diff options
Diffstat (limited to 'src/gallium/drivers/r300')
32 files changed, 2340 insertions, 1307 deletions
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile index 121b65063f..afddcb161f 100644 --- a/src/gallium/drivers/r300/Makefile +++ b/src/gallium/drivers/r300/Makefile @@ -4,8 +4,8 @@ include $(TOP)/configs/current LIBNAME = r300 C_SOURCES = \ + r300_blit.c \ r300_chipset.c \ - r300_clear.c \ r300_context.c \ r300_debug.c \ r300_emit.c \ @@ -17,13 +17,13 @@ C_SOURCES = \ r300_state.c \ r300_state_derived.c \ r300_state_invariant.c \ - r300_vbo.c \ r300_vs.c \ r300_texture.c \ r300_tgsi_to_rc.c LIBRARY_INCLUDES = \ - -I$(TOP)/src/mesa/drivers/dri/r300/compiler + -I$(TOP)/src/mesa/drivers/dri/r300/compiler \ + -I$(TOP)/src/gallium/winsys/drm/radeon/core COMPILER_ARCHIVE = $(TOP)/src/mesa/drivers/dri/r300/compiler/libr300compiler.a diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript index 97989040d2..183aa17f9b 100644 --- a/src/gallium/drivers/r300/SConscript +++ b/src/gallium/drivers/r300/SConscript @@ -4,13 +4,18 @@ r300compiler = SConscript('#/src/mesa/drivers/dri/r300/compiler/SConscript') env = env.Clone() # add the paths for r300compiler -env.Append(CPPPATH = ['#/src/mesa/drivers/dri/r300/compiler', '#/include', '#/src/mesa']) +env.Append(CPPPATH = [ + '#/src/mesa/drivers/dri/r300/compiler', + '#/src/gallium/winsys/drm/radeon/core', + '#/include', + '#/src/mesa', +]) r300 = env.ConvenienceLibrary( target = 'r300', source = [ + 'r300_blit.c', 'r300_chipset.c', - 'r300_clear.c', 'r300_context.c', 'r300_debug.c', 'r300_emit.c', diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c new file mode 100644 index 0000000000..c14414fff6 --- /dev/null +++ b/src/gallium/drivers/r300/r300_blit.c @@ -0,0 +1,130 @@ +/* + * Copyright 2009 Marek Olšák <maraeo@gmail.com> + * + * 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 + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, + * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR + * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE + * USE OR OTHER DEALINGS IN THE SOFTWARE. */ + +#include "r300_blit.h" +#include "r300_context.h" + +#include "util/u_rect.h" + +static void r300_blitter_save_states(struct r300_context* r300) +{ + util_blitter_save_blend(r300->blitter, r300->blend_state.state); + util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state); + util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state); + util_blitter_save_fragment_shader(r300->blitter, r300->fs); + util_blitter_save_vertex_shader(r300->blitter, r300->vs); +} + +/* Clear currently bound buffers. */ +void r300_clear(struct pipe_context* pipe, + unsigned buffers, + const float* rgba, + double depth, + unsigned stencil) +{ + /* XXX Implement fastfill. + * + * If fastfill is enabled, a few facts should be considered: + * + * 1) Zbuffer must be micro-tiled and whole microtiles must be + * written. + * + * 2) ZB_DEPTHCLEARVALUE is used to clear a zbuffer and Z Mask must be + * equal to 0. + * + * 3) RB3D_COLOR_CLEAR_VALUE is used to clear a colorbuffer and + * RB3D_COLOR_CHANNEL_MASK must be equal to 0. + * + * 4) ZB_CB_CLEAR can be used to make the ZB units help in clearing + * the colorbuffer. The color clear value is supplied through both + * RB3D_COLOR_CLEAR_VALUE and ZB_DEPTHCLEARVALUE, and the colorbuffer + * must be set in ZB_DEPTHOFFSET and ZB_DEPTHPITCH in addition to + * RB3D_COLOROFFSET and RB3D_COLORPITCH. It's obvious that the zbuffer + * will not be cleared and multiple render targets cannot be cleared + * this way either. + * + * 5) For 16-bit integer buffering, compression causes a hung with one or + * two samples and should not be used. + * + * 6) Fastfill must not be used if reading of compressed Z data is disabled + * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE), + * i.e. it cannot be used to compress the zbuffer. + * (what the hell does that mean and how does it fit in clearing + * the buffers?) + * + * - Marek + */ + + struct r300_context* r300 = r300_context(pipe); + + r300_blitter_save_states(r300); + + util_blitter_clear(r300->blitter, + r300->framebuffer_state.width, + r300->framebuffer_state.height, + r300->framebuffer_state.nr_cbufs, + buffers, rgba, depth, stencil); +} + +/* Copy a block of pixels from one surface to another. */ +void r300_surface_copy(struct pipe_context* pipe, + struct pipe_surface* dst, + unsigned dstx, unsigned dsty, + struct pipe_surface* src, + unsigned srcx, unsigned srcy, + unsigned width, unsigned height) +{ + struct r300_context* r300 = r300_context(pipe); + + /* Yeah we have to save all those states to ensure this blitter operation + * is really transparent. The states will be restored by the blitter once + * copying is done. */ + r300_blitter_save_states(r300); + util_blitter_save_framebuffer(r300->blitter, &r300->framebuffer_state); + + util_blitter_save_fragment_sampler_states( + r300->blitter, r300->sampler_count, (void**)r300->sampler_states); + + util_blitter_save_fragment_sampler_textures( + r300->blitter, r300->texture_count, + (struct pipe_texture**)r300->textures); + + /* Do a copy */ + util_blitter_copy(r300->blitter, + dst, dstx, dsty, src, srcx, srcy, width, height, TRUE); +} + +/* Fill a region of a surface with a constant value. */ +void r300_surface_fill(struct pipe_context* pipe, + struct pipe_surface* dst, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height, + unsigned value) +{ + struct r300_context* r300 = r300_context(pipe); + + r300_blitter_save_states(r300); + util_blitter_save_framebuffer(r300->blitter, &r300->framebuffer_state); + + util_blitter_fill(r300->blitter, + dst, dstx, dsty, width, height, value); +} diff --git a/src/gallium/drivers/r300/r300_clear.h b/src/gallium/drivers/r300/r300_blit.h index b8fcdf273c..029e4f98e7 100644 --- a/src/gallium/drivers/r300/r300_clear.h +++ b/src/gallium/drivers/r300/r300_blit.h @@ -1,5 +1,5 @@ /* - * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2008 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -20,10 +20,11 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ -#ifndef R300_CLEAR_H -#define R300_CLEAR_H +#ifndef R300_BLIT_H +#define R300_BLIT_H struct pipe_context; +struct pipe_surface; void r300_clear(struct pipe_context* pipe, unsigned buffers, @@ -31,4 +32,17 @@ void r300_clear(struct pipe_context* pipe, double depth, unsigned stencil); -#endif /* R300_CLEAR_H */ +void r300_surface_copy(struct pipe_context* pipe, + struct pipe_surface* dst, + unsigned dstx, unsigned dsty, + struct pipe_surface* src, + unsigned srcx, unsigned srcy, + unsigned width, unsigned height); + +void r300_surface_fill(struct pipe_context* pipe, + struct pipe_surface* dst, + unsigned dstx, unsigned dsty, + unsigned width, unsigned height, + unsigned value); + +#endif /* R300_BLIT_H */ diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c index 51fdb82ff3..92de297ef1 100644 --- a/src/gallium/drivers/r300/r300_chipset.c +++ b/src/gallium/drivers/r300/r300_chipset.c @@ -33,6 +33,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) /* Reasonable defaults */ caps->num_vert_fpus = 4; caps->has_tcl = debug_get_bool_option("RADEON_NO_TCL", FALSE) ? FALSE : TRUE; + caps->is_r400 = FALSE; caps->is_r500 = FALSE; caps->high_second_pipe = FALSE; @@ -123,6 +124,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x4A54: caps->family = CHIP_FAMILY_R420; caps->num_vert_fpus = 6; + caps->is_r400 = TRUE; break; case 0x5548: @@ -136,6 +138,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x5D57: caps->family = CHIP_FAMILY_R423; caps->num_vert_fpus = 6; + caps->is_r400 = TRUE; break; case 0x554C: @@ -147,6 +150,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x5D4A: caps->family = CHIP_FAMILY_R430; caps->num_vert_fpus = 6; + caps->is_r400 = TRUE; break; case 0x5D4C: @@ -157,6 +161,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x5D52: caps->family = CHIP_FAMILY_R480; caps->num_vert_fpus = 6; + caps->is_r400 = TRUE; break; case 0x4B48: @@ -166,6 +171,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x4B4C: caps->family = CHIP_FAMILY_R481; caps->num_vert_fpus = 6; + caps->is_r400 = TRUE; break; case 0x5E4C: @@ -182,6 +188,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x5E4D: caps->family = CHIP_FAMILY_RV410; caps->num_vert_fpus = 6; + caps->is_r400 = TRUE; break; case 0x5954: @@ -212,6 +219,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x791F: caps->family = CHIP_FAMILY_RS690; caps->has_tcl = FALSE; + caps->is_r400 = TRUE; break; case 0x793F: @@ -219,6 +227,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x7942: caps->family = CHIP_FAMILY_RS600; caps->has_tcl = FALSE; + caps->is_r400 = TRUE; break; case 0x796C: @@ -227,6 +236,7 @@ void r300_parse_chipset(struct r300_capabilities* caps) case 0x796F: caps->family = CHIP_FAMILY_RS740; caps->has_tcl = FALSE; + caps->is_r400 = TRUE; break; case 0x7100: diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h index 0633a8b8a7..2808486492 100644 --- a/src/gallium/drivers/r300/r300_chipset.h +++ b/src/gallium/drivers/r300/r300_chipset.h @@ -40,11 +40,18 @@ struct r300_capabilities { unsigned num_z_pipes; /* Whether or not TCL is physically present */ boolean has_tcl; + /* Whether or not this is R400. The differences compared to their R3xx + * cousins are: + * - Extended fragment shader registers + * - Blend LTE/GTE thresholds */ + boolean is_r400; /* Whether or not this is an RV515 or newer; R500s have many differences * that require extra consideration, compared to their R3xx cousins: * - Extra bit of width and height on texture sizes * - Blend color is split across two registers - * - Universal Shader (US) block used for fragment shaders */ + * - Blend LTE/GTE thresholds + * - Universal Shader (US) block used for fragment shaders + * - FP16 blending and multisampling */ boolean is_r500; /* Whether or not the second pixel pipe is accessed with the high bit */ boolean high_second_pipe; diff --git a/src/gallium/drivers/r300/r300_clear.c b/src/gallium/drivers/r300/r300_clear.c deleted file mode 100644 index 02d6d504fc..0000000000 --- a/src/gallium/drivers/r300/r300_clear.c +++ /dev/null @@ -1,38 +0,0 @@ -/* - * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> - * - * 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 - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM, - * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR - * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE - * USE OR OTHER DEALINGS IN THE SOFTWARE. */ - -#include "r300_clear.h" -#include "r300_context.h" - -#include "util/u_clear.h" - -/* Clears currently bound buffers. */ -void r300_clear(struct pipe_context* pipe, - unsigned buffers, - const float* rgba, - double depth, - unsigned stencil) -{ - /* XXX we can and should do one clear if both color and zs are set */ - util_clear(pipe, &r300_context(pipe)->framebuffer_state, - buffers, rgba, depth, stencil); -} diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c index ae23329b83..5e4f6552c3 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -28,14 +28,16 @@ #include "util/u_memory.h" #include "util/u_simple_list.h" -#include "r300_clear.h" +#include "r300_blit.h" #include "r300_context.h" +#include "r300_emit.h" #include "r300_flush.h" #include "r300_query.h" #include "r300_render.h" #include "r300_screen.h" #include "r300_state_derived.h" #include "r300_state_invariant.h" +#include "r300_texture.h" #include "r300_winsys.h" static enum pipe_error r300_clear_hash_table(void* key, void* value, @@ -51,6 +53,8 @@ static void r300_destroy_context(struct pipe_context* context) struct r300_context* r300 = r300_context(context); struct r300_query* query, * temp; + util_blitter_destroy(r300->blitter); + util_hash_table_foreach(r300->shader_hash_table, r300_clear_hash_table, NULL); util_hash_table_destroy(r300->shader_hash_table); @@ -66,10 +70,13 @@ static void r300_destroy_context(struct pipe_context* context) FREE(query); } - FREE(r300->blend_color_state); + FREE(r300->blend_color_state.state); + FREE(r300->clip_state.state); FREE(r300->rs_block); - FREE(r300->scissor_state); - FREE(r300->viewport_state); + FREE(r300->scissor_state.state); + FREE(r300->vertex_info); + FREE(r300->viewport_state.state); + FREE(r300->ztop_state.state); FREE(r300); } @@ -103,8 +110,37 @@ static void r300_flush_cb(void *data) cs_context_copy->context.flush(&cs_context_copy->context, 0, NULL); } +#define R300_INIT_ATOM(atomname, atomsize) \ + r300->atomname##_state.name = #atomname; \ + r300->atomname##_state.state = NULL; \ + r300->atomname##_state.size = atomsize; \ + r300->atomname##_state.emit = r300_emit_##atomname##_state; \ + r300->atomname##_state.dirty = FALSE; \ + insert_at_tail(&r300->atom_list, &r300->atomname##_state); + +static void r300_setup_atoms(struct r300_context* r300) +{ + /* Create the actual atom list. + * + * Each atom is examined and emitted in the order it appears here, which + * can affect performance and conformance if not handled with care. + * + * Some atoms never change size, others change every emit. This is just + * an upper bound on each atom, to keep the emission machinery from + * underallocating space. */ + make_empty_list(&r300->atom_list); + R300_INIT_ATOM(ztop, 2); + R300_INIT_ATOM(blend, 8); + R300_INIT_ATOM(blend_color, 3); + R300_INIT_ATOM(clip, 29); + R300_INIT_ATOM(dsa, 8); + R300_INIT_ATOM(rs, 22); + R300_INIT_ATOM(scissor, 3); + R300_INIT_ATOM(viewport, 9); +} + struct pipe_context* r300_create_context(struct pipe_screen* screen, - struct r300_winsys* r300_winsys) + struct radeon_winsys* radeon_winsys) { struct r300_context* r300 = CALLOC_STRUCT(r300_context); struct r300_screen* r300screen = r300_screen(screen); @@ -112,9 +148,9 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, if (!r300) return NULL; - r300->winsys = r300_winsys; + r300->winsys = radeon_winsys; - r300->context.winsys = (struct pipe_winsys*)r300_winsys; + r300->context.winsys = (struct pipe_winsys*)radeon_winsys; r300->context.screen = screen; r300_init_debug(r300); @@ -122,16 +158,27 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->context.destroy = r300_destroy_context; r300->context.clear = r300_clear; + r300->context.surface_copy = r300_surface_copy; + r300->context.surface_fill = r300_surface_fill; - if (r300screen->caps->has_tcl) - { + if (r300screen->caps->has_tcl) { r300->context.draw_arrays = r300_draw_arrays; r300->context.draw_elements = r300_draw_elements; r300->context.draw_range_elements = r300_draw_range_elements; - } - else - { - assert(0); + } else { + r300->context.draw_arrays = r300_swtcl_draw_arrays; + r300->context.draw_elements = r300_draw_elements; + r300->context.draw_range_elements = r300_swtcl_draw_range_elements; + + /* Create a Draw. This is used for SW TCL. */ + r300->draw = draw_create(); + /* Enable our renderer. */ + draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); + /* Enable Draw's clipping. */ + draw_set_driver_clipping(r300->draw, FALSE); + /* Force Draw to never do viewport transform, since we can do + * transform in hardware, always. */ + draw_set_viewport_state(r300->draw, &r300_viewport_identity); } r300->context.is_texture_referenced = r300_is_texture_referenced; @@ -140,20 +187,15 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->shader_hash_table = util_hash_table_create(r300_shader_key_hash, r300_shader_key_compare); - r300->blend_color_state = CALLOC_STRUCT(r300_blend_color_state); + r300_setup_atoms(r300); + + r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state); + r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state); r300->rs_block = CALLOC_STRUCT(r300_rs_block); - r300->scissor_state = CALLOC_STRUCT(r300_scissor_state); - r300->viewport_state = CALLOC_STRUCT(r300_viewport_state); - - /* Create a Draw. This is used for vert collation and SW TCL. */ - r300->draw = draw_create(); - /* Enable our renderer. */ - draw_set_rasterize_stage(r300->draw, r300_draw_stage(r300)); - /* Disable Draw's clipping if TCL is present. */ - draw_set_driver_clipping(r300->draw, r300_screen(screen)->caps->has_tcl); - /* Force Draw to never do viewport transform, since (again) we can do - * transform in hardware, always. */ - draw_set_viewport_state(r300->draw, &r300_viewport_identity); + r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state); + r300->vertex_info = CALLOC_STRUCT(r300_vertex_info); + r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state); + r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state); /* Open up the OQ BO. */ r300->oqbo = screen->buffer_create(screen, 4096, @@ -173,5 +215,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen, r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300); r300->dirty_state = R300_NEW_KITCHEN_SINK; r300->dirty_hw++; + + r300->blitter = util_blitter_create(&r300->context); + return &r300->context; } diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index f954ba7f9a..682b9179c8 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -25,12 +25,33 @@ #include "draw/draw_vertex.h" +#include "util/u_blitter.h" + #include "pipe/p_context.h" #include "pipe/p_inlines.h" +struct r300_context; + struct r300_fragment_shader; struct r300_vertex_shader; +struct r300_atom { + /* List pointers. */ + struct r300_atom *prev, *next; + /* Name, for debugging. */ + const char* name; + /* Opaque state. */ + void* state; + /* Emit the state to the context. */ + void (*emit)(struct r300_context*, void*); + /* Upper bound on number of dwords to emit. */ + unsigned size; + /* Whether this atom should be emitted. */ + boolean dirty; + /* Another dirty flag that is never automatically cleared. */ + boolean always_dirty; +}; + struct r300_blend_state { uint32_t blend_control; /* R300_RB3D_CBLEND: 0x4e04 */ uint32_t alpha_blend_control; /* R300_RB3D_ABLEND: 0x4e08 */ @@ -60,11 +81,6 @@ struct r300_rs_state { /* Draw-specific rasterizer state */ struct pipe_rasterizer_state rs; - /* Whether or not to enable the VTE. This is referenced at the very - * last moment during emission of VTE state, to decide whether or not - * the VTE should be used for transformation. */ - boolean enable_vte; - uint32_t vap_control_status; /* R300_VAP_CNTL_STATUS: 0x2140 */ uint32_t point_size; /* R300_GA_POINT_SIZE: 0x421c */ uint32_t point_minmax; /* R300_GA_POINT_MINMAX: 0x4230 */ @@ -89,14 +105,15 @@ struct r300_rs_block { }; struct r300_sampler_state { + struct pipe_sampler_state state; + uint32_t filter0; /* R300_TX_FILTER0: 0x4400 */ uint32_t filter1; /* R300_TX_FILTER1: 0x4440 */ uint32_t border_color; /* R300_TX_BORDER_COLOR: 0x45c0 */ -}; -struct r300_scissor_state { - uint32_t scissor_top_left; /* R300_SC_SCISSORS_TL: 0x43e0 */ - uint32_t scissor_bottom_right; /* R300_SC_SCISSORS_BR: 0x43e4 */ + /* Min/max LOD must be clamped to [0, last_level], thus + * it's dependent on a currently bound texture */ + unsigned min_lod, max_lod; }; struct r300_texture_state { @@ -119,24 +136,17 @@ struct r300_ztop_state { uint32_t z_buffer_top; /* R300_ZB_ZTOP: 0x4f14 */ }; -#define R300_NEW_BLEND 0x00000001 -#define R300_NEW_BLEND_COLOR 0x00000002 -#define R300_NEW_CLIP 0x00000004 -#define R300_NEW_DSA 0x00000008 #define R300_NEW_FRAMEBUFFERS 0x00000010 #define R300_NEW_FRAGMENT_SHADER 0x00000020 #define R300_NEW_FRAGMENT_SHADER_CONSTANTS 0x00000040 -#define R300_NEW_RASTERIZER 0x00000080 #define R300_NEW_RS_BLOCK 0x00000100 #define R300_NEW_SAMPLER 0x00000200 #define R300_ANY_NEW_SAMPLERS 0x0001fe00 -#define R300_NEW_SCISSOR 0x00020000 #define R300_NEW_TEXTURE 0x00040000 #define R300_ANY_NEW_TEXTURES 0x03fc0000 #define R300_NEW_VERTEX_FORMAT 0x04000000 #define R300_NEW_VERTEX_SHADER 0x08000000 #define R300_NEW_VERTEX_SHADER_CONSTANTS 0x10000000 -#define R300_NEW_VIEWPORT 0x20000000 #define R300_NEW_QUERY 0x40000000 #define R300_NEW_KITCHEN_SINK 0x7fffffff @@ -178,6 +188,12 @@ struct r300_query { struct r300_query* next; }; +enum r300_buffer_tiling { + R300_BUFFER_LINEAR = 0, + R300_BUFFER_TILED, + R300_BUFFER_SQUARETILED +}; + struct r300_texture { /* Parent class */ struct pipe_texture tex; @@ -214,16 +230,14 @@ struct r300_texture { /* Registers carrying texture format data. */ struct r300_texture_state state; + + /* Buffer tiling */ + enum r300_buffer_tiling microtile, macrotile; }; struct r300_vertex_info { /* Parent class */ struct vertex_info vinfo; - /* Map of vertex attributes into PVS memory for HW TCL, - * or GA memory for SW TCL. */ - int vs_tab[16]; - /* Map of rasterizer attributes from GB through RS to US. */ - int fs_tab[16]; /* R300_VAP_PROG_STREAK_CNTL_[0-7] */ uint32_t vap_prog_stream_cntl[8]; @@ -238,9 +252,11 @@ struct r300_context { struct pipe_context context; /* The interface to the windowing system, etc. */ - struct r300_winsys* winsys; + struct radeon_winsys* winsys; /* Draw module. Used mostly for SW TCL. */ struct draw_context* draw; + /* Accelerated blit support. */ + struct blitter_context* blitter; /* Vertex buffer for rendering. */ struct pipe_buffer* vbo; @@ -260,38 +276,40 @@ struct r300_context { struct r300_vertex_info* vertex_info; /* Various CSO state objects. */ + /* Beginning of atom list. */ + struct r300_atom atom_list; /* Blend state. */ - struct r300_blend_state* blend_state; + struct r300_atom blend_state; /* Blend color state. */ - struct r300_blend_color_state* blend_color_state; + struct r300_atom blend_color_state; /* User clip planes. */ - struct pipe_clip_state clip_state; + struct r300_atom clip_state; /* Shader constants. */ struct r300_constant_buffer shader_constants[PIPE_SHADER_TYPES]; /* Depth, stencil, and alpha state. */ - struct r300_dsa_state* dsa_state; + struct r300_atom dsa_state; /* Fragment shader. */ struct r300_fragment_shader* fs; /* Framebuffer state. We currently don't need our own version of this. */ struct pipe_framebuffer_state framebuffer_state; /* Rasterizer state. */ - struct r300_rs_state* rs_state; + struct r300_atom rs_state; /* RS block state. */ struct r300_rs_block* rs_block; /* Sampler states. */ struct r300_sampler_state* sampler_states[8]; int sampler_count; /* Scissor state. */ - struct r300_scissor_state* scissor_state; + struct r300_atom scissor_state; /* Texture states. */ struct r300_texture* textures[8]; int texture_count; /* Vertex shader. */ struct r300_vertex_shader* vs; /* Viewport state. */ - struct r300_viewport_state* viewport_state; + struct r300_atom viewport_state; /* ZTOP state. */ - struct r300_ztop_state ztop_state; + struct r300_atom ztop_state; /* Vertex buffers for Gallium. */ struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS]; @@ -304,6 +322,8 @@ struct r300_context { uint32_t dirty_state; /* Flag indicating whether or not the HW is dirty. */ uint32_t dirty_hw; + /* Whether the TCL engine should be in bypass mode. */ + boolean tcl_bypass; /** Combination of DBG_xxx flags */ unsigned debug; diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h index 5342488d0d..151f72b0fe 100644 --- a/src/gallium/drivers/r300/r300_cs.h +++ b/src/gallium/drivers/r300/r300_cs.h @@ -26,7 +26,8 @@ #include "util/u_math.h" #include "r300_reg.h" -#include "r300_winsys.h" + +#include "radeon_winsys.h" /* Yes, I know macros are ugly. However, they are much prettier than the code * that they neatly hide away, and don't have the cost of function setup,so @@ -50,11 +51,11 @@ #define CS_LOCALS(context) \ struct r300_context* const cs_context_copy = (context); \ - struct r300_winsys* cs_winsys = cs_context_copy->winsys; \ - int cs_count = 0 + struct radeon_winsys* cs_winsys = cs_context_copy->winsys; \ + int cs_count = 0; (void) cs_count; #define CHECK_CS(size) \ - cs_winsys->check_cs(cs_winsys, (size)) + assert(cs_winsys->check_cs(cs_winsys, (size))) #define BEGIN_CS(size) do { \ CHECK_CS(size); \ @@ -114,6 +115,15 @@ cs_count -= 3; \ } while (0) +#define OUT_CS_RELOC_NO_OFFSET(bo, rd, wd, flags) do { \ + DBG(cs_context_copy, DBG_CS, "r300: writing relocation for buffer %p, " \ + "domains (%d, %d, %d)\n", \ + bo, rd, wd, flags); \ + assert(bo); \ + cs_winsys->write_cs_reloc(cs_winsys, bo, rd, wd, flags); \ + cs_count -= 2; \ +} while (0) + #define END_CS do { \ if (VERY_VERBOSE_CS) { \ DBG(cs_context_copy, DBG_CS, "r300: END_CS in %s (%s:%d)\n", __FUNCTION__, \ diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index b44c7bdbb3..9f93327e59 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -22,7 +23,9 @@ /* r300_emit: Functions for emitting state. */ +#include "util/u_format.h" #include "util/u_math.h" +#include "util/u_simple_list.h" #include "r300_context.h" #include "r300_cs.h" @@ -34,23 +37,31 @@ #include "r300_texture.h" #include "r300_vs.h" -void r300_emit_blend_state(struct r300_context* r300, - struct r300_blend_state* blend) +void r300_emit_blend_state(struct r300_context* r300, void* state) { + struct r300_blend_state* blend = (struct r300_blend_state*)state; CS_LOCALS(r300); + BEGIN_CS(8); - OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3); - OUT_CS(blend->blend_control); - OUT_CS(blend->alpha_blend_control); - OUT_CS(blend->color_channel_mask); OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop); + OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3); + if (r300->framebuffer_state.nr_cbufs) { + OUT_CS(blend->blend_control); + OUT_CS(blend->alpha_blend_control); + OUT_CS(blend->color_channel_mask); + } else { + OUT_CS(0); + OUT_CS(0); + OUT_CS(0); + /* XXX also disable fastfill here once it's supported */ + } OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither); END_CS; } -void r300_emit_blend_color_state(struct r300_context* r300, - struct r300_blend_color_state* bc) +void r300_emit_blend_color_state(struct r300_context* r300, void* state) { + struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state; struct r300_screen* r300screen = r300_screen(r300->context.screen); CS_LOCALS(r300); @@ -67,9 +78,9 @@ void r300_emit_blend_color_state(struct r300_context* r300, } } -void r300_emit_clip_state(struct r300_context* r300, - struct pipe_clip_state* clip) +void r300_emit_clip_state(struct r300_context* r300, void* state) { + struct pipe_clip_state* clip = (struct pipe_clip_state*)state; int i; struct r300_screen* r300screen = r300_screen(r300->context.screen); CS_LOCALS(r300); @@ -97,13 +108,13 @@ void r300_emit_clip_state(struct r300_context* r300, } -void r300_emit_dsa_state(struct r300_context* r300, - struct r300_dsa_state* dsa) +void r300_emit_dsa_state(struct r300_context* r300, void* state) { + struct r300_dsa_state* dsa = (struct r300_dsa_state*)state; struct r300_screen* r300screen = r300_screen(r300->context.screen); CS_LOCALS(r300); - BEGIN_CS(r300screen->caps->is_r500 ? 10 : 8); + BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6); OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function); /* not needed since we use the 8bit alpha ref */ @@ -112,10 +123,16 @@ void r300_emit_dsa_state(struct r300_context* r300, }*/ OUT_CS_REG_SEQ(R300_ZB_CNTL, 3); - OUT_CS(dsa->z_buffer_control); - OUT_CS(dsa->z_stencil_control); + + if (r300->framebuffer_state.zsbuf) { + OUT_CS(dsa->z_buffer_control); + OUT_CS(dsa->z_stencil_control); + } else { + OUT_CS(0); + OUT_CS(0); + } + OUT_CS(dsa->stencil_ref_mask); - OUT_CS_REG(R300_ZB_ZTOP, r300->ztop_state.z_buffer_top); /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */ if (r300screen->caps->is_r500) { @@ -129,7 +146,11 @@ static const float * get_shader_constant( struct rc_constant * constant, struct r300_constant_buffer * externals) { - static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 }; + struct r300_viewport_state* viewport = + (struct r300_viewport_state*)r300->viewport_state.state; + static float vec[4] = { 0.0, 0.0, 0.0, 1.0 }; + struct pipe_texture *tex; + switch(constant->Type) { case RC_CONSTANT_EXTERNAL: return externals->constants[constant->u.External]; @@ -137,11 +158,58 @@ static const float * get_shader_constant( case RC_CONSTANT_IMMEDIATE: return constant->u.Immediate; + case RC_CONSTANT_STATE: + switch (constant->u.State[0]) { + /* Factor for converting rectangle coords to + * normalized coords. Should only show up on non-r500. */ + case RC_STATE_R300_TEXRECT_FACTOR: + tex = &r300->textures[constant->u.State[1]]->tex; + vec[0] = 1.0 / tex->width0; + vec[1] = 1.0 / tex->height0; + break; + + /* Texture compare-fail value. */ + /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient, + * this is always (0,0,0,0), right? */ + case RC_STATE_SHADOW_AMBIENT: + vec[3] = 0; + break; + + case RC_STATE_R300_VIEWPORT_SCALE: + if (r300->tcl_bypass) { + vec[0] = 1; + vec[1] = 1; + vec[2] = 1; + } else { + vec[0] = viewport->xscale; + vec[1] = viewport->yscale; + vec[2] = viewport->zscale; + } + break; + + case RC_STATE_R300_VIEWPORT_OFFSET: + if (!r300->tcl_bypass) { + vec[0] = viewport->xoffset; + vec[1] = viewport->yoffset; + vec[2] = viewport->zoffset; + } + break; + + default: + debug_printf("r300: Implementation error: " + "Unknown RC_CONSTANT type %d\n", constant->u.State[0]); + } + break; + default: - debug_printf("r300: Implementation error: Unhandled constant type %i\n", - constant->Type); - return zero; + debug_printf("r300: Implementation error: " + "Unhandled constant type %d\n", constant->Type); } + + /* This should either be (0, 0, 0, 1), which should be a relatively safe + * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE + * state factors. */ + return vec; } /* Convert a normal single-precision float into the 7.16 format @@ -245,6 +313,22 @@ void r300_emit_fs_constant_buffer(struct r300_context* r300, END_CS; } +static void r300_emit_fragment_depth_config(struct r300_context* r300, + struct r300_fragment_shader* fs) +{ + CS_LOCALS(r300); + + BEGIN_CS(4); + if (r300_fragment_shader_writes_depth(fs)) { + OUT_CS_REG(R300_FG_DEPTH_SRC, R300_FG_DEPTH_SRC_SHADER); + OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W24 | R300_W_SRC_US); + } else { + OUT_CS_REG(R300_FG_DEPTH_SRC, R300_FG_DEPTH_SRC_SCAN); + OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W0 | R300_W_SRC_US); + } + END_CS; +} + void r500_emit_fragment_program_code(struct r300_context* r300, struct rX00_fragment_program_code* generic_code) { @@ -254,7 +338,7 @@ void r500_emit_fragment_program_code(struct r300_context* r300, BEGIN_CS(13 + ((code->inst_end + 1) * 6)); - OUT_CS_REG(R500_US_CONFIG, 0); + OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx); OUT_CS_REG(R500_US_CODE_RANGE, R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); @@ -308,7 +392,13 @@ void r300_emit_fb_state(struct r300_context* r300, int i; CS_LOCALS(r300); - BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 4); + /* Shouldn't fail unless there is a bug in the state tracker. */ + assert(fb->nr_cbufs <= 4); + + BEGIN_CS((10 * fb->nr_cbufs) + (2 * (4 - fb->nr_cbufs)) + + (fb->zsbuf ? 