intel_bo_alloc_for_render(intel->bufmgr, "region", @@ -122,7 +185,16 @@ intel_region_alloc(struct intel_context *intel, pitch * cpp * height, 64); } - return intel_region_alloc_internal(intel, cpp, width, height, pitch, buffer); + region = intel_region_alloc_internal(intel, cpp, width, height, + pitch, buffer); + + if (tiling != I915_TILING_NONE) { + assert(((pitch * cpp) & 511) == 0); + drm_intel_bo_set_tiling(buffer, &tiling, pitch * cpp); + drm_intel_bo_get_tiling(buffer, ®ion->tiling, ®ion->bit_6_swizzle); + } + + return region; } struct intel_region * @@ -158,7 +230,7 @@ void intel_region_reference(struct intel_region **dst, struct intel_region *src) { if (src) - DBG("%s %p %d\n", __FUNCTION__, src, src->refcount); + _DBG("%s %p %d\n", __FUNCTION__, src, src->refcount); assert(*dst == NULL); if (src) { @@ -172,10 +244,12 @@ intel_region_release(struct intel_region **region_handle) { struct intel_region *region = *region_handle; - if (region == NULL) + if (region == NULL) { + _DBG("%s NULL\n", __FUNCTION__); return; + } - DBG("%s %p %d\n", __FUNCTION__, region, region->refcount - 1); + _DBG("%s %p %d\n", __FUNCTION__, region, region->refcount - 1); ASSERT(region->refcount > 0); region->refcount--; @@ -251,7 +325,7 @@ intel_region_data(struct intel_context *intel, { GLboolean locked = GL_FALSE; - DBG("%s\n", __FUNCTION__); + _DBG("%s\n", __FUNCTION__); if (intel == NULL) return; @@ -293,7 +367,7 @@ intel_region_copy(struct intel_context *intel, GLuint src_offset, GLuint srcx, GLuint srcy, GLuint width, GLuint height) { - DBG("%s\n", __FUNCTION__); + _DBG("%s\n", __FUNCTION__); if (intel == NULL) return; @@ -326,7 +400,7 @@ intel_region_fill(struct intel_context *intel, GLuint dstx, GLuint dsty, GLuint width, GLuint height, GLuint color) { - DBG("%s\n", __FUNCTION__); + _DBG("%s\n", __FUNCTION__); if (intel == NULL) return; @@ -356,6 +430,8 @@ intel_region_attach_pbo(struct intel_context *intel, if (region->pbo == pbo) return; + _DBG("%s %p %p\n", __FUNCTION__, region, pbo); + /* If there is already a pbo attached, break the cow tie now. * Don't call intel_region_release_pbo() as that would * unnecessarily allocate a new buffer we would have to immediately @@ -385,6 +461,7 @@ void intel_region_release_pbo(struct intel_context *intel, struct intel_region *region) { + _DBG("%s %p\n", __FUNCTION__, region); assert(region->buffer == region->pbo->buffer); region->pbo->region = NULL; region->pbo = NULL; @@ -412,7 +489,7 @@ intel_region_cow(struct intel_context *intel, struct intel_region *region) assert(region->cpp * region->pitch * region->height == pbo->Base.Size); - DBG("%s (%d bytes)\n", __FUNCTION__, pbo->Base.Size); + _DBG("%s %p (%d bytes)\n", __FUNCTION__, region, pbo->Base.Size); /* Now blit from the texture buffer to the new buffer: */ @@ -459,6 +536,10 @@ intel_recreate_static(struct intel_context *intel, if (region == NULL) { region = calloc(sizeof(*region), 1); region->refcount = 1; + _DBG("%s creating new region %p\n", __FUNCTION__, region); + } + else { + _DBG("%s %p\n", __FUNCTION__, region); } if (intel->ctx.Visual.rgbBits == 24) diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h index 45e2bf4e77..bd3c8e7325 100644 --- a/src/mesa/drivers/dri/intel/intel_regions.h +++ b/src/mesa/drivers/dri/intel/intel_regions.h @@ -73,7 +73,8 @@ struct intel_region * copied by calling intel_reference_region(). */ struct intel_region *intel_region_alloc(struct intel_context *intel, - GLuint cpp, GLuint width, + uint32_t tiling, + GLuint cpp, GLuint width, GLuint height, GLuint pitch, GLboolean expect_accelerated_upload); diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c index 65e62947ef..8da96ede64 100644 --- a/src/mesa/drivers/dri/intel/intel_screen.c +++ b/src/mesa/drivers/dri/intel/intel_screen.c @@ -49,6 +49,10 @@ #include "i915_drm.h" #include "i830_dri.h" +#define DRI_CONF_TEXTURE_TILING(def) \ + DRI_CONF_OPT_BEGIN(texture_tiling, bool, def) \ + DRI_CONF_DESC(en, "Enable texture tiling") \ + DRI_CONF_OPT_END \ PUBLIC const char __driConfigOptions[] = DRI_CONF_BEGIN @@ -64,6 +68,17 @@ PUBLIC const char __driConfigOptions[] = DRI_CONF_ENUM(1, "Enable reuse of all sizes of buffer objects") DRI_CONF_DESC_END DRI_CONF_OPT_END + +#ifdef I915 + DRI_CONF_TEXTURE_TILING(false) +#else + DRI_CONF_TEXTURE_TILING(true) +#endif + + DRI_CONF_OPT_BEGIN(early_z, bool, false) + DRI_CONF_DESC(en, "Enable early Z in classic mode (unstable, 945-only).") + DRI_CONF_OPT_END + DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY DRI_CONF_FORCE_S3TC_ENABLE(false) @@ -76,7 +91,7 @@ PUBLIC const char __driConfigOptions[] = DRI_CONF_SECTION_END DRI_CONF_END; -const GLuint __driNConfigOptions = 8; +const GLuint __driNConfigOptions = 10; #ifdef USE_NEW_INTERFACE static PFNGLXCREATECONTEXTMODES create_context_modes = NULL; @@ -236,7 +251,7 @@ intel_get_param(__DRIscreenPrivate *psp, int param, int *value) ret = drmCommandWriteRead(psp->fd, DRM_I915_GETPARAM, &gp, sizeof(gp)); if (ret) { - fprintf(stderr, "drm_i915_getparam: %d\n", ret); + _mesa_warning(NULL, "drm_i915_getparam: %d", ret); return GL_FALSE; } diff --git a/src/mesa/drivers/dri/intel/intel_tex.c b/src/mesa/drivers/dri/intel/intel_tex.c index ae0994b183..df63f29a42 100644 --- a/src/mesa/drivers/dri/intel/intel_tex.c +++ b/src/mesa/drivers/dri/intel/intel_tex.c @@ -158,81 +158,11 @@ timed_memcpy(void *dest, const void *src, size_t n) } #endif /* DO_DEBUG */ -/** - * Generate new mipmap data from BASE+1 to BASE+p (the minimally-sized mipmap - * level). - * - * The texture object's miptree must be mapped. - * - * It would be really nice if this was just called by Mesa whenever mipmaps - * needed to be regenerated, rather than us having to remember to do so in - * each texture image modification path. - * - * This function should also include an accelerated path. - */ -void -intel_generate_mipmap(GLcontext *ctx, GLenum target, - struct gl_texture_object *texObj) -{ - struct intel_context *intel = intel_context(ctx); - struct intel_texture_object *intelObj = intel_texture_object(texObj); - GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1; - int face, i; - - _mesa_generate_mipmap(ctx, target, texObj); - - /* Update the level information in our private data in the new images, since - * it didn't get set as part of a normal TexImage path. - */ - for (face = 0; face < nr_faces; face++) { - for (i = texObj->BaseLevel + 1; i < texObj->MaxLevel; i++) { - struct intel_texture_image *intelImage; - - intelImage = intel_texture_image(texObj->Image[face][i]); - if (intelImage == NULL) - break; - - intelImage->level = i; - intelImage->face = face; - /* Unreference the miptree to signal that the new Data is a bare - * pointer from mesa. - */ - intel_miptree_release(intel, &intelImage->mt); - } - } -} - -static void intelGenerateMipmap(GLcontext *ctx, GLenum target, struct gl_texture_object *texObj) -{ - struct intel_context *intel = intel_context(ctx); - struct intel_texture_object *intelObj = intel_texture_object(texObj); - - intel_tex_map_level_images(intel, intelObj, texObj->BaseLevel); - intel_generate_mipmap(ctx, target, texObj); - intel_tex_unmap_level_images(intel, intelObj, texObj->BaseLevel); -} - void intelInitTextureFuncs(struct dd_function_table *functions) { functions->ChooseTextureFormat = intelChooseTextureFormat; - functions->TexImage1D = intelTexImage1D; - functions->TexImage2D = intelTexImage2D; - functions->TexImage3D = intelTexImage3D; - functions->TexSubImage1D = intelTexSubImage1D; - functions->TexSubImage2D = intelTexSubImage2D; - functions->TexSubImage3D = intelTexSubImage3D; - functions->CopyTexImage1D = intelCopyTexImage1D; - functions->CopyTexImage2D = intelCopyTexImage2D; - functions->CopyTexSubImage1D = intelCopyTexSubImage1D; - functions->CopyTexSubImage2D = intelCopyTexSubImage2D; - functions->GetTexImage = intelGetTexImage; - functions->GenerateMipmap = intelGenerateMipmap; - - /* compressed texture functions */ - functions->CompressedTexImage2D = intelCompressedTexImage2D; - functions->CompressedTexSubImage2D = intelCompressedTexSubImage2D; - functions->GetCompressedTexImage = intelGetCompressedTexImage; + functions->GenerateMipmap = intel_generate_mipmap; functions->NewTextureObject = intelNewTextureObject; functions->NewTextureImage = intelNewTextureImage; diff --git a/src/mesa/drivers/dri/intel/intel_tex.h b/src/mesa/drivers/dri/intel/intel_tex.h index f5372d82fb..471aa2a240 100644 --- a/src/mesa/drivers/dri/intel/intel_tex.h +++ b/src/mesa/drivers/dri/intel/intel_tex.h @@ -35,116 +35,17 @@ void intelInitTextureFuncs(struct dd_function_table *functions); +void intelInitTextureImageFuncs(struct dd_function_table *functions); + +void intelInitTextureSubImageFuncs(struct dd_function_table *functions); + +void intelInitTextureCopyImageFuncs(struct dd_function_table *functions); + const struct gl_texture_format *intelChooseTextureFormat(GLcontext * ctx, GLint internalFormat, GLenum format, GLenum type); - -void intelTexImage3D(GLcontext * ctx, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint depth, - GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelTexSubImage3D(GLcontext * ctx, - GLenum target, - GLint level, - GLint xoffset, GLint yoffset, GLint zoffset, - GLsizei width, GLsizei height, GLsizei depth, - GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelTexImage2D(GLcontext * ctx, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelTexSubImage2D(GLcontext * ctx, - GLenum target, - GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLsizei height, - GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelTexImage1D(GLcontext * ctx, - GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint border, - GLenum format, GLenum type, const void *pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelTexSubImage1D(GLcontext * ctx, - GLenum target, - GLint level, - GLint xoffset, - GLsizei width, - GLenum format, GLenum type, - const GLvoid * pixels, - const struct gl_pixelstore_attrib *packing, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level, - GLenum internalFormat, - GLint x, GLint y, GLsizei width, GLint border); - -void intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level, - GLenum internalFormat, - GLint x, GLint y, GLsizei width, GLsizei height, - GLint border); - -void intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, - GLint xoffset, GLint x, GLint y, GLsizei width); - -void intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, - GLint xoffset, GLint yoffset, - GLint x, GLint y, GLsizei width, GLsizei height); - -void intelGetTexImage(GLcontext * ctx, GLenum target, GLint level, - GLenum format, GLenum type, GLvoid * pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ); - -void intelCompressedTexSubImage2D(GLcontext * ctx, - GLenum target, - GLint level, - GLint xoffset, GLint yoffset, - GLsizei width, GLsizei height, - GLenum format, GLsizei imageSize, - const GLvoid * pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - -void intelGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level, - GLvoid *pixels, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage); - void intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname, unsigned long long offset, GLint depth, GLuint pitch); void intelSetTexBuffer(__DRIcontext *pDRICtx, diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c index 90bbb8c6bb..260235b1eb 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_copy.c +++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c @@ -118,8 +118,12 @@ do_copy_texsubimage(struct intel_context *intel, dstx += x - orig_x; dsty += y - orig_y; - /* image_offset may be non-page-aligned, but that's illegal for tiling. */ - assert(intelImage->mt->region->tiling == I915_TILING_NONE); + /* Can't blit to tiled buffers with non-tile-aligned offset. */ + if (intelImage->mt->region->tiling != I915_TILING_NONE && + (image_offset & 4095) != 0) { + UNLOCK_HARDWARE(intel); + return GL_FALSE; + } if (ctx->ReadBuffer->Name == 0) { /* reading from a window, adjust x, y */ @@ -158,17 +162,14 @@ do_copy_texsubimage(struct intel_context *intel, /* GL_SGIS_generate_mipmap */ if (intelImage->level == texObj->BaseLevel && texObj->GenerateMipmap) { - ctx->Driver.GenerateMipmap(ctx, target, texObj); + intel_generate_mipmap(ctx, target, texObj); } return GL_TRUE; } - - - -void +static void intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level, GLenum internalFormat, GLint x, GLint y, GLsizei width, GLint border) @@ -214,7 +215,8 @@ intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level, width, border); } -void + +static void intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level, GLenum internalFormat, GLint x, GLint y, GLsizei width, GLsizei height, @@ -262,7 +264,7 @@ intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level, } -void +static void intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, GLint xoffset, GLint x, GLint y, GLsizei width) { @@ -287,8 +289,7 @@ intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, } - -void +static void intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, GLint xoffset, GLint yoffset, GLint x, GLint y, GLsizei width, GLsizei height) @@ -301,7 +302,6 @@ intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, _mesa_select_tex_image(ctx, texObj, target, level); GLenum internalFormat = texImage->InternalFormat; - /* Need to check texture is compatible with source format. */ @@ -316,3 +316,13 @@ intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, xoffset, yoffset, x, y, width, height); } } + + +void +intelInitTextureCopyImageFuncs(struct dd_function_table *functions) +{ + functions->CopyTexImage1D = intelCopyTexImage1D; + functions->CopyTexImage2D = intelCopyTexImage2D; + functions->CopyTexSubImage1D = intelCopyTexSubImage1D; + functions->CopyTexSubImage2D = intelCopyTexSubImage2D; +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c index 1f192dafbe..e9a3823078 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_image.c +++ b/src/mesa/drivers/dri/intel/intel_tex_image.c @@ -208,7 +208,7 @@ try_pbo_upload(struct intel_context *intel, if (!pbo || intel->ctx._ImageTransferState || unpack->SkipPixels || unpack->SkipRows) { - _mesa_printf("%s: failure 1\n", __FUNCTION__); + DBG("%s: failure 1\n", __FUNCTION__); return GL_FALSE; } @@ -248,7 +248,6 @@ try_pbo_upload(struct intel_context *intel, } - static GLboolean try_pbo_zcopy(struct intel_context *intel, struct intel_texture_image *intelImage, @@ -264,7 +263,7 @@ try_pbo_zcopy(struct intel_context *intel, if (!pbo || intel->ctx._ImageTransferState || unpack->SkipPixels || unpack->SkipRows) { - _mesa_printf("%s: failure 1\n", __FUNCTION__); + DBG("%s: failure 1\n", __FUNCTION__); return GL_FALSE; } @@ -283,7 +282,7 @@ try_pbo_zcopy(struct intel_context *intel, dst_stride = intelImage->mt->pitch; if (src_stride != dst_stride || dst_offset != 0 || src_offset != 0) { - _mesa_printf("%s: failure 2\n", __FUNCTION__); + DBG("%s: failure 2\n", __FUNCTION__); return GL_FALSE; } @@ -293,10 +292,6 @@ try_pbo_zcopy(struct intel_context *intel, } - - - - static void intelTexImage(GLcontext * ctx, GLint dims, @@ -307,7 +302,8 @@ intelTexImage(GLcontext * ctx, GLenum format, GLenum type, const void *pixels, const struct gl_pixelstore_attrib *unpack, struct gl_texture_object *texObj, - struct gl_texture_image *texImage, GLsizei imageSize, int compressed) + struct gl_texture_image *texImage, GLsizei imageSize, + GLboolean compressed) { struct intel_context *intel = intel_context(ctx); struct intel_texture_object *intelObj = intel_texture_object(texObj); @@ -316,7 +312,6 @@ intelTexImage(GLcontext * ctx, GLint postConvHeight = height; GLint texelBytes, sizeInBytes; GLuint dstRowStride = 0, srcRowStride = texImage->RowStride; - GLboolean needs_map; DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__, _mesa_lookup_enum_by_nr(target), level, width, height, depth, border); @@ -464,8 +459,6 @@ intelTexImage(GLcontext * ctx, DBG("pbo upload failed\n"); } - - /* intelCopyTexImage calls this function with pixels == NULL, with * the expectation that the mipmap tree will be set up but nothing * more will be done. This is where those calls return: @@ -482,15 +475,8 @@ intelTexImage(GLcontext * ctx, LOCK_HARDWARE(intel); - /* Two cases where we need a mapping of the miptree: when the user supplied - * data is mapped as well (non-PBO, memcpy upload) or when we're going to do - * (software) mipmap generation. - */ - needs_map = (pixels != NULL) || (level == texObj->BaseLevel && - texObj->GenerateMipmap); - if (intelImage->mt) { - if (needs_map) + if (pixels != NULL) texImage->Data = intel_miptree_image_map(intel, intelImage->mt, intelImage->face, @@ -547,25 +533,26 @@ intelTexImage(GLcontext * ctx, format, type, pixels, unpack)) { _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage"); } - - /* GL_SGIS_generate_mipmap */ - if (level == texObj->BaseLevel && texObj->GenerateMipmap) { - intel_generate_mipmap(ctx, target, texObj); - } } _mesa_unmap_teximage_pbo(ctx, unpack); if (intelImage->mt) { - if (needs_map) + if (pixels != NULL) intel_miptree_image_unmap(intel, intelImage->mt); texImage->Data = NULL; } UNLOCK_HARDWARE(intel); + + /* GL_SGIS_generate_mipmap */ + if (level == texObj->BaseLevel && texObj->GenerateMipmap) { + intel_generate_mipmap(ctx, target, texObj); + } } -void + +static void intelTexImage3D(GLcontext * ctx, GLenum target, GLint level, GLint internalFormat, @@ -578,11 +565,11 @@ intelTexImage3D(GLcontext * ctx, { intelTexImage(ctx, 3, target, level, internalFormat, width, height, depth, border, - format, type, pixels, unpack, texObj, texImage, 0, 0); + format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); } -void +static void intelTexImage2D(GLcontext * ctx, GLenum target, GLint level, GLint internalFormat, @@ -594,10 +581,11 @@ intelTexImage2D(GLcontext * ctx, { intelTexImage(ctx, 2, target, level, internalFormat, width, height, 1, border, - format, type, pixels, unpack, texObj, texImage, 0, 0); + format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); } -void + +static void intelTexImage1D(GLcontext * ctx, GLenum target, GLint level, GLint internalFormat, @@ -609,21 +597,24 @@ intelTexImage1D(GLcontext * ctx, { intelTexImage(ctx, 1, target, level, internalFormat, width, 1, 1, border, - format, type, pixels, unpack, texObj, texImage, 0, 0); + format, type, pixels, unpack, texObj, texImage, 0, GL_FALSE); } -void intelCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level, - GLint internalFormat, - GLint width, GLint height, GLint border, - GLsizei imageSize, const GLvoid *data, - struct gl_texture_object *texObj, - struct gl_texture_image *texImage ) + +static void +intelCompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level, + GLint