10 : 0) + 6); + + /* Flush and free renderbuffer caches. */ OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT, R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS | R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D); @@ -316,6 +406,10 @@ void r300_emit_fb_state(struct r300_context* r300, R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE | R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE); + /* Set the number of colorbuffers. */ + OUT_CS_REG(R300_RB3D_CCTL, R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs)); + + /* Set up colorbuffers. */ for (i = 0; i < fb->nr_cbufs; i++) { surf = fb->cbufs[i]; tex = (struct r300_texture*)surf->texture; @@ -326,13 +420,21 @@ void r300_emit_fb_state(struct r300_context* r300, OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1); OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] | - r300_translate_colorformat(tex->tex.format), 0, - RADEON_GEM_DOMAIN_VRAM, 0); + r300_translate_colorformat(tex->tex.format) | + R300_COLOR_TILE(tex->macrotile) | + R300_COLOR_MICROTILE(tex->microtile), + 0, RADEON_GEM_DOMAIN_VRAM, 0); OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), r300_translate_out_fmt(surf->format)); } + /* Disable unused colorbuffers. */ + for (; i < 4; i++) { + OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED); + } + + /* Set up a zbuffer. */ if (fb->zsbuf) { surf = fb->zsbuf; tex = (struct r300_texture*)surf->texture; @@ -344,8 +446,10 @@ void r300_emit_fb_state(struct r300_context* r300, OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format)); OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1); - OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level], 0, - RADEON_GEM_DOMAIN_VRAM, 0); + OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] | + R300_DEPTHMACROTILE(tex->macrotile) | + R300_DEPTHMICROTILE(tex->microtile), + 0, RADEON_GEM_DOMAIN_VRAM, 0); } END_CS; @@ -360,8 +464,6 @@ static void r300_emit_query_start(struct r300_context *r300) if (!query) return; - /* XXX This will almost certainly not return good results - * for overlapping queries. */ BEGIN_CS(4); if (caps->family == CHIP_FAMILY_RV530) { OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL); @@ -479,8 +581,9 @@ void r300_emit_query_end(struct r300_context* r300) r300_emit_query_finish(r300, query); } -void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs) +void r300_emit_rs_state(struct r300_context* r300, void* state) { + struct r300_rs_state* rs = (struct r300_rs_state*)state; CS_LOCALS(r300); BEGIN_CS(22); @@ -543,15 +646,46 @@ void r300_emit_rs_block_state(struct r300_context* r300, END_CS; } -void r300_emit_scissor_state(struct r300_context* r300, - struct r300_scissor_state* scissor) +void r300_emit_scissor_state(struct r300_context* r300, void* state) { + unsigned minx, miny, maxx, maxy; + uint32_t top_left, bottom_right; + struct r300_screen* r300screen = r300_screen(r300->context.screen); + struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state; CS_LOCALS(r300); + minx = miny = 0; + maxx = r300->framebuffer_state.width; + maxy = r300->framebuffer_state.height; + + if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) { + minx = MAX2(minx, scissor->minx); + miny = MAX2(miny, scissor->miny); + maxx = MIN2(maxx, scissor->maxx); + maxy = MIN2(maxy, scissor->maxy); + } + + if (r300screen->caps->is_r500) { + top_left = + (minx << R300_SCISSORS_X_SHIFT) | + (miny << R300_SCISSORS_Y_SHIFT); + bottom_right = + ((maxx - 1) << R300_SCISSORS_X_SHIFT) | + ((maxy - 1) << R300_SCISSORS_Y_SHIFT); + } else { + /* Offset of 1440 in non-R500 chipsets. */ + top_left = + ((minx + 1440) << R300_SCISSORS_X_SHIFT) | + ((miny + 1440) << R300_SCISSORS_Y_SHIFT); + bottom_right = + (((maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) | + (((maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT); + } + BEGIN_CS(3); OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); - OUT_CS(scissor->scissor_top_left); - OUT_CS(scissor->scissor_bottom_right); + OUT_CS(top_left); + OUT_CS(bottom_right); END_CS; } @@ -561,6 +695,8 @@ void r300_emit_texture(struct r300_context* r300, unsigned offset) { uint32_t filter0 = sampler->filter0; + uint32_t format0 = tex->state.format0; + unsigned min_level, max_level; CS_LOCALS(r300); /* to emulate 1D textures through 2D ones correctly */ @@ -569,90 +705,91 @@ void r300_emit_texture(struct r300_context* r300, filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE); } + /* determine min/max levels */ + /* the MAX_MIP level is the largest (finest) one */ + max_level = MIN2(sampler->max_lod, tex->tex.last_level); + min_level = MIN2(sampler->min_lod, max_level); + format0 |= R300_TX_NUM_LEVELS(max_level); + filter0 |= R300_TX_MAX_MIP_LEVEL(min_level); + BEGIN_CS(16); OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 | (offset << 28)); OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1); OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color); - OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), tex->state.format0); + OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0); OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1); OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2); OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1); - OUT_CS_RELOC(tex->buffer, 0, - RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0); + OUT_CS_RELOC(tex->buffer, + R300_TXO_MACRO_TILE(tex->macrotile) | + R300_TXO_MICRO_TILE(tex->microtile), + RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0); END_CS; } -/* XXX I can't read this and that's not good */ -void r300_emit_aos(struct r300_context* r300, unsigned offset) +static boolean r300_validate_aos(struct r300_context *r300) { struct pipe_vertex_buffer *vbuf = r300->vertex_buffer; struct pipe_vertex_element *velem = r300->vertex_element; - CS_LOCALS(r300); int i; - unsigned aos_count = r300->vertex_element_count; + /* Check if formats and strides are aligned to the size of DWORD. */ + for (i = 0; i < r300->vertex_element_count; i++) { + if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 || + util_format_get_blocksize(velem[i].src_format) % 4 != 0) { + return FALSE; + } + } + return TRUE; +} + +void r300_emit_aos(struct r300_context* r300, unsigned offset) +{ + struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer; + struct pipe_vertex_element *velem = r300->vertex_element; + int i; + unsigned size1, size2, aos_count = r300->vertex_element_count; unsigned packet_size = (aos_count * 3 + 1) / 2; + CS_LOCALS(r300); + + /* XXX Move this checking to a more approriate place. */ + if (!r300_validate_aos(r300)) { + /* XXX We should fallback using Draw. */ + assert(0); + } + BEGIN_CS(2 + packet_size + aos_count * 2); OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size); OUT_CS(aos_count); + for (i = 0; i < aos_count - 1; i += 2) { - int buf_num1 = velem[i].vertex_buffer_index; - int buf_num2 = velem[i+1].vertex_buffer_index; - assert(vbuf[buf_num1].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0); - assert(vbuf[buf_num2].stride % 4 == 0 && pf_get_size(velem[i+1].src_format) % 4 == 0); - OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num1].stride << 6) | - (pf_get_size(velem[i+1].src_format) << 14) | (vbuf[buf_num2].stride << 22)); - OUT_CS(vbuf[buf_num1].buffer_offset + velem[i].src_offset + - offset * vbuf[buf_num1].stride); - OUT_CS(vbuf[buf_num2].buffer_offset + velem[i+1].src_offset + - offset * vbuf[buf_num2].stride); + vb1 = &vbuf[velem[i].vertex_buffer_index]; + vb2 = &vbuf[velem[i+1].vertex_buffer_index]; + size1 = util_format_get_blocksize(velem[i].src_format); + size2 = util_format_get_blocksize(velem[i+1].src_format); + + OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) | + R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride)); + OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride); + OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride); } + if (aos_count & 1) { - int buf_num = velem[i].vertex_buffer_index; - assert(vbuf[buf_num].stride % 4 == 0 && pf_get_size(velem[i].src_format) % 4 == 0); - OUT_CS((pf_get_size(velem[i].src_format) >> 2) | (vbuf[buf_num].stride << 6)); - OUT_CS(vbuf[buf_num].buffer_offset + velem[i].src_offset + - offset * vbuf[buf_num].stride); + vb1 = &vbuf[velem[i].vertex_buffer_index]; + size1 = util_format_get_blocksize(velem[i].src_format); + + OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride)); + OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride); } - /* XXX bare CS reloc */ for (i = 0; i < aos_count; i++) { - cs_winsys->write_cs_reloc(cs_winsys, - vbuf[velem[i].vertex_buffer_index].buffer, - RADEON_GEM_DOMAIN_GTT, - 0, - 0); - cs_count -= 2; + OUT_CS_RELOC_NO_OFFSET(vbuf[velem[i].vertex_buffer_index].buffer, + RADEON_GEM_DOMAIN_GTT, 0, 0); } END_CS; } -#if 0 -void r300_emit_draw_packet(struct r300_context* r300) -{ - CS_LOCALS(r300); - - DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, " - "vertex size %d\n", r300->vbo, - r300->vertex_info->vinfo.size); - /* Set the pointer to our vertex buffer. The emitted values are this: - * PACKET3 [3D_LOAD_VBPNTR] - * COUNT [1] - * FORMAT [size | stride << 8] - * OFFSET [offset into BO] - * VBPNTR [relocated BO] - */ - BEGIN_CS(7); - OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3); - OUT_CS(1); - OUT_CS(r300->vertex_info->vinfo.size | - (r300->vertex_info->vinfo.size << 8)); - OUT_CS(r300->vbo_offset); - OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0); - END_CS; -} -#endif void r300_emit_vertex_format_state(struct r300_context* r300) { @@ -690,12 +827,22 @@ void r300_emit_vertex_format_state(struct r300_context* r300) END_CS; } + void r300_emit_vertex_program_code(struct r300_context* r300, struct r300_vertex_program_code* code) { int i; struct r300_screen* r300screen = r300_screen(r300->context.screen); unsigned instruction_count = code->length / 4; + + int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72; + int input_count = MAX2(util_bitcount(code->InputsRead), 1); + int output_count = MAX2(util_bitcount(code->OutputsWritten), 1); + int temp_count = MAX2(code->num_temporaries, 1); + int pvs_num_slots = MIN3(vtx_mem_size / input_count, + vtx_mem_size / output_count, 10); + int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6); + CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { @@ -708,8 +855,7 @@ void r300_emit_vertex_program_code(struct r300_context* r300, /* R300_VAP_PVS_CODE_CNTL_0 * R300_VAP_PVS_CONST_CNTL * R300_VAP_PVS_CODE_CNTL_1 - * See the r5xx docs for instructions on how to use these. - * XXX these could be optimized to select better values... */ + * See the r5xx docs for instructions on how to use these. */ OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3); OUT_CS(R300_PVS_FIRST_INST(0) | R300_PVS_XYZW_VALID_INST(instruction_count - 1) | @@ -722,10 +868,11 @@ void r300_emit_vertex_program_code(struct r300_context* r300, for (i = 0; i < code->length; i++) OUT_CS(code->body.d[i]); - OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(10) | - R300_PVS_NUM_CNTLRS(5) | + OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) | + R300_PVS_NUM_CNTLRS(pvs_num_controllers) | R300_PVS_NUM_FPUS(r300screen->caps->num_vert_fpus) | - R300_PVS_VF_MAX_VTX_NUM(12)); + R300_PVS_VF_MAX_VTX_NUM(12) | + (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0)); END_CS; } @@ -768,25 +915,57 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300, END_CS; } -void r300_emit_viewport_state(struct r300_context* r300, - struct r300_viewport_state* viewport) +void r300_emit_viewport_state(struct r300_context* r300, void* state) { + struct r300_viewport_state* viewport = (struct r300_viewport_state*)state; CS_LOCALS(r300); - BEGIN_CS(9); - OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); - OUT_CS_32F(viewport->xscale); - OUT_CS_32F(viewport->xoffset); - OUT_CS_32F(viewport->yscale); - OUT_CS_32F(viewport->yoffset); - OUT_CS_32F(viewport->zscale); - OUT_CS_32F(viewport->zoffset); - - if (r300->rs_state->enable_vte) { - OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); - } else { + if (r300->tcl_bypass) { + BEGIN_CS(2); OUT_CS_REG(R300_VAP_VTE_CNTL, 0); + END_CS; + } else { + BEGIN_CS(9); + OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6); + OUT_CS_32F(viewport->xscale); + OUT_CS_32F(viewport->xoffset); + OUT_CS_32F(viewport->yscale); + OUT_CS_32F(viewport->yoffset); + OUT_CS_32F(viewport->zscale); + OUT_CS_32F(viewport->zoffset); + OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control); + END_CS; + } +} + +void r300_emit_texture_count(struct r300_context* r300) +{ + uint32_t tx_enable = 0; + int i; + CS_LOCALS(r300); + + /* Notice that texture_count and sampler_count are just sizes + * of the respective arrays. We still have to check for the individual + * elements. */ + for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) { + if (r300->textures[i]) { + tx_enable |= 1 << i; + } } + + BEGIN_CS(2); + OUT_CS_REG(R300_TX_ENABLE, tx_enable); + END_CS; + +} + +void r300_emit_ztop_state(struct r300_context* r300, void* state) +{ + struct r300_ztop_state* ztop = (struct r300_ztop_state*)state; + CS_LOCALS(r300); + + BEGIN_CS(2); + OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top); END_CS; } @@ -794,9 +973,8 @@ void r300_flush_textures(struct r300_context* r300) { CS_LOCALS(r300); - BEGIN_CS(4); + BEGIN_CS(2); OUT_CS_REG(R300_TX_INVALTAGS, 0); - OUT_CS_REG(R300_TX_ENABLE, (1 << r300->texture_count) - 1); END_CS; } @@ -814,17 +992,30 @@ void r300_emit_dirty_state(struct r300_context* r300) { struct r300_screen* r300screen = r300_screen(r300->context.screen); struct r300_texture* tex; - int i, dirty_tex = 0; + struct r300_atom* atom; + unsigned i, dwords = 1024; + int dirty_tex = 0; boolean invalid = FALSE; - if (!(r300->dirty_state)) { - return; + /* Check the required number of dwords against the space remaining in the + * current CS object. If we need more, then flush. */ + + foreach(atom, &r300->atom_list) { + if (atom->dirty || atom->always_dirty) { + dwords += atom->size; + } + } + + /* Make sure we have at least 2*1024 spare dwords. */ + /* XXX It would be nice to know the number of dwords we really need to + * XXX emit. */ + if (!r300->winsys->check_cs(r300->winsys, dwords)) { + r300->context.flush(&r300->context, 0, NULL); } /* Clean out BOs. */ r300->winsys->reset_bos(r300->winsys); - /* XXX check size */ validate: /* Color buffers... */ for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) { @@ -858,10 +1049,12 @@ validate: } } /* ...occlusion query buffer... */ - if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo, - 0, RADEON_GEM_DOMAIN_GTT)) { - r300->context.flush(&r300->context, 0, NULL); - goto validate; + if (r300->dirty_state & R300_NEW_QUERY) { + if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo, + 0, RADEON_GEM_DOMAIN_GTT)) { + r300->context.flush(&r300->context, 0, NULL); + goto validate; + } } /* ...and vertex buffer. */ if (r300->vbo) { @@ -889,40 +1082,30 @@ validate: r300->dirty_state &= ~R300_NEW_QUERY; } - if (r300->dirty_state & R300_NEW_BLEND) { - r300_emit_blend_state(r300, r300->blend_state); - r300->dirty_state &= ~R300_NEW_BLEND; - } - - if (r300->dirty_state & R300_NEW_BLEND_COLOR) { - r300_emit_blend_color_state(r300, r300->blend_color_state); - r300->dirty_state &= ~R300_NEW_BLEND_COLOR; - } - - if (r300->dirty_state & R300_NEW_CLIP) { - r300_emit_clip_state(r300, &r300->clip_state); - r300->dirty_state &= ~R300_NEW_CLIP; - } - - if (r300->dirty_state & R300_NEW_DSA) { - r300_emit_dsa_state(r300, r300->dsa_state); - r300->dirty_state &= ~R300_NEW_DSA; + foreach(atom, &r300->atom_list) { + if (atom->dirty || atom->always_dirty) { + atom->emit(r300, atom->state); + atom->dirty = FALSE; + } } if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) { + r300_emit_fragment_depth_config(r300, r300->fs); if (r300screen->caps->is_r500) { - r500_emit_fragment_program_code(r300, &r300->fs->code); + r500_emit_fragment_program_code(r300, &r300->fs->shader->code); } else { - r300_emit_fragment_program_code(r300, &r300->fs->code); + r300_emit_fragment_program_code(r300, &r300->fs->shader->code); } r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER; } if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER_CONSTANTS) { if (r300screen->caps->is_r500) { - r500_emit_fs_constant_buffer(r300, &r300->fs->code.constants); + r500_emit_fs_constant_buffer(r300, + &r300->fs->shader->code.constants); } else { - r300_emit_fs_constant_buffer(r300, &r300->fs->code.constants); + r300_emit_fs_constant_buffer(r300, + &r300->fs->shader->code.constants); } r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS; } @@ -932,24 +1115,16 @@ validate: r300->dirty_state &= ~R300_NEW_FRAMEBUFFERS; } - if (r300->dirty_state & R300_NEW_RASTERIZER) { - r300_emit_rs_state(r300, r300->rs_state); - r300->dirty_state &= ~R300_NEW_RASTERIZER; - } - if (r300->dirty_state & R300_NEW_RS_BLOCK) { r300_emit_rs_block_state(r300, r300->rs_block); r300->dirty_state &= ~R300_NEW_RS_BLOCK; } - if (r300->dirty_state & R300_NEW_SCISSOR) { - r300_emit_scissor_state(r300, r300->scissor_state); - r300->dirty_state &= ~R300_NEW_SCISSOR; - } - /* Samplers and textures are tracked separately but emitted together. */ if (r300->dirty_state & (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) { + r300_emit_texture_count(r300); + for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) { if (r300->dirty_state & ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) { @@ -966,11 +1141,6 @@ validate: r300->dirty_state &= ~(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES); } - if (r300->dirty_state & R300_NEW_VIEWPORT) { - r300_emit_viewport_state(r300, r300->viewport_state); - r300->dirty_state &= ~R300_NEW_VIEWPORT; - } - if (dirty_tex) { r300_flush_textures(r300); } diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h index 7c83c5166d..05a6bfeae8 100644 --- a/src/gallium/drivers/r300/r300_emit.h +++ b/src/gallium/drivers/r300/r300_emit.h @@ -31,17 +31,13 @@ struct r300_vertex_program_code; void r300_emit_aos(struct r300_context* r300, unsigned offset); -void r300_emit_blend_state(struct r300_context* r300, - struct r300_blend_state* blend); +void r300_emit_blend_state(struct r300_context* r300, void* state); -void r300_emit_blend_color_state(struct r300_context* r300, - struct r300_blend_color_state* bc); +void r300_emit_blend_color_state(struct r300_context* r300, void* state); -void r300_emit_clip_state(struct r300_context* r300, - struct pipe_clip_state* clip); +void r300_emit_clip_state(struct r300_context* r300, void* state); -void r300_emit_dsa_state(struct r300_context* r300, - struct r300_dsa_state* dsa); +void r300_emit_dsa_state(struct r300_context* r300, void* state); void r300_emit_fragment_program_code(struct r300_context* r300, struct rX00_fragment_program_code* generic_code); @@ -63,13 +59,12 @@ void r300_emit_query_begin(struct r300_context* r300, void r300_emit_query_end(struct r300_context* r300); -void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs); +void r300_emit_rs_state(struct r300_context* r300, void* state); void r300_emit_rs_block_state(struct r300_context* r300, struct r300_rs_block* rs); -void r300_emit_scissor_state(struct r300_context* r300, - struct r300_scissor_state* scissor); +void r300_emit_scissor_state(struct r300_context* r300, void* state); void r300_emit_texture(struct r300_context* r300, struct r300_sampler_state* sampler, @@ -89,8 +84,11 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300, void r300_emit_vertex_shader(struct r300_context* r300, struct r300_vertex_shader* vs); -void r300_emit_viewport_state(struct r300_context* r300, - struct r300_viewport_state* viewport); +void r300_emit_viewport_state(struct r300_context* r300, void* state); + +void r300_emit_texture_count(struct r300_context* r300); + +void r300_emit_ztop_state(struct r300_context* r300, void* state); void r300_flush_textures(struct r300_context* r300); diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c index 4282357521..59819cb106 100644 --- a/src/gallium/drivers/r300/r300_flush.c +++ b/src/gallium/drivers/r300/r300_flush.c @@ -37,6 +37,7 @@ static void r300_flush(struct pipe_context* pipe, { struct r300_context *r300 = r300_context(pipe); struct r300_query *query; + struct r300_atom *atom; CS_LOCALS(r300); (void) cs_count; @@ -55,7 +56,15 @@ static void r300_flush(struct pipe_context* pipe, r300_emit_invariant_state(r300); r300->dirty_state = R300_NEW_KITCHEN_SINK; r300->dirty_hw = 0; + + /* New kitchen sink, baby. */ + foreach(atom, &r300->atom_list) { + if (atom->state) { + atom->dirty = TRUE; + } + } } + /* reset flushed query */ foreach(query, &r300->query_list) { query->flushed = TRUE; diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c index 29ddc84c41..60ea9c171d 100644 --- a/src/gallium/drivers/r300/r300_fs.c +++ b/src/gallium/drivers/r300/r300_fs.c @@ -1,6 +1,7 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> * Joakim Sindholt <opensource@zhasha.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -21,6 +22,9 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ +#include "util/u_math.h" +#include "util/u_memory.h" + #include "tgsi/tgsi_dump.h" #include "r300_context.h" @@ -31,6 +35,45 @@ #include "radeon_code.h" #include "radeon_compiler.h" +/* Convert info about FS input semantics to r300_shader_semantics. */ +void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, + struct r300_shader_semantics* fs_inputs) +{ + int i; + unsigned index; + + r300_shader_semantics_reset(fs_inputs); + + for (i = 0; i < info->num_inputs; i++) { + index = info->input_semantic_index[i]; + + switch (info->input_semantic_name[i]) { + case TGSI_SEMANTIC_COLOR: + assert(index <= ATTR_COLOR_COUNT); + fs_inputs->color[index] = i; + break; + + case TGSI_SEMANTIC_GENERIC: + assert(index <= ATTR_GENERIC_COUNT); + fs_inputs->generic[index] = i; + break; + + case TGSI_SEMANTIC_FOG: + assert(index == 0); + fs_inputs->fog = i; + break; + + case TGSI_SEMANTIC_POSITION: + assert(index == 0); + fs_inputs->wpos = i; + break; + + default: + assert(0); + } + } +} + static void find_output_registers(struct r300_fragment_program_compiler * compiler, struct r300_fragment_shader * fs) { @@ -58,61 +101,69 @@ static void allocate_hardware_inputs( void (*allocate)(void * data, unsigned input, unsigned hwreg), void * mydata) { - struct tgsi_shader_info* info = &((struct r300_fragment_shader*)c->UserData)->info; - int total_colors = 0; - int colors = 0; - int total_generic = 0; - int generic = 0; - int i; - - for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - total_colors++; - break; - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - total_generic++; - break; + struct r300_shader_semantics* inputs = + (struct r300_shader_semantics*)c->UserData; + int i, reg = 0; + + /* Allocate input registers. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (inputs->color[i] != ATTR_UNUSED) { + allocate(mydata, inputs->color[i], reg++); + } + } + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (inputs->generic[i] != ATTR_UNUSED) { + allocate(mydata, inputs->generic[i], reg++); } } + if (inputs->fog != ATTR_UNUSED) { + allocate(mydata, inputs->fog, reg++); + } + if (inputs->wpos != ATTR_UNUSED) { + allocate(mydata, inputs->wpos, reg++); + } +} - for(i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - allocate(mydata, i, colors); - colors++; - break; - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - allocate(mydata, i, total_colors + generic); - generic++; - break; +static void get_compare_state( + struct r300_context* r300, + struct r300_fragment_program_external_state* state, + unsigned shadow_samplers) +{ + memset(state, 0, sizeof(*state)); + + for (int i = 0; i < r300->sampler_count; i++) { + struct r300_sampler_state* s = r300->sampler_states[i]; + + if (s && s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { + /* XXX Gallium doesn't provide us with any information regarding + * this mode, so we are screwed. I'm setting 0 = LUMINANCE. */ + state->unit[i].depth_texture_mode = 0; + + /* Fortunately, no need to translate this. */ + state->unit[i].texture_compare_func = s->state.compare_func; } } } -void r300_translate_fragment_shader(struct r300_context* r300, - struct r300_fragment_shader* fs) +static void r300_translate_fragment_shader( + struct r300_context* r300, + struct r300_fragment_shader_code* shader) { + struct r300_fragment_shader* fs = r300->fs; struct r300_fragment_program_compiler compiler; struct tgsi_to_rc ttr; + int wpos = fs->inputs.wpos; + /* Setup the compiler. */ memset(&compiler, 0, sizeof(compiler)); rc_init(&compiler.Base); compiler.Base.Debug = DBG_ON(r300, DBG_FP); - compiler.code = &fs->code; + compiler.code = &shader->code; + compiler.state = shader->compare_state; compiler.is_r500 = r300_screen(r300->context.screen)->caps->is_r500; compiler.AllocateHwInputs = &allocate_hardware_inputs; - compiler.UserData = fs; - - /* TODO: Program compilation depends on texture compare modes, - * which are sampler state. Therefore, programs need to be recompiled - * depending on this state as in the classic Mesa driver. - * - * This is not yet handled correctly. - */ + compiler.UserData = &fs->inputs; find_output_registers(&compiler, fs); @@ -127,15 +178,76 @@ void r300_translate_fragment_shader(struct r300_context* r300, r300_tgsi_to_rc(&ttr, fs->state.tokens); + fs->shadow_samplers = compiler.Base.Program.ShadowSamplers; + + /** + * Transform the program to support WPOS. + * + * Introduce a small fragment at the start of the program that will be + * the only code that directly reads the WPOS input. + * All other code pieces that reference that input will be rewritten + * to read from a newly allocated temporary. */ + if (wpos != ATTR_UNUSED) { + /* Moving the input to some other reg is not really necessary. */ + rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE); + } + /* Invoke the compiler */ r3xx_compile_fragment_program(&compiler); if (compiler.Base.Error) { /* XXX failover maybe? */ DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n", compiler.Base.ErrorMsg); + assert(0); } /* And, finally... */ rc_destroy(&compiler.Base); - fs->translated = TRUE; +} + +boolean r300_pick_fragment_shader(struct r300_context* r300) +{ + struct r300_fragment_shader* fs = r300->fs; + struct r300_fragment_program_external_state state; + struct r300_fragment_shader_code* ptr; + + if (!fs->first) { + /* Build the fragment shader for the first time. */ + fs->first = fs->shader = CALLOC_STRUCT(r300_fragment_shader_code); + + /* BTW shadow samplers will be known after the first translation, + * therefore we set ~0, which means it should look at all sampler + * states. This choice doesn't have any impact on the correctness. */ + get_compare_state(r300, &fs->shader->compare_state, ~0); + r300_translate_fragment_shader(r300, fs->shader); + return TRUE; + + } else if (fs->shadow_samplers) { + get_compare_state(r300, &state, fs->shadow_samplers); + + /* Check if the currently-bound shader has been compiled + * with the texture-compare state we need. */ + if (memcmp(&fs->shader->compare_state, &state, sizeof(state)) != 0) { + /* Search for the right shader. */ + ptr = fs->first; + while (ptr) { + if (memcmp(&ptr->compare_state, &state, sizeof(state)) == 0) { + fs->shader = ptr; + return TRUE; + } + ptr = ptr->next; + } + + /* Not found, gotta compile a new one. */ + ptr = CALLOC_STRUCT(r300_fragment_shader_code); + ptr->next = fs->first; + fs->first = fs->shader = ptr; + + ptr->compare_state = state; + r300_translate_fragment_shader(r300, ptr); + return TRUE; + } + } + + return FALSE; } diff --git a/src/gallium/drivers/r300/r300_fs.h b/src/gallium/drivers/r300/r300_fs.h index e831c30301..40ce874353 100644 --- a/src/gallium/drivers/r300/r300_fs.h +++ b/src/gallium/drivers/r300/r300_fs.h @@ -1,6 +1,7 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> * Joakim Sindholt <opensource@zhasha.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -25,31 +26,46 @@ #define R300_FS_H #include "pipe/p_state.h" - #include "tgsi/tgsi_scan.h" - #include "radeon_code.h" +#include "r300_shader_semantics.h" + +struct r300_fragment_shader_code { + struct r300_fragment_program_external_state compare_state; + struct rX00_fragment_program_code code; + + struct r300_fragment_shader_code* next; +}; struct r300_fragment_shader { /* Parent class */ struct pipe_shader_state state; + struct tgsi_shader_info info; + struct r300_shader_semantics inputs; - /* Has this shader been translated yet? */ - boolean translated; + /* Bits 0-15: TRUE if it's a shadow sampler, FALSE otherwise. */ + unsigned shadow_samplers; - /* Compiled code */ - struct rX00_fragment_program_code code; + /* Currently-bound fragment shader. */ + struct r300_fragment_shader_code* shader; + + /* List of the same shaders compiled with different texture-compare + * states. */ + struct r300_fragment_shader_code* first; }; +void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, + struct r300_shader_semantics* fs_inputs); -void r300_translate_fragment_shader(struct r300_context* r300, - struct r300_fragment_shader* fs); +/* Return TRUE if the shader was switched and should be re-emitted. */ +boolean r300_pick_fragment_shader(struct r300_context* r300); -static inline boolean r300_fragment_shader_writes_depth(struct r300_fragment_shader *fs) +static INLINE boolean r300_fragment_shader_writes_depth(struct r300_fragment_shader *fs) { if (!fs) - return FALSE; - return (fs->code.writes_depth) ? TRUE : FALSE; + return FALSE; + return (fs->shader->code.writes_depth) ? TRUE : FALSE; } + #endif /* R300_FS_H */ diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h index 744ea6568d..361813891f 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -661,20 +661,20 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_GB_SUPER_TILE_B (1 << 15) # define R300_GB_SUBPIXEL_1_12 (0 << 16) # define R300_GB_SUBPIXEL_1_16 (1 << 16) -# define GB_TILE_CONFIG_QUADS_PER_RAS_4 (0 << 17) -# define GB_TILE_CONFIG_QUADS_PER_RAS_8 (1 << 17) -# define GB_TILE_CONFIG_QUADS_PER_RAS_16 (2 << 17) -# define GB_TILE_CONFIG_QUADS_PER_RAS_32 (3 << 17) -# define GB_TILE_CONFIG_BB_SCAN_INTERCEPT (0 << 19) -# define GB_TILE_CONFIG_BB_SCAN_BOUND_BOX (1 << 19) -# define GB_TILE_CONFIG_ALT_SCAN_EN_LR (0 << 20) -# define GB_TILE_CONFIG_ALT_SCAN_EN_LRL (1 << 20) -# define GB_TILE_CONFIG_ALT_OFFSET (0 << 21) -# define GB_TILE_CONFIG_SUBPRECISION (0 << 22) -# define GB_TILE_CONFIG_ALT_TILING_DEF (0 << 23) -# define GB_TILE_CONFIG_ALT_TILING_3_2 (1 << 23) -# define GB_TILE_CONFIG_Z_EXTENDED_24_1 (0 << 24) -# define GB_TILE_CONFIG_Z_EXTENDED_S25_1 (1 << 24) +# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_4 (0 << 17) +# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_8 (1 << 17) +# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_16 (2 << 17) +# define R300_GB_TILE_CONFIG_QUADS_PER_RAS_32 (3 << 17) +# define R300_GB_TILE_CONFIG_BB_SCAN_INTERCEPT (0 << 19) +# define R300_GB_TILE_CONFIG_BB_SCAN_BOUND_BOX (1 << 19) +# define R300_GB_TILE_CONFIG_ALT_SCAN_EN_LR (0 << 20) +# define R300_GB_TILE_CONFIG_ALT_SCAN_EN_LRL (1 << 20) +# define R300_GB_TILE_CONFIG_ALT_OFFSET (0 << 21) +# define R300_GB_TILE_CONFIG_SUBPRECISION (0 << 22) +# define R300_GB_TILE_CONFIG_ALT_TILING_DEF (0 << 23) +# define R300_GB_TILE_CONFIG_ALT_TILING_3_2 (1 << 23) +# define R300_GB_TILE_CONFIG_Z_EXTENDED_24_1 (0 << 24) +# define R300_GB_TILE_CONFIG_Z_EXTENDED_S25_1 (1 << 24) /* Specifies the sizes of the various FIFO`s in the sc/rs/us. This register must be the first one written */ #define R300_GB_FIFO_SIZE 0x4024 @@ -700,9 +700,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_OFIFO_HIGHWATER_SHIFT 22 /* two bits only */ # define R300_CUBE_FIFO_HIGHWATER_COL_SHIFT 24 -#define GB_Z_PEQ_CONFIG 0x4028 -# define GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_4_4 (0 << 