internalFormat, + GLint width, GLint height, GLint border, + GLsizei imageSize, const GLvoid *data, + struct gl_texture_object *texObj, + struct gl_texture_image *texImage ) { intelTexImage(ctx, 2, target, level, internalFormat, width, height, 1, border, - 0, 0, data, &ctx->Unpack, texObj, texImage, imageSize, 1); + 0, 0, data, &ctx->Unpack, texObj, texImage, imageSize, GL_TRUE); } + /** * Need to map texture image into memory before copying image data, * then unmap it. @@ -632,11 +623,17 @@ static void intel_get_tex_image(GLcontext * ctx, GLenum target, GLint level, GLenum format, GLenum type, GLvoid * pixels, struct gl_texture_object *texObj, - struct gl_texture_image *texImage, int compressed) + struct gl_texture_image *texImage, GLboolean compressed) { struct intel_context *intel = intel_context(ctx); struct intel_texture_image *intelImage = intel_texture_image(texImage); + /* If we're reading from a texture that has been rendered to, need to + * make sure rendering is complete. + * We could probably predicate this on texObj->_RenderToTexture + */ + intelFlush(ctx); + /* Map */ if (intelImage->mt) { /* Image is stored in hardware format in a buffer managed by the @@ -680,28 +677,29 @@ intel_get_tex_image(GLcontext * ctx, GLenum target, GLint level, } } -void + +static void intelGetTexImage(GLcontext * ctx, GLenum target, GLint level, GLenum format, GLenum type, GLvoid * pixels, struct gl_texture_object *texObj, struct gl_texture_image *texImage) { intel_get_tex_image(ctx, target, level, format, type, pixels, - texObj, texImage, 0); - - + texObj, texImage, GL_FALSE); } -void + +static void intelGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level, GLvoid *pixels, struct gl_texture_object *texObj, struct gl_texture_image *texImage) { intel_get_tex_image(ctx, target, level, 0, 0, pixels, - texObj, texImage, 1); + texObj, texImage, GL_TRUE); } + void intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname, unsigned long long offset, GLint depth, GLuint pitch) @@ -810,3 +808,16 @@ intelSetTexBuffer(__DRIcontext *pDRICtx, GLint target, __DRIdrawable *dPriv) */ intelSetTexBuffer2(pDRICtx, target, GLX_TEXTURE_FORMAT_RGBA_EXT, dPriv); } + + +void +intelInitTextureImageFuncs(struct dd_function_table *functions) +{ + functions->TexImage1D = intelTexImage1D; + functions->TexImage2D = intelTexImage2D; + functions->TexImage3D = intelTexImage3D; + functions->GetTexImage = intelGetTexImage; + + functions->CompressedTexImage2D = intelCompressedTexImage2D; + functions->GetCompressedTexImage = intelGetCompressedTexImage; +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_layout.c b/src/mesa/drivers/dri/intel/intel_tex_layout.c index e6f9a41779..2c1b722b7f 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_layout.c +++ b/src/mesa/drivers/dri/intel/intel_tex_layout.c @@ -52,7 +52,9 @@ GLuint intel_compressed_alignment(GLenum internalFormat) return alignment; } -void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tree *mt ) +void i945_miptree_layout_2d( struct intel_context *intel, + struct intel_mipmap_tree *mt, + uint32_t tiling ) { GLint align_h = 2, align_w = 4; GLuint level; @@ -92,7 +94,7 @@ void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tr /* Pitch must be a whole number of dwords, even though we * express it in texels. */ - mt->pitch = intel_miptree_pitch_align (intel, mt, mt->pitch); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->pitch); mt->total_height = 0; for ( level = mt->first_level ; level <= mt->last_level ; level++ ) { diff --git a/src/mesa/drivers/dri/intel/intel_tex_layout.h b/src/mesa/drivers/dri/intel/intel_tex_layout.h index dbc90e6f9b..7bc25b6bcb 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_layout.h +++ b/src/mesa/drivers/dri/intel/intel_tex_layout.h @@ -38,5 +38,7 @@ static GLuint minify( GLuint d ) return MAX2(1, d>>1); } -extern void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tree *mt ); +extern void i945_miptree_layout_2d(struct intel_context *intel, + struct intel_mipmap_tree *mt, + uint32_t tiling); extern GLuint intel_compressed_alignment(GLenum); diff --git a/src/mesa/drivers/dri/intel/intel_tex_subimage.c b/src/mesa/drivers/dri/intel/intel_tex_subimage.c index f86de56897..1f27131dac 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_subimage.c +++ b/src/mesa/drivers/dri/intel/intel_tex_subimage.c @@ -101,11 +101,6 @@ intelTexSubimage(GLcontext * ctx, _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage"); } - /* GL_SGIS_generate_mipmap */ - if (level == texObj->BaseLevel && texObj->GenerateMipmap) { - intel_generate_mipmap(ctx, target, texObj); - } - _mesa_unmap_teximage_pbo(ctx, packing); if (intelImage->mt) { @@ -114,13 +109,15 @@ intelTexSubimage(GLcontext * ctx, } UNLOCK_HARDWARE(intel); -} - - + /* GL_SGIS_generate_mipmap */ + if (level == texObj->BaseLevel && texObj->GenerateMipmap) { + intel_generate_mipmap(ctx, target, texObj); + } +} -void +static void intelTexSubImage3D(GLcontext * ctx, GLenum target, GLint level, @@ -132,18 +129,15 @@ intelTexSubImage3D(GLcontext * ctx, struct gl_texture_object *texObj, struct gl_texture_image *texImage) { - intelTexSubimage(ctx, 3, target, level, xoffset, yoffset, zoffset, width, height, depth, format, type, pixels, packing, texObj, texImage); - } - -void +static void intelTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, @@ -155,17 +149,15 @@ intelTexSubImage2D(GLcontext * ctx, struct gl_texture_object *texObj, struct gl_texture_image *texImage) { - intelTexSubimage(ctx, 2, target, level, xoffset, yoffset, 0, width, height, 1, format, type, pixels, packing, texObj, texImage); - } -void +static void intelTexSubImage1D(GLcontext * ctx, GLenum target, GLint level, @@ -182,10 +174,9 @@ intelTexSubImage1D(GLcontext * ctx, xoffset, 0, 0, width, 1, 1, format, type, pixels, packing, texObj, texImage); - } -void +static void intelCompressedTexSubImage2D(GLcontext * ctx, GLenum target, GLint level, @@ -199,3 +190,14 @@ intelCompressedTexSubImage2D(GLcontext * ctx, fprintf(stderr, "stubbed CompressedTexSubImage2D: %dx%d@%dx%d\n", width, height, xoffset, yoffset); } + + + +void +intelInitTextureSubImageFuncs(struct dd_function_table *functions) +{ + functions->TexSubImage1D = intelTexSubImage1D; + functions->TexSubImage2D = intelTexSubImage2D; + functions->TexSubImage3D = intelTexSubImage3D; + functions->CompressedTexSubImage2D = intelCompressedTexSubImage2D; +} diff --git a/src/mesa/drivers/dri/intel/intel_tex_validate.c b/src/mesa/drivers/dri/intel/intel_tex_validate.c index 05a375e1f3..b5cb7597d1 100644 --- a/src/mesa/drivers/dri/intel/intel_tex_validate.c +++ b/src/mesa/drivers/dri/intel/intel_tex_validate.c @@ -241,7 +241,7 @@ intel_tex_map_level_images(struct intel_context *intel, struct intel_texture_image *intelImage = intel_texture_image(intelObj->base.Image[face][level]); - if (intelImage->mt) { + if (intelImage && intelImage->mt) { intelImage->base.Data = intel_miptree_image_map(intel, intelImage->mt, @@ -268,7 +268,7 @@ intel_tex_unmap_level_images(struct intel_context *intel, struct intel_texture_image *intelImage = intel_texture_image(intelObj->base.Image[face][level]); - if (intelImage->mt) { + if (intelImage && intelImage->mt) { intel_miptree_image_unmap(intel, intelImage->mt); intelImage->base.Data = NULL; } diff --git a/src/mesa/drivers/dri/r200/r200_state.c b/src/mesa/drivers/dri/r200/r200_state.c index 2fcc87c0f5..81ee1ed022 100644 --- a/src/mesa/drivers/dri/r200/r200_state.c +++ b/src/mesa/drivers/dri/r200/r200_state.c @@ -2480,11 +2480,11 @@ void r200ValidateState( GLcontext *ctx ) r200ContextPtr rmesa = R200_CONTEXT(ctx); GLuint new_state = rmesa->NewGLState; - if (new_state & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { + if (new_state & _NEW_BUFFERS) { r200UpdateDrawBuffer(ctx); } - if (new_state & (_NEW_TEXTURE | _NEW_PROGRAM)) { + if (new_state & (_NEW_TEXTURE | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS)) { r200UpdateTextureState( ctx ); new_state |= rmesa->NewGLState; /* may add TEXTURE_MATRIX */ r200UpdateLocalViewer( ctx ); @@ -2523,6 +2523,7 @@ void r200ValidateState( GLcontext *ctx ) } if (new_state & (_NEW_PROGRAM| + _NEW_PROGRAM_CONSTANTS | /* need to test for pretty much anything due to possible parameter bindings */ _NEW_MODELVIEW|_NEW_PROJECTION|_NEW_TRANSFORM| _NEW_LIGHT|_NEW_TEXTURE|_NEW_TEXTURE_MATRIX| diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c index 873cde4414..2f45429cf2 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog.c +++ b/src/mesa/drivers/dri/r300/r300_fragprog.c @@ -470,7 +470,8 @@ void r300TranslateFragmentShader(r300ContextPtr r300, fp->translated = GL_TRUE; if (fp->error || (RADEON_DEBUG & DEBUG_PIXEL)) r300FragmentProgramDump(fp, &fp->code); - r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM); + r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM | + _NEW_PROGRAM_CONSTANTS); } update_params(r300, fp); diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c index 79f0b3625c..07299f3b36 100644 --- a/src/mesa/drivers/dri/r300/r300_state.c +++ b/src/mesa/drivers/dri/r300/r300_state.c @@ -1109,7 +1109,7 @@ void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state) struct gl_program_parameter_list *paramList; GLuint i; - if (!(new_state & (_NEW_BUFFERS | _NEW_PROGRAM))) + if (!(new_state & (_NEW_BUFFERS | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS))) return; fp = (struct r300_fragment_program *)ctx->FragmentProgram._Current; @@ -2357,11 +2357,12 @@ void r300UpdateShaders(r300ContextPtr rmesa) hw_tcl_on = future_hw_tcl_on = 0; r300ResetHwState(rmesa); - r300UpdateStateParameters(ctx, _NEW_PROGRAM); + r300UpdateStateParameters(ctx, _NEW_PROGRAM | + _NEW_PROGRAM_CONSTANTS); return; } } - r300UpdateStateParameters(ctx, _NEW_PROGRAM); + r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS); } static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, @@ -2588,7 +2589,7 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state) _tnl_InvalidateState(ctx, new_state); _ae_invalidate_state(ctx, new_state); - if (new_state & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { + if (new_state & _NEW_BUFFERS) { r300UpdateDrawBuffer(ctx); } diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.c b/src/mesa/drivers/dri/r300/r500_fragprog.c index 292573de89..300559d0b4 100644 --- a/src/mesa/drivers/dri/r300/r500_fragprog.c +++ b/src/mesa/drivers/dri/r300/r500_fragprog.c @@ -501,7 +501,8 @@ void r500TranslateFragmentShader(r300ContextPtr r300, _mesa_reference_program(r300->radeon.glCtx, &compiler.program, 0); - r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM); + r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM | + _NEW_PROGRAM_CONSTANTS); if (RADEON_DEBUG & DEBUG_PIXEL) { if (fp->translated) { diff --git a/src/mesa/drivers/dri/radeon/radeon_state.c b/src/mesa/drivers/dri/radeon/radeon_state.c index b6561001e7..4432f85691 100644 --- a/src/mesa/drivers/dri/radeon/radeon_state.c +++ b/src/mesa/drivers/dri/radeon/radeon_state.c @@ -2255,7 +2255,7 @@ void radeonValidateState( GLcontext *ctx ) radeonContextPtr rmesa = RADEON_CONTEXT(ctx); GLuint new_state = rmesa->NewGLState; - if (new_state & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) { + if (new_state & _NEW_BUFFERS) { radeonUpdateDrawBuffer(ctx); } |