0) -# define GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8 (1 << 0) +#define R300_GB_Z_PEQ_CONFIG 0x4028 +# define R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_4_4 (0 << 0) +# define R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8 (1 << 0) /* Specifies various polygon specific selects (fog, depth, perspective). */ #define R300_GB_SELECT 0x401c @@ -725,39 +725,39 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* Specifies the graphics pipeline configuration for antialiasing. */ #define R300_GB_AA_CONFIG 0x4020 -# define GB_AA_CONFIG_AA_DISABLE (0 << 0) -# define GB_AA_CONFIG_AA_ENABLE (1 << 0) -# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2 (0 << 1) -# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3 (1 << 1) -# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4 (2 << 1) -# define GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6 (3 << 1) +# define R300_GB_AA_CONFIG_AA_DISABLE (0 << 0) +# define R300_GB_AA_CONFIG_AA_ENABLE (1 << 0) +# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2 (0 << 1) +# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3 (1 << 1) +# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4 (2 << 1) +# define R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6 (3 << 1) /* Selects which of 4 pipes are active. */ -#define GB_PIPE_SELECT 0x402c -# define GB_PIPE_SELECT_PIPE0_ID_SHIFT 0 -# define GB_PIPE_SELECT_PIPE1_ID_SHIFT 2 -# define GB_PIPE_SELECT_PIPE2_ID_SHIFT 4 -# define GB_PIPE_SELECT_PIPE3_ID_SHIFT 6 -# define GB_PIPE_SELECT_PIPE_MASK_SHIFT 8 -# define GB_PIPE_SELECT_MAX_PIPE 12 -# define GB_PIPE_SELECT_BAD_PIPES 14 -# define GB_PIPE_SELECT_CONFIG_PIPES 18 +#define R300_GB_PIPE_SELECT 0x402c +# define R300_GB_PIPE_SELECT_PIPE0_ID_SHIFT 0 +# define R300_GB_PIPE_SELECT_PIPE1_ID_SHIFT 2 +# define R300_GB_PIPE_SELECT_PIPE2_ID_SHIFT 4 +# define R300_GB_PIPE_SELECT_PIPE3_ID_SHIFT 6 +# define R300_GB_PIPE_SELECT_PIPE_MASK_SHIFT 8 +# define R300_GB_PIPE_SELECT_MAX_PIPE 12 +# define R300_GB_PIPE_SELECT_BAD_PIPES 14 +# define R300_GB_PIPE_SELECT_CONFIG_PIPES 18 /* Specifies the sizes of the various FIFO`s in the sc/rs. */ -#define GB_FIFO_SIZE1 0x4070 +#define R300_GB_FIFO_SIZE1 0x4070 /* High water mark for SC input fifo */ -# define GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_SHIFT 0 -# define GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_MASK 0x0000003f +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_SHIFT 0 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_IFIFO_MASK 0x0000003f /* High water mark for SC input fifo (B) */ -# define GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_SHIFT 6 -# define GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_MASK 0x00000fc0 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_SHIFT 6 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_BFIFO_MASK 0x00000fc0 /* High water mark for RS colors' fifo */ -# define GB_FIFO_SIZE1_SC_HIGHWATER_COL_SHIFT 12 -# define GB_FIFO_SIZE1_SC_HIGHWATER_COL_MASK 0x0003f000 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_COL_SHIFT 12 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_COL_MASK 0x0003f000 /* High water mark for RS textures' fifo */ -# define GB_FIFO_SIZE1_SC_HIGHWATER_TEX_SHIFT 18 -# define GB_FIFO_SIZE1_SC_HIGHWATER_TEX_MASK 0x00fc0000 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_TEX_SHIFT 18 +# define R300_GB_FIFO_SIZE1_SC_HIGHWATER_TEX_MASK 0x00fc0000 /* This table specifies the source location and format for up to 16 texture * addresses (i[0]:i[15]) and four colors (c[0]:c[3]) @@ -1293,7 +1293,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R500_RS_INST_TEX_ID(x) ((x) << 0) #define R500_RS_INST_TEX_CN_WRITE (1 << 4) #define R500_RS_INST_TEX_ADDR_SHIFT 5 -# define R500_RS_INST_TEX_ADDR(x) ((x) << 0) +# define R500_RS_INST_TEX_ADDR(x) ((x) << 5) #define R500_RS_INST_COL_ID_SHIFT 12 # define R500_RS_INST_COL_ID(x) ((x) << 12) #define R500_RS_INST_COL_CN_NO_WRITE (0 << 16) @@ -1463,6 +1463,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_MIN_FILTER_MIP_NEAREST (1 << 13) # define R300_TX_MIN_FILTER_MIP_LINEAR (2 << 13) # define R300_TX_MIN_FILTER_MIP_MASK (3 << 13) +# define R300_TX_MAX_MIP_LEVEL_SHIFT 17 +# define R300_TX_MAX_MIP_LEVEL_MASK (0xf << 17) # define R300_TX_MAX_ANISO_1_TO_1 (0 << 21) # define R300_TX_MAX_ANISO_2_TO_1 (1 << 21) # define R300_TX_MAX_ANISO_4_TO_1 (2 << 21) @@ -1471,6 +1473,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_MAX_ANISO_MASK (7 << 21) # define R300_TX_WRAP_S(x) ((x) << 0) # define R300_TX_WRAP_T(x) ((x) << 3) +# define R300_TX_MAX_MIP_LEVEL(x) ((x) << 17) #define R300_TX_FILTER1_0 0x4440 # define R300_CHROMA_KEY_MODE_DISABLE 0 @@ -1500,8 +1503,6 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_TX_HEIGHTMASK_MASK (2047 << 11) # define R300_TX_DEPTHMASK_SHIFT 22 # define R300_TX_DEPTHMASK_MASK (0xf << 22) -# define R300_TX_MAX_MIP_LEVEL_SHIFT 26 -# define R300_TX_MAX_MIP_LEVEL_MASK (0xf << 26) # define R300_TX_SIZE_PROJECTED (1 << 30) # define R300_TX_PITCH_EN (1 << 31) # define R300_TX_WIDTH(x) ((x) << 0) @@ -1618,18 +1619,20 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #define R300_TX_OFFSET_5 0x4554 #define R300_TX_OFFSET_6 0x4558 #define R300_TX_OFFSET_7 0x455C - /* BEGIN: Guess from R200 */ + # define R300_TXO_ENDIAN_NO_SWAP (0 << 0) # define R300_TXO_ENDIAN_BYTE_SWAP (1 << 0) # define R300_TXO_ENDIAN_WORD_SWAP (2 << 0) # define R300_TXO_ENDIAN_HALFDW_SWAP (3 << 0) -# define R300_TXO_MACRO_TILE (1 << 2) +# define R300_TXO_MACRO_TILE_LINEAR (0 << 2) +# define R300_TXO_MACRO_TILE_TILED (1 << 2) +# define R300_TXO_MACRO_TILE(x) ((x) << 2) # define R300_TXO_MICRO_TILE_LINEAR (0 << 3) -# define R300_TXO_MICRO_TILE (1 << 3) -# define R300_TXO_MICRO_TILE_SQUARE (2 << 3) +# define R300_TXO_MICRO_TILE_TILED (1 << 3) +# define R300_TXO_MICRO_TILE_TILED_SQUARE (2 << 3) +# define R300_TXO_MICRO_TILE(x) ((x) << 3) # define R300_TXO_OFFSET_MASK 0xffffffe0 # define R300_TXO_OFFSET_SHIFT 5 - /* END: Guess from R200 */ /* 32 bit chroma key */ #define R300_TX_CHROMA_KEY_0 0x4580 @@ -2144,6 +2147,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. /* Unpipelined. */ #define R300_RB3D_CCTL 0x4e00 +# define R300_RB3D_CCTL_NUM_MULTIWRITES(x) (MAX2(((x)-1), 0) << 5) # define R300_RB3D_CCTL_NUM_MULTIWRITES_1_BUFFER (0 << 5) # define R300_RB3D_CCTL_NUM_MULTIWRITES_2_BUFFERS (1 << 5) # define R300_RB3D_CCTL_NUM_MULTIWRITES_3_BUFFERS (2 << 5) @@ -2184,6 +2188,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1 (4 << 3) # define R300_DISCARD_SRC_PIXELS_SRC_COLOR_1 (5 << 3) # define R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1 (6 << 3) +# define R500_SRC_ALPHA_0_NO_READ (1 << 30) +# define R500_SRC_ALPHA_1_NO_READ (1 << 31) /* the following are shared between CBLEND and ABLEND */ # define R300_FCN_MASK (3 << 12) @@ -2279,9 +2285,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_COLORPITCH_MASK 0x00003FFE # define R300_COLOR_TILE_DISABLE (0 << 16) # define R300_COLOR_TILE_ENABLE (1 << 16) +# define R300_COLOR_TILE(x) ((x) << 16) # define R300_COLOR_MICROTILE_DISABLE (0 << 17) # define R300_COLOR_MICROTILE_ENABLE (1 << 17) # define R300_COLOR_MICROTILE_ENABLE_SQUARE (2 << 17) /* Only available in 16-bit */ +# define R300_COLOR_MICROTILE(x) ((x) << 17) # define R300_COLOR_ENDIAN_NO_SWAP (0 << 19) # define R300_COLOR_ENDIAN_WORD_SWAP (1 << 19) # define R300_COLOR_ENDIAN_DWORD_SWAP (2 << 19) @@ -2540,9 +2548,11 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. # define R300_DEPTHPITCH_MASK 0x00003FFC # define R300_DEPTHMACROTILE_DISABLE (0 << 16) # define R300_DEPTHMACROTILE_ENABLE (1 << 16) +# define R300_DEPTHMACROTILE(x) ((x) << 16) # define R300_DEPTHMICROTILE_LINEAR (0 << 17) # define R300_DEPTHMICROTILE_TILED (1 << 17) # define R300_DEPTHMICROTILE_TILED_SQUARE (2 << 17) +# define R300_DEPTHMICROTILE(x) ((x) << 17) # define R300_DEPTHENDIAN_NO_SWAP (0 << 18) # define R300_DEPTHENDIAN_WORD_SWAP (1 << 18) # define R300_DEPTHENDIAN_DWORD_SWAP (2 << 18) @@ -3292,6 +3302,11 @@ enum { */ #define R300_PACKET3_3D_LOAD_VBPNTR 0x00002F00 +# define R300_VBPNTR_SIZE0(x) ((x) >> 2) +# define R300_VBPNTR_STRIDE0(x) (((x) >> 2) << 8) +# define R300_VBPNTR_SIZE1(x) (((x) >> 2) << 16) +# define R300_VBPNTR_STRIDE1(x) (((x) >> 2) << 24) + #define R300_PACKET3_INDX_BUFFER 0x00003300 # define R300_INDX_BUFFER_DST_SHIFT 0 # define R300_INDX_BUFFER_SKIP_SHIFT 16 diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index 62e1456ed3..710d850163 100644 --- a/src/gallium/drivers/r300/r300_render.c +++ b/src/gallium/drivers/r300/r300_render.c @@ -26,6 +26,8 @@ #include "draw/draw_context.h" #include "draw/draw_vbuf.h" +#include "indices/u_indices.h" + #include "pipe/p_inlines.h" #include "util/u_memory.h" @@ -37,7 +39,6 @@ #include "r300_reg.h" #include "r300_render.h" #include "r300_state_derived.h" -#include "r300_vbo.h" /* r300_render: Vertex and index buffer primitive emission. */ #define R300_MAX_VBO_SIZE (1024 * 1024) @@ -70,14 +71,100 @@ uint32_t r300_translate_primitive(unsigned prim) } } +static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, + unsigned mode) +{ + struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state; + uint32_t color_control = rs->color_control; + + /* By default (see r300_state.c:r300_create_rs_state) color_control is + * initialized to provoking the first vertex. + * + * Triangle fans must be reduced to the second vertex, not the first, in + * Gallium flatshade-first mode, as per the GL spec. + * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt) + * + * Quads never provoke correctly in flatshade-first mode. The first + * vertex is never considered as provoking, so only the second, third, + * and fourth vertices can be selected, and both "third" and "last" modes + * select the fourth vertex. This is probably due to D3D lacking quads. + * + * Similarly, polygons reduce to the first, not the last, vertex, when in + * "last" mode, and all other modes start from the second vertex. + * + * ~ C. + */ + + if (rs->rs.flatshade_first) { + switch (mode) { + case PIPE_PRIM_TRIANGLE_FAN: + color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND; + break; + case PIPE_PRIM_QUADS: + case PIPE_PRIM_QUAD_STRIP: + case PIPE_PRIM_POLYGON: + color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; + break; + default: + color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST; + break; + } + } else { + color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; + } + + return color_control; +} + +static void r300_emit_draw_immediate(struct r300_context *r300, + unsigned mode, + unsigned start, + unsigned count) +{ + struct pipe_buffer* vbo = r300->vertex_buffer[0].buffer; + unsigned vertex_size = r300->vertex_buffer[0].stride / sizeof(float); + unsigned i; + uint32_t* map; + CS_LOCALS(r300); + + map = (uint32_t*)pipe_buffer_map_range(r300->context.screen, vbo, + start * vertex_size, count * vertex_size, + PIPE_BUFFER_USAGE_CPU_READ); + + BEGIN_CS(10 + count * vertex_size); + OUT_CS_REG(R300_GA_COLOR_CONTROL, + r300_provoking_vertex_fixes(r300, mode)); + OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size); + OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0); + OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); + OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size); + OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) | + r300_translate_primitive(mode)); + //debug_printf("r300: Immd %d verts, %d attrs\n", count, vertex_size); + for (i = 0; i < count * vertex_size; i++) { + if (i % vertex_size == 0) { + //debug_printf("r300: -- vert --\n"); + } + //debug_printf("r300: 0x%08x\n", *map); + OUT_CS(*map); + map++; + } + END_CS; + + pipe_buffer_unmap(r300->context.screen, vbo); +} + static void r300_emit_draw_arrays(struct r300_context *r300, unsigned mode, unsigned count) { CS_LOCALS(r300); - BEGIN_CS(4); - OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count); + BEGIN_CS(8); + OUT_CS_REG(R300_GA_COLOR_CONTROL, + r300_provoking_vertex_fixes(r300, mode)); + OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0); + OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0); OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) | r300_translate_primitive(mode)); @@ -102,7 +189,10 @@ static void r300_emit_draw_elements(struct r300_context *r300, assert((start * indexSize) % 4 == 0); assert(offset_dwords == 0); - BEGIN_CS(10); + BEGIN_CS(14); + OUT_CS_REG(R300_GA_COLOR_CONTROL, + r300_provoking_vertex_fixes(r300, mode)); + OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex); OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex); OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0); if (indexSize == 4) { @@ -157,35 +247,85 @@ validate: return TRUE; } +static struct pipe_buffer* r300_translate_elts(struct r300_context* r300, + struct pipe_buffer* elts, + unsigned* size, + unsigned* mode, + unsigned* count) +{ + struct pipe_screen* screen = r300->context.screen; + struct pipe_buffer* new_elts; + void *in_map, *out_map; + unsigned out_prim, out_index_size, out_nr; + u_translate_func out_translate; + + (void)u_index_translator(~0, *mode, *size, *count, PV_LAST, PV_LAST, + &out_prim, &out_index_size, &out_nr, &out_translate); + + new_elts = screen->buffer_create(screen, 32, + PIPE_BUFFER_USAGE_INDEX | + PIPE_BUFFER_USAGE_CPU_WRITE | + PIPE_BUFFER_USAGE_GPU_READ, + out_index_size * out_nr); + + in_map = pipe_buffer_map(screen, elts, PIPE_BUFFER_USAGE_CPU_READ); + out_map = pipe_buffer_map(screen, new_elts, PIPE_BUFFER_USAGE_CPU_WRITE); + + out_translate(in_map, *count, out_map); + + pipe_buffer_unmap(screen, elts); + pipe_buffer_unmap(screen, new_elts); + + *size = out_index_size; + *mode = out_prim; + *count = out_nr; + + return new_elts; +} + /* This is the fast-path drawing & emission for HW TCL. */ -boolean r300_draw_range_elements(struct pipe_context* pipe, - struct pipe_buffer* indexBuffer, - unsigned indexSize, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count) +void r300_draw_range_elements(struct pipe_context* pipe, + struct pipe_buffer* indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count) { struct r300_context* r300 = r300_context(pipe); + struct pipe_buffer* orgIndexBuffer = indexBuffer; if (!u_trim_pipe_prim(mode, &count)) { - return FALSE; + return; } if (count > 65535) { - return FALSE; + /* XXX: use aux/indices functions to split this into smaller + * primitives. + */ + return; } r300_update_derived_state(r300); if (!r300_setup_vertex_buffers(r300)) { - return FALSE; + return; } - setup_vertex_attributes(r300); + if (indexSize == 1) { + indexBuffer = r300_translate_elts(r300, indexBuffer, + &indexSize, &mode, &count); + } - setup_index_buffer(r300, indexBuffer, indexSize); + if (!r300->winsys->add_buffer(r300->winsys, indexBuffer, + RADEON_GEM_DOMAIN_GTT, 0)) { + goto cleanup; + } + + if (!r300->winsys->validate(r300->winsys)) { + goto cleanup; + } r300_emit_dirty_state(r300); @@ -194,47 +334,52 @@ boolean r300_draw_range_elements(struct pipe_context* pipe, r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex, mode, start, count); - return TRUE; +cleanup: + if (indexBuffer != orgIndexBuffer) { + pipe->screen->buffer_destroy(indexBuffer); + } } /* Simple helpers for context setup. Should probably be moved to util. */ -boolean r300_draw_elements(struct pipe_context* pipe, - struct pipe_buffer* indexBuffer, - unsigned indexSize, unsigned mode, - unsigned start, unsigned count) +void r300_draw_elements(struct pipe_context* pipe, + struct pipe_buffer* indexBuffer, + unsigned indexSize, unsigned mode, + unsigned start, unsigned count) { - return pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0, - mode, start, count); + pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0, + mode, start, count); } -boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode, - unsigned start, unsigned count) +void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, + unsigned start, unsigned count) { struct r300_context* r300 = r300_context(pipe); if (!u_trim_pipe_prim(mode, &count)) { - return FALSE; + return; } if (count > 65535) { - return FALSE; + /* XXX: driver needs to handle this -- use the functions in + * aux/indices to split this into several smaller primitives. + */ + return; } r300_update_derived_state(r300); if (!r300_setup_vertex_buffers(r300)) { - return FALSE; + return; } - setup_vertex_attributes(r300); - r300_emit_dirty_state(r300); - r300_emit_aos(r300, start); - - r300_emit_draw_arrays(r300, mode, count); - - return TRUE; + if (FALSE && count <= 4 && r300->vertex_buffer_count == 1) { + r300_emit_draw_immediate(r300, mode, start, count); + } else { + r300_emit_aos(r300, start); + r300_emit_draw_arrays(r300, mode, count); + } } /**************************************************************************** @@ -243,7 +388,7 @@ boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode, ***************************************************************************/ /* SW TCL arrays, using Draw. */ -boolean r300_swtcl_draw_arrays(struct pipe_context* pipe, +void r300_swtcl_draw_arrays(struct pipe_context* pipe, unsigned mode, unsigned start, unsigned count) @@ -252,7 +397,7 @@ boolean r300_swtcl_draw_arrays(struct pipe_context* pipe, int i; if (!u_trim_pipe_prim(mode, &count)) { - return FALSE; + return; } for (i = 0; i < r300->vertex_buffer_count; i++) { @@ -265,8 +410,9 @@ boolean r300_swtcl_draw_arrays(struct pipe_context* pipe, draw_set_mapped_element_buffer(r300->draw, 0, NULL); draw_set_mapped_constant_buffer(r300->draw, - r300->shader_constants[PIPE_SHADER_VERTEX].constants, - r300->shader_constants[PIPE_SHADER_VERTEX].count * + PIPE_SHADER_VERTEX, + r300->shader_constants[PIPE_SHADER_VERTEX].constants, + r300->shader_constants[PIPE_SHADER_VERTEX].count * (sizeof(float) * 4)); draw_arrays(r300->draw, mode, start, count); @@ -275,12 +421,10 @@ boolean r300_swtcl_draw_arrays(struct pipe_context* pipe, pipe_buffer_unmap(pipe->screen, r300->vertex_buffer[i].buffer); draw_set_mapped_vertex_buffer(r300->draw, i, NULL); } - - return TRUE; } /* SW TCL elements, using Draw. */ -boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe, +void r300_swtcl_draw_range_elements(struct pipe_context* pipe, struct pipe_buffer* indexBuffer, unsigned indexSize, unsigned minIndex, @@ -291,9 +435,10 @@ boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); int i; + void* indices; if (!u_trim_pipe_prim(mode, &count)) { - return FALSE; + return; } for (i = 0; i < r300->vertex_buffer_count; i++) { @@ -303,12 +448,13 @@ boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe, draw_set_mapped_vertex_buffer(r300->draw, i, buf); } - void* indices = pipe_buffer_map(pipe->screen, indexBuffer, - PIPE_BUFFER_USAGE_CPU_READ); + indices = pipe_buffer_map(pipe->screen, indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_element_buffer_range(r300->draw, indexSize, minIndex, maxIndex, indices); draw_set_mapped_constant_buffer(r300->draw, + PIPE_SHADER_VERTEX, r300->shader_constants[PIPE_SHADER_VERTEX].constants, r300->shader_constants[PIPE_SHADER_VERTEX].count * (sizeof(float) * 4)); @@ -323,8 +469,6 @@ boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe, pipe_buffer_unmap(pipe->screen, indexBuffer); draw_set_mapped_element_buffer_range(r300->draw, 0, start, start + count - 1, NULL); - - return TRUE; } /* Object for rendering using Draw. */ @@ -400,7 +544,7 @@ static void* r300_render_map_vertices(struct vbuf_render* render) r300render->vbo_ptr = pipe_buffer_map(screen, r300render->vbo, PIPE_BUFFER_USAGE_CPU_WRITE); - return (r300render->vbo_ptr + r300render->vbo_offset); + return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset); } static void r300_render_unmap_vertices(struct vbuf_render* render, diff --git a/src/gallium/drivers/r300/r300_render.h b/src/gallium/drivers/r300/r300_render.h index da83069083..27b5e6a963 100644 --- a/src/gallium/drivers/r300/r300_render.h +++ b/src/gallium/drivers/r300/r300_render.h @@ -25,35 +25,35 @@ uint32_t r300_translate_primitive(unsigned prim); -boolean r300_draw_range_elements(struct pipe_context* pipe, - struct pipe_buffer* indexBuffer, - unsigned indexSize, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count); - -boolean r300_draw_elements(struct pipe_context* pipe, - struct pipe_buffer* indexBuffer, - unsigned indexSize, unsigned mode, - unsigned start, unsigned count); - -boolean r300_draw_arrays(struct pipe_context* pipe, unsigned mode, - unsigned start, unsigned count); - -boolean r300_swtcl_draw_arrays(struct pipe_context* pipe, - unsigned mode, - unsigned start, - unsigned count); - -boolean r300_swtcl_draw_range_elements(struct pipe_context* pipe, - struct pipe_buffer* indexBuffer, - unsigned indexSize, - unsigned minIndex, - unsigned maxIndex, - unsigned mode, - unsigned start, - unsigned count); +void r300_draw_range_elements(struct pipe_context* pipe, + struct pipe_buffer* indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count); + +void r300_draw_elements(struct pipe_context* pipe, + struct pipe_buffer* indexBuffer, + unsigned indexSize, unsigned mode, + unsigned start, unsigned count); + +void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, + unsigned start, unsigned count); + +void r300_swtcl_draw_arrays(struct pipe_context* pipe, + unsigned mode, + unsigned start, + unsigned count); + +void r300_swtcl_draw_range_elements(struct pipe_context* pipe, + struct pipe_buffer* indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count); #endif /* R300_RENDER_H */ diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c index 390b63007e..287664b1d2 100644 --- a/src/gallium/drivers/r300/r300_screen.c +++ b/src/gallium/drivers/r300/r300_screen.c @@ -21,13 +21,15 @@ * USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include "pipe/p_inlines.h" +#include "util/u_format.h" #include "util/u_memory.h" #include "util/u_simple_screen.h" #include "r300_context.h" #include "r300_screen.h" #include "r300_texture.h" -#include "r300_winsys.h" + +#include "radeon_winsys.h" /* Return the identifier behind whom the brave coders responsible for this * amalgamation of code, sweat, and duct tape, routinely obscure their names. @@ -81,6 +83,7 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) switch (param) { case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS: + case PIPE_CAP_MAX_COMBINED_SAMPLERS: /* XXX I'm told this goes up to 16 */ return 8; case PIPE_CAP_NPOT_TEXTURES: @@ -140,6 +143,12 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) return 0; case PIPE_CAP_BLEND_EQUATION_SEPARATE: return 1; + case PIPE_CAP_SM3: + if (r300screen->caps->is_r500) { + return 1; + } else { + return 0; + } default: debug_printf("r300: Implementation error: Bad param %d\n", param); @@ -219,12 +228,18 @@ static boolean check_tex_format(enum pipe_format format, uint32_t usage, /* Z buffer or texture */ case PIPE_FORMAT_Z16_UNORM: + retval = usage & + (PIPE_TEXTURE_USAGE_DEPTH_STENCIL | + PIPE_TEXTURE_USAGE_SAMPLER); + break; + + /* 24bit Z buffer can only be used as a texture on R500. */ case PIPE_FORMAT_Z24X8_UNORM: /* Z buffer with stencil or texture */ case PIPE_FORMAT_Z24S8_UNORM: retval = usage & (PIPE_TEXTURE_USAGE_DEPTH_STENCIL | - PIPE_TEXTURE_USAGE_SAMPLER); + (is_r500 ? PIPE_TEXTURE_USAGE_SAMPLER : 0)); break; /* Definitely unsupported formats. */ @@ -311,14 +326,10 @@ r300_get_tex_transfer(struct pipe_screen *screen, trans = CALLOC_STRUCT(r300_transfer); if (trans) { pipe_texture_reference(&trans->transfer.texture, texture); - trans->transfer.format = texture->format; trans->transfer.x = x; trans->transfer.y = y; trans->transfer.width = w; trans->transfer.height = h; - trans->transfer.block = texture->block; - trans->transfer.nblocksx = texture->nblocksx[level]; - trans->transfer.nblocksy = texture->nblocksy[level]; trans->transfer.stride = r300_texture_get_stride(tex, level); trans->transfer.usage = usage; @@ -344,6 +355,7 @@ static void* r300_transfer_map(struct pipe_screen* screen, { struct r300_texture* tex = (struct r300_texture*)transfer->texture; char* map; + enum pipe_format format = tex->tex.format; map = pipe_buffer_map(screen, tex->buffer, pipe_transfer_buffer_flags(transfer)); @@ -353,8 +365,8 @@ static void* r300_transfer_map(struct pipe_screen* screen, } return map + r300_transfer(transfer)->offset + - transfer->y / transfer->block.height * transfer->stride + - transfer->x / transfer->block.width * transfer->block.size; + transfer->y / util_format_get_blockheight(format) * transfer->stride + + transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format); } static void r300_transfer_unmap(struct pipe_screen* screen, @@ -372,7 +384,7 @@ static void r300_destroy_screen(struct pipe_screen* pscreen) FREE(r300screen); } -struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys) +struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys) { struct r300_screen* r300screen = CALLOC_STRUCT(r300_screen); struct r300_capabilities* caps = CALLOC_STRUCT(r300_capabilities); @@ -380,14 +392,14 @@ struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys) if (!r300screen || !caps) return NULL; - caps->pci_id = r300_winsys->pci_id; - caps->num_frag_pipes = r300_winsys->gb_pipes; - caps->num_z_pipes = r300_winsys->z_pipes; + caps->pci_id = radeon_winsys->pci_id; + caps->num_frag_pipes = radeon_winsys->gb_pipes; + caps->num_z_pipes = radeon_winsys->z_pipes; r300_parse_chipset(caps); r300screen->caps = caps; - r300screen->screen.winsys = (struct pipe_winsys*)r300_winsys; + r300screen->screen.winsys = (struct pipe_winsys*)radeon_winsys; r300screen->screen.destroy = r300_destroy_screen; r300screen->screen.get_name = r300_get_name; r300screen->screen.get_vendor = r300_get_vendor; diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h index 41df31f670..2217988add 100644 --- a/src/gallium/drivers/r300/r300_screen.h +++ b/src/gallium/drivers/r300/r300_screen.h @@ -27,6 +27,8 @@ #include "r300_chipset.h" +struct radeon_winsys; + struct r300_screen { /* Parent class */ struct pipe_screen screen; @@ -56,6 +58,6 @@ r300_transfer(struct pipe_transfer* transfer) } /* Creates a new r300 screen. */ -struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys); +struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys); #endif /* R300_SCREEN_H */ diff --git a/src/gallium/drivers/r300/r300_shader_semantics.h b/src/gallium/drivers/r300/r300_shader_semantics.h new file mode 100644 index 0000000000..6796841b29 --- /dev/null +++ b/src/gallium/drivers/r300/r300_shader_semantics.h @@ -0,0 +1,66 @@ +/* + * Copyright 2009 Marek Olšák <maraeo@gmail.com> + * + * 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 + * on the rights to use, copy, modify, merge, publish, distribute, sub + * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL + * THE AUTHOR(S) AND/OR THEIR 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. */ + +#ifndef R300_SHADER_SEMANTICS_H +#define R300_SHADER_SEMANTICS_H + +#define ATTR_UNUSED (-1) +#define ATTR_COLOR_COUNT 2 +#define ATTR_GENERIC_COUNT 16 + +/* This structure contains information about what attributes are written by VS + * or read by FS. (but not both) It's much easier to work with than + * tgsi_shader_info. + * + * The variables contain indices to tgsi_shader_info semantics and those + * indices are nothing else than input/output register numbers. */ +struct r300_shader_semantics { + int pos; + int psize; + int color[ATTR_COLOR_COUNT]; + int bcolor[ATTR_COLOR_COUNT]; + int generic[ATTR_GENERIC_COUNT]; + int fog; + int wpos; +}; + +static INLINE void r300_shader_semantics_reset( + struct r300_shader_semantics* info) +{ + int i; + + info->pos = ATTR_UNUSED; + info->psize = ATTR_UNUSED; + info->fog = ATTR_UNUSED; + info->wpos = ATTR_UNUSED; + + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + info->color[i] = ATTR_UNUSED; + info->bcolor[i] = ATTR_UNUSED; + } + + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + info->generic[i] = ATTR_UNUSED; + } +} + +#endif diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index d1eced61db..281ff68449 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -41,6 +42,120 @@ /* r300_state: Functions used to intialize state context by translating * Gallium state objects into semi-native r300 state objects. */ +static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA, + unsigned dstRGB, unsigned dstA) +{ + /* If the blend equation is ADD or REVERSE_SUBTRACT, + * SRC_ALPHA == 0, and the following state is set, the colorbuffer + * will not be changed. + * Notice that the dst factors are the src factors inverted. */ + return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || + srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || + srcRGB == PIPE_BLENDFACTOR_ZERO) && + (srcA == PIPE_BLENDFACTOR_SRC_COLOR || + srcA == PIPE_BLENDFACTOR_SRC_ALPHA || + srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || + srcA == PIPE_BLENDFACTOR_ZERO) && + (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ONE) && + (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ONE); +} + +static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA, + unsigned dstRGB, unsigned dstA) +{ + /* If the blend equation is ADD or REVERSE_SUBTRACT, + * SRC_ALPHA == 1, and the following state is set, the colorbuffer + * will not be changed. + * Notice that the dst factors are the src factors inverted. */ + return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + srcRGB == PIPE_BLENDFACTOR_ZERO) && + (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || + srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + srcA == PIPE_BLENDFACTOR_ZERO) && + (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ONE) && + (dstA == PIPE_BLENDFACTOR_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ONE); +} + +static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA, + unsigned dstRGB, unsigned dstA) +{ + /* If the blend equation is ADD or REVERSE_SUBTRACT, + * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer + * will not be changed. + * Notice that the dst factors are the src factors inverted. */ + return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || + srcRGB == PIPE_BLENDFACTOR_ZERO) && + (srcA == PIPE_BLENDFACTOR_ZERO) && + (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || + dstRGB == PIPE_BLENDFACTOR_ONE) && + (dstA == PIPE_BLENDFACTOR_ONE); +} + +static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA, + unsigned dstRGB, unsigned dstA) +{ + /* If the blend equation is ADD or REVERSE_SUBTRACT, + * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer + * will not be changed. + * Notice that the dst factors are the src factors inverted. */ + return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || + srcRGB == PIPE_BLENDFACTOR_ZERO) && + (srcA == PIPE_BLENDFACTOR_ZERO) && + (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || + dstRGB == PIPE_BLENDFACTOR_ONE) && + (dstA == PIPE_BLENDFACTOR_ONE); +} + +static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA, + unsigned dstRGB, unsigned dstA) +{ + /* If the blend equation is ADD or REVERSE_SUBTRACT, + * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set, + * the colorbuffer will not be changed. + * Notice that the dst factors are the src factors inverted. */ + return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR || + srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA || + srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || + srcRGB == PIPE_BLENDFACTOR_ZERO) && + (srcA == PIPE_BLENDFACTOR_SRC_COLOR || + srcA == PIPE_BLENDFACTOR_SRC_ALPHA || + srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE || + srcA == PIPE_BLENDFACTOR_ZERO) && + (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || + dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ONE) && + (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ONE); +} + +static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA, + unsigned dstRGB, unsigned dstA) +{ + /* If the blend equation is ADD or REVERSE_SUBTRACT, + * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set, + * the colorbuffer will not be changed. + * Notice that the dst factors are the src factors inverted. */ + return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR || + srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + srcRGB == PIPE_BLENDFACTOR_ZERO) && + (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR || + srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + srcA == PIPE_BLENDFACTOR_ZERO) && + (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR || + dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ONE) && + (dstA == PIPE_BLENDFACTOR_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ONE); +} + /* Create a new blend state based on the CSO blend state. * * This encompasses alpha blending, logic/raster ops, and blend dithering. */ @@ -66,7 +181,11 @@ static void* r300_create_blend_state(struct pipe_context* pipe, ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) | ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT); - /* optimization: some operations do not require the destination color */ + /* Optimization: some operations do not require the destination color. + * + * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled, + * otherwise blending gives incorrect results. It seems to be + * a hardware bug. */ if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN || eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX || dstRGB != PIPE_BLENDFACTOR_ZERO || @@ -78,11 +197,81 @@ static void* r300_create_blend_state(struct pipe_context* pipe, srcA == PIPE_BLENDFACTOR_DST_COLOR || srcA == PIPE_BLENDFACTOR_DST_ALPHA || srcA == PIPE_BLENDFACTOR_INV_DST_COLOR || - srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA) + srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA || + srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) { + /* Enable reading from the colorbuffer. */ blend->blend_control |= R300_READ_ENABLE; - /* XXX implement the optimization with DISCARD_SRC_PIXELS*/ - /* XXX implement the optimization with SRC_ALPHA_?_NO_READ */ + if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) { + /* Optimization: Depending on incoming pixels, we can + * conditionally disable the reading in hardware... */ + if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN && + eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) { + /* Disable reading if SRC_ALPHA == 0. */ + if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ZERO) && + (dstA == PIPE_BLENDFACTOR_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ZERO)) { + blend->blend_control |= R500_SRC_ALPHA_0_NO_READ; + } + + /* Disable reading if SRC_ALPHA == 1. */ + if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstRGB == PIPE_BLENDFACTOR_ZERO) && + (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR || + dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA || + dstA == PIPE_BLENDFACTOR_ZERO)) { + blend->blend_control |= R500_SRC_ALPHA_1_NO_READ; + } + } + } + } + + /* Optimization: discard pixels which don't change the colorbuffer. + * + * The code below is non-trivial and some math is involved. + * + * Discarding pixels must be disabled when FP16 AA is enabled. + * This is a hardware bug. Also, this implementation wouldn't work + * with FP blending enabled and equation clamping disabled. + * + * Equations other than ADD are rarely used and therefore won't be + * optimized. */ + if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) && + (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) { + /* ADD: X+Y + * REVERSE_SUBTRACT: Y-X + * + * The idea is: + * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1, + * then CB will not be changed. + * + * Given the srcFactor and dstFactor variables, we can derive + * what src and dst should be equal to and discard appropriate + * pixels. + */ + if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) { + blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0; + } else if (blend_discard_if_src_alpha_1(srcRGB, srcA, + dstRGB, dstA)) { + blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1; + } else if (blend_discard_if_src_color_0(srcRGB, srcA, + dstRGB, dstA)) { + blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0; + } else if (blend_discard_if_src_color_1(srcRGB, srcA, + dstRGB, dstA)) { + blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1; + } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA, + dstRGB, dstA)) { + blend->blend_control |= + R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0; + } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA, + dstRGB, dstA)) { + blend->blend_control |= + R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1; + } + } /* separate alpha */ if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) { @@ -128,8 +317,8 @@ static void r300_bind_blend_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - r300->blend_state = (struct r300_blend_state*)state; - r300->dirty_state |= R300_NEW_BLEND; + r300->blend_state.state = state; + r300->blend_state.dirty = TRUE; } /* Free blend state. */ @@ -151,19 +340,24 @@ static void r300_set_blend_color(struct pipe_context* pipe, const struct pipe_blend_color* color) { struct r300_context* r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(pipe->screen); + struct r300_blend_color_state* state = + (struct r300_blend_color_state*)r300->blend_color_state.state; + union util_color uc; - util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, - &r300->blend_color_state->blend_color); + util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); + state->blend_color = uc.ui; /* XXX if FP16 blending is enabled, we should use the FP16 format */ - r300->blend_color_state->blend_color_red_alpha = + state->blend_color_red_alpha = float_to_fixed10(color->color[0]) | (float_to_fixed10(color->color[3]) << 16); - r300->blend_color_state->blend_color_green_blue = + state->blend_color_green_blue = float_to_fixed10(color->color[2]) | (float_to_fixed10(color->color[1]) << 16); - r300->dirty_state |= R300_NEW_BLEND_COLOR; + r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2; + r300->blend_color_state.dirty = TRUE; } static void r300_set_clip_state(struct pipe_context* pipe, @@ -172,12 +366,15 @@ static void r300_set_clip_state(struct pipe_context* pipe, struct r300_context* r300 = r300_context(pipe); if (r300_screen(pipe->screen)->caps->has_tcl) { - r300->clip_state = *state; - r300->dirty_state |= R300_NEW_CLIP; + memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state)); + r300->clip_state.size = 29; } else { draw_flush(r300->draw); draw_set_clip_state(r300->draw, state); + r300->clip_state.size = 2; } + + r300->clip_state.dirty = TRUE; } /* Create a new depth, stencil, and alpha state based on the CSO dsa state. @@ -270,9 +467,11 @@ static void r300_bind_dsa_state(struct pipe_context* pipe, void* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_screen* r300screen = r300_screen(pipe->screen); - r300->dsa_state = (struct r300_dsa_state*)state; - r300->dirty_state |= R300_NEW_DSA; + r300->dsa_state.state = state; + r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6; + r300->dsa_state.dirty = TRUE; } /* Free DSA state. */ @@ -282,13 +481,6 @@ static void r300_delete_dsa_state(struct pipe_context* pipe, FREE(state); } -static void r300_set_edgeflags(struct pipe_context* pipe, - const unsigned* bitfield) -{ - /* XXX you know it's bad when i915 has this blank too */ - /* XXX and even worse, I have no idea WTF the bitfield is */ -} - static void r300_set_framebuffer_state(struct pipe_context* pipe, const struct pipe_framebuffer_state* state) @@ -301,7 +493,12 @@ static void r300->framebuffer_state = *state; + /* Don't rely on the order of states being set for the first time. */ r300->dirty_state |= R300_NEW_FRAMEBUFFERS; + + r300->blend_state.dirty = TRUE; + r300->dsa_state.dirty = TRUE; + r300->scissor_state.dirty = TRUE; } /* Create fragment shader state. */ @@ -317,6 +514,7 @@ static void* r300_create_fs_state(struct pipe_context* pipe, fs->state.tokens = tgsi_dup_tokens(shader->tokens); tgsi_scan_shader(shader->tokens, &fs->info); + r300_shader_read_fs_inputs(&fs->info, &fs->inputs); return (void*)fs; } @@ -330,11 +528,14 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) if (fs == NULL) { r300->fs = NULL; return; - } else if (!fs->translated) { - r300_translate_fragment_shader(r300, fs); } r300->fs = fs; + r300_pick_fragment_shader(r300); + + if (r300->vs && r300_vertex_shader_setup_wpos(r300)) { + r300->dirty_state |= R300_NEW_VERTEX_FORMAT; + } r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS; } @@ -343,7 +544,14 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader) static void r300_delete_fs_state(struct pipe_context* pipe, void* shader) { struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader; - rc_constants_destroy(&fs->code.constants); + struct r300_fragment_shader_code *tmp, *ptr = fs->first; + + while (ptr) { + tmp = ptr; + ptr = ptr->next; + rc_constants_destroy(&tmp->code.constants); + FREE(tmp); + } FREE((void*)fs->state.tokens); FREE(shader); } @@ -369,8 +577,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe, /* Copy rasterizer state for Draw. */ rs->rs = *state; - rs->enable_vte = !state->bypass_vs_clip_and_viewport; - #ifdef PIPE_ARCH_LITTLE_ENDIAN rs->vap_control_status = R300_VC_NO_SWAP; #else @@ -382,8 +588,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe, if (state->bypass_vs_clip_and_viewport || !r300_screen(pipe->screen)->caps->has_tcl) { rs->vap_control_status |= R300_VAP_TCL_BYPASS; - } else { - rs->rs.bypass_vs_clip_and_viewport = TRUE; } rs->point_size = pack_float_16_6x(state->point_size) | @@ -474,10 +678,6 @@ static void* r300_create_rs_state(struct pipe_context* pipe, rs->color_control = R300_SHADE_MODEL_SMOOTH; } - if (!state->flatshade_first) { - rs->color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST; - } - return (void*)rs; } @@ -492,12 +692,23 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state) draw_set_rasterizer_state(r300->draw, &rs->rs); } - r300->rs_state = rs; + if (rs) { + r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport; + } else { + r300->tcl_bypass = FALSE; + } + + r300->rs_state.state = rs; + r300->rs_state.dirty = TRUE; + /* XXX Why is this still needed, dammit!? */ + r300->scissor_state.dirty = TRUE; + r300->viewport_state.dirty = TRUE; + /* XXX Clean these up when we move to atom emits */ - r300->dirty_state |= R300_NEW_RASTERIZER; r300->dirty_state |= R300_NEW_RS_BLOCK; - r300->dirty_state |= R300_NEW_SCISSOR; - r300->dirty_state |= R300_NEW_VIEWPORT; + if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { + r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; + } } /* Free rasterizer state. */ @@ -513,6 +724,9 @@ static void* struct r300_context* r300 = r300_context(pipe); struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state); int lod_bias; + union util_color uc; + + sampler->state = *state; sampler->filter0 |= (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) | @@ -521,7 +735,13 @@ static void* sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter, state->mag_img_filter, - state->min_mip_filter); + state->min_mip_filter, + state->max_anisotropy > 1.0); + + /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */ + /* We must pass these to the emit function to clamp them properly. */ + sampler->min_lod = MAX2((unsigned)state->min_lod, 0); + sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0); lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1); @@ -529,8 +749,8 @@ static void* sampler->filter1 |= r300_anisotropy(state->max_anisotropy); - util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, - &sampler->border_color); + util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc); + sampler->border_color = uc.ui; /* R500-specific fixups and optimizations */ if (r300_screen(r300->context.screen)->caps->is_r500) { @@ -559,6 +779,20 @@ static void r300_bind_sampler_states(struct pipe_context* pipe, } r300->sampler_count = count; + + /* Pick a fragment shader based on the texture compare state. */ + if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) { + if (r300_pick_fragment_shader(r300)) { + r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | + R300_NEW_FRAGMENT_SHADER_CONSTANTS; + } + } +} + +static void r300_lacks_vertex_textures(struct pipe_context* pipe, + unsigned count, + void** states) +{ } static void r300_delete_sampler_state(struct pipe_context* pipe, void* state) @@ -571,6 +805,7 @@ static void r300_set_sampler_textures(struct pipe_context* pipe, struct pipe_texture** texture) { struct r300_context* r300 = r300_context(pipe); + boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500; int i; /* XXX magic num */ @@ -578,13 +813,18 @@ static void r300_set_sampler_textures(struct pipe_context* pipe, return; } - r300->context.flush(&r300->context, 0, NULL); - for (i = 0; i < count; i++) { if (r300->textures[i] != (struct r300_texture*)texture[i]) { pipe_texture_reference((struct pipe_texture**)&r300->textures[i], texture[i]); r300->dirty_state |= (R300_NEW_TEXTURE << i); + + /* R300-specific - set the texrect factor in a fragment shader */ + if (!is_r500 && r300->textures[i]->is_npot) { + /* XXX It would be nice to re-emit just 1 constant, + * XXX not all of them */ + r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; + } } } @@ -604,60 +844,51 @@ static void r300_set_scissor_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - if (r300_screen(r300->context.screen)->caps->is_r500) { - r300->scissor_state->scissor_top_left = - (state->minx << R300_SCISSORS_X_SHIFT) | - (state->miny << R300_SCISSORS_Y_SHIFT); - r300->scissor_state->scissor_bottom_right = - ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) | - ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT); - } else { - /* Offset of 1440 in non-R500 chipsets. */ - r300->scissor_state->scissor_top_left = - ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) | - ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT); - r300->scissor_state->scissor_bottom_right = - (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) | - (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT); - } + memcpy(r300->scissor_state.state, state, + sizeof(struct pipe_scissor_state)); - r300->dirty_state |= R300_NEW_SCISSOR; + r300->scissor_state.dirty = TRUE; } static void r300_set_viewport_state(struct pipe_context* pipe, const struct pipe_viewport_state* state) { struct r300_context* r300 = r300_context(pipe); + struct r300_viewport_state* viewport = + (struct r300_viewport_state*)r300->viewport_state.state; /* Do the transform in HW. */ - r300->viewport_state->vte_control = R300_VTX_W0_FMT; + viewport->vte_control = R300_VTX_W0_FMT; if (state->scale[0] != 1.0f) { - r300->viewport_state->xscale = state->scale[0]; - r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA; + viewport->xscale = state->scale[0]; + viewport->vte_control |= R300_VPORT_X_SCALE_ENA; } if (state->scale[1] != 1.0f) { - r300->viewport_state->yscale = state->scale[1]; - r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA; + viewport->yscale = state->scale[1]; + viewport->vte_control |= R300_VPORT_Y_SCALE_ENA; } if (state->scale[2] != 1.0f) { - r300->viewport_state->zscale = state->scale[2]; - r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA; + viewport->zscale = state->scale[2]; + viewport->vte_control |= R300_VPORT_Z_SCALE_ENA; } if (state->translate[0] != 0.0f) { - r300->viewport_state->xoffset = state->translate[0]; - r300->viewport_state->vte_control |= R300_VPORT_X_OFFSET_ENA; + viewport->xoffset = state->translate[0]; + viewport->vte_control |= R300_VPORT_X_OFFSET_ENA; } if (state->translate[1] != 0.0f) { - r300->viewport_state->yoffset = state->translate[1]; - r300->viewport_state->vte_control |= R300_VPORT_Y_OFFSET_ENA; + viewport->yoffset = state->translate[1]; + viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA; } if (state->translate[2] != 0.0f) { - r300->viewport_state->zoffset = state->translate[2]; - r300->viewport_state->vte_control |= R300_VPORT_Z_OFFSET_ENA; + viewport->zoffset = state->translate[2]; + viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA; } - r300->dirty_state |= R300_NEW_VIEWPORT; + r300->viewport_state.dirty = TRUE; + if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) { + r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS; + } } static void r300_set_vertex_buffers(struct pipe_context* pipe, @@ -674,6 +905,8 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe, draw_flush(r300->draw); draw_set_vertex_buffers(r300->draw, count, buffers); } + + r300->dirty_state |= R300_NEW_VERTEX_FORMAT; } static void r300_set_vertex_elements(struct pipe_context* pipe, @@ -706,9 +939,6 @@ static void* r300_create_vs_state(struct pipe_context* pipe, tgsi_scan_shader(shader->tokens, &vs->info); - /* Appease Draw. */ - vs->draw = draw_create_vertex_shader(r300->draw, shader); - return (void*)vs; } else { return draw_create_vertex_shader(r300->draw, shader); @@ -719,8 +949,6 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) { struct r300_context* r300 = r300_context(pipe); - draw_flush(r300->draw); - if (r300_screen(pipe->screen)->caps->has_tcl) { struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; @@ -731,10 +959,16 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader) r300_translate_vertex_shader(r300, vs); } - draw_bind_vertex_shader(r300->draw, vs->draw); r300->vs = vs; - r300->dirty_state |= R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS; + if (r300->fs) { + r300_vertex_shader_setup_wpos(r300); + } + + r300->dirty_state |= + R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS | + R300_NEW_VERTEX_FORMAT; } else { + draw_flush(r300->draw); draw_bind_vertex_shader(r300->draw, (struct draw_vertex_shader*)shader); } @@ -748,7 +982,6 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader; rc_constants_destroy(&vs->code.constants); - draw_delete_vertex_shader(r300->draw, vs->draw); FREE((void*)vs->state.tokens); FREE(shader); } else { @@ -798,8 +1031,6 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state; r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state; - r300->context.set_edgeflags = r300_set_edgeflags; - r300->context.set_framebuffer_state = r300_set_framebuffer_state; r300->context.create_fs_state = r300_create_fs_state; @@ -813,10 +1044,11 @@ void r300_init_state_functions(struct r300_context* r300) r300->context.delete_rasterizer_state = r300_delete_rs_state; r300->context.create_sampler_state = r300_create_sampler_state; - r300->context.bind_sampler_states = r300_bind_sampler_states; + r300->context.bind_fragment_sampler_states = r300_bind_sampler_states; + r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures; r300->context.delete_sampler_state = r300_delete_sampler_state; - r300->context.set_sampler_textures = r300_set_sampler_textures; + r300->context.set_fragment_sampler_textures = r300_set_sampler_textures; r300->context.set_scissor_state = r300_set_scissor_state; diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 55430a9ad6..192846411b 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -1,5 +1,6 @@ /* * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -28,6 +29,7 @@ #include "r300_context.h" #include "r300_fs.h" #include "r300_screen.h" +#include "r300_shader_semantics.h" #include "r300_state_derived.h" #include "r300_state_inlines.h" #include "r300_vs.h" @@ -47,8 +49,8 @@ struct r300_shader_derived_value { unsigned r300_shader_key_hash(void* key) { struct r300_shader_key* shader_key = (struct r300_shader_key*)key; - unsigned vs = (unsigned)shader_key->vs; - unsigned fs = (unsigned)shader_key->fs; + unsigned vs = (intptr_t)shader_key->vs; + unsigned fs = (intptr_t)shader_key->fs; return (vs << 16) | (fs & 0xffff); } @@ -61,209 +63,152 @@ int r300_shader_key_compare(void* key1, void* key2) { (shader_key1->fs == shader_key2->fs); } -/* Set up the vs_tab and routes. */ -static void r300_vs_tab_routes(struct r300_context* r300, - struct r300_vertex_info* vformat) +static void r300_draw_emit_attrib(struct r300_context* r300, + enum attrib_emit emit, + enum interp_mode interp, + int index) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct vertex_info* vinfo = &vformat->vinfo; - int* tab = vformat->vs_tab; - boolean pos = FALSE, psize = FALSE, fog = FALSE; - int i, texs = 0, cols = 0; - struct tgsi_shader_info* info; + struct tgsi_shader_info* info = &r300->vs->info; + int output; - if (r300screen->caps->has_tcl) { - /* Use vertex shader to determine required routes. */ - info = &r300->vs->info; + output = draw_find_shader_output(r300->draw, + info->output_semantic_name[index], + info->output_semantic_index[index]); + draw_emit_vertex_attr(&r300->vertex_info->vinfo, emit, interp, output); +} + +static void r300_draw_emit_all_attribs(struct r300_context* r300) +{ + struct r300_shader_semantics* vs_outputs = &r300->vs->outputs; + int i, gen_count; + + /* Position. */ + if (vs_outputs->pos != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->pos); } else { - /* Use fragment shader to determine required routes. */ - info = &r300->fs->info; + assert(0); } - assert(info->num_inputs <= 16); - - if (!r300screen->caps->has_tcl || !r300->rs_state->enable_vte) - { - for (i = 0; i < info->num_inputs; i++) { - switch (r300->vs->code.inputs[i]) { - case TGSI_SEMANTIC_POSITION: - pos = TRUE; - tab[i] = 0; - break; - case TGSI_SEMANTIC_COLOR: - tab[i] = 2 + cols; - cols++; - break; - case TGSI_SEMANTIC_PSIZE: - assert(psize == FALSE); - psize = TRUE; - tab[i] = 15; - break; - case TGSI_SEMANTIC_FOG: - assert(fog == FALSE); - fog = TRUE; - /* Fall through */ - case TGSI_SEMANTIC_GENERIC: - tab[i] = 6 + texs; - texs++; - break; - default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); - break; - } - } + /* Point size. */ + if (vs_outputs->psize != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS, + vs_outputs->psize); } - else - { - /* Just copy vert attribs over as-is. */ - for (i = 0; i < info->num_inputs; i++) { - tab[i] = i; - } - for (i = 0; i < info->num_outputs; i++) { - switch (info->output_semantic_name[i]) { - case TGSI_SEMANTIC_POSITION: - pos = TRUE; - break; - case TGSI_SEMANTIC_COLOR: - cols++; - break; - case TGSI_SEMANTIC_PSIZE: - psize = TRUE; - break; - case TGSI_SEMANTIC_FOG: - fog = TRUE; - /* Fall through */ - case TGSI_SEMANTIC_GENERIC: - texs++; - break; - default: - debug_printf("r300: Unknown vertex output %d\n", - info->output_semantic_name[i]); - break; - } + /* Colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->color[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR, + vs_outputs->color[i]); } } - /* XXX magic */ - assert(texs <= 8); + /* XXX Back-face colors. */ - /* Do the actual vertex_info setup. - * - * vertex_info has four uints of hardware-specific data in it. - * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL - * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM - * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0 - * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */ - - vinfo->hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */ - - /* We need to add vertex position attribute only for SW TCL case, - * for HW TCL case it could be generated by vertex shader */ - if (!pos && !r300screen->caps->has_tcl) { - debug_printf("r300: Forcing vertex position attribute emit...\n"); - /* Make room for the position attribute - * at the beginning of the tab. */ - for (i = 15; i > 0; i--) { - tab[i] = tab[i-1]; + /* Texture coordinates. */ + gen_count = 0; + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (vs_outputs->generic[i] != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->generic[i]); + gen_count++; } - tab[0] = 0; } - /* Position. */ - if (r300->draw) { - draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0)); + /* Fog coordinates. */ + if (vs_outputs->fog != ATTR_UNUSED) { + r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE, + vs_outputs->fog); + gen_count++; } - vinfo->hwfmt[1] |= R300_INPUT_CNTL_POS; - vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; - /* Point size. */ - if (psize) { - if (r300->draw) { - draw_emit_vertex_attr(vinfo, EMIT_1F_PSIZE, INTERP_POS, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0)); - } - vinfo->hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; - } + /* XXX magic */ + assert(gen_count <= 8); +} - /* Colors. */ - for (i = 0; i < cols; i++) { - if (r300->draw) { - draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_LINEAR, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i)); - } - vinfo->hwfmt[1] |= R300_INPUT_CNTL_COLOR; - vinfo->hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i); +/* Update the PSC tables. */ +static void r300_vertex_psc(struct r300_context* r300) +{ + struct r300_vertex_info *vformat = r300->vertex_info; + uint16_t type, swizzle; + enum pipe_format format; + unsigned i; + int identity[16] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15}; + int* stream_tab; + + /* If TCL is bypassed, map vertex streams to equivalent VS output + * locations. */ + if (r300->tcl_bypass) { + stream_tab = r300->vs->stream_loc_notcl; + } else { + stream_tab = identity; } - /* Init i right here, increment it if fog is enabled. - * This gets around a double-increment problem. */ - i = 0; + /* Vertex shaders have no semantics on their inputs, + * so PSC should just route stuff based on the vertex elements, + * and not on attrib information. */ + DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements" + " in psc\n", + r300->vs->info.num_inputs, + r300->vertex_element_count); - /* Fog. This is a special-cased texcoord. */ - if (fog) { - i++; - if (r300->draw) { - draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0)); - } - vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i); - vinfo->hwfmt[3] |= (4 << (3 * i)); - } + for (i = 0; i < r300->vertex_element_count; i++) { + format = r300->vertex_element[i].src_format; + + type = r300_translate_vertex_data_type(format) | + (stream_tab[i] << R300_DST_VEC_LOC_SHIFT); + swizzle = r300_translate_vertex_data_swizzle(format); - /* Texcoords. */ - for (; i < texs; i++) { - if (r300->draw) { - draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, - draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i)); + if (i & 1) { + vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; + } else { + vformat->vap_prog_stream_cntl[i >> 1] |= type; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; } - vinfo->hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i); - vinfo->hwfmt[3] |= (4 << (3 * i)); } - draw_compute_vertex_size(vinfo); + assert(i <= 15); + + /* Set the last vector in the PSC. */ + if (i) { + i -= 1; + } + vformat->vap_prog_stream_cntl[i >> 1] |= + (R300_LAST_VEC << (i & 1 ? 16 : 0)); } -/* Update the PSC tables. */ -static void r300_vertex_psc(struct r300_context* r300, - struct r300_vertex_info* vformat) +/* Update the PSC tables for SW TCL, using Draw. */ +static void r300_swtcl_vertex_psc(struct r300_context* r300) { - struct r300_screen* r300screen = r300_screen(r300->context.screen); + struct r300_vertex_info *vformat = r300->vertex_info; struct vertex_info* vinfo = &vformat->vinfo; - int* tab = vformat->vs_tab; uint16_t type, swizzle; enum pipe_format format; unsigned i, attrib_count; + int* vs_output_tab = r300->vs->stream_loc_notcl; - /* Vertex shaders have no semantics on their inputs, - * so PSC should just route stuff based on their info, - * and not on attrib information. */ - if (r300screen->caps->has_tcl) { - attrib_count = r300->vs->info.num_inputs; - DBG(r300, DBG_DRAW, "r300: routing %d attribs in psc for vs\n", - attrib_count); - } else { - attrib_count = vinfo->num_attribs; - DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); - for (i = 0; i < attrib_count; i++) { - DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," - " tab %d\n", vinfo->attrib[i].src_index, - vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, - tab[i]); - } + /* For each Draw attribute, route it to the fragment shader according + * to the vs_output_tab. */ + attrib_count = vinfo->num_attribs; + DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count); + for (i = 0; i < attrib_count; i++) { + DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d," + " vs_output_tab %d\n", vinfo->attrib[i].src_index, + vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit, + vs_output_tab[i]); } for (i = 0; i < attrib_count; i++) { /* Make sure we have a proper destination for our attribute. */ - assert(tab[i] != -1); + assert(vs_output_tab[i] != -1); format = draw_translate_vinfo_format(vinfo->attrib[i].emit); /* Obtain the type of data in this attribute. */ type = r300_translate_vertex_data_type(format) | - tab[i] << R300_DST_VEC_LOC_SHIFT; + vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT; /* Obtain the swizzle for this attribute. Note that the default * swizzle in the hardware is not XYZW! */ @@ -272,12 +217,10 @@ static void r300_vertex_psc(struct r300_context* r300, /* Add the attribute to the PSC table. */ if (i & 1) { vformat->vap_prog_stream_cntl[i >> 1] |= type << 16; - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16; } else { - vformat->vap_prog_stream_cntl[i >> 1] |= type << 0; - - vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 0; + vformat->vap_prog_stream_cntl[i >> 1] |= type; + vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle; } } @@ -289,185 +232,212 @@ static void r300_vertex_psc(struct r300_context* r300, (R300_LAST_VEC << (i & 1 ? 16 : 0)); } -/* Set up the mappings from GB to US, for RS block. */ -static void r300_update_fs_tab(struct r300_context* r300, - struct r300_vertex_info* vformat) +static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr, + boolean swizzle_0001) +{ + rs->ip[id] |= R300_RS_COL_PTR(ptr); + if (swizzle_0001) { + rs->ip[id] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001); + } else { + rs->ip[id] |= R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA); + } + rs->inst[id] |= R300_RS_INST_COL_ID(id); +} + +static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset) { - struct tgsi_shader_info* info = &r300->fs->info; - int i, cols = 0, texs = 0, cols_emitted = 0; - int* tab = vformat->fs_tab; + rs->inst[id] |= R300_RS_INST_COL_CN_WRITE | + R300_RS_INST_COL_ADDR(fp_offset); +} - for (i = 0; i < 16; i++) { - tab[i] = -1; +static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr, + boolean swizzle_X001) +{ + if (swizzle_X001) { + rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + R300_RS_SEL_S(R300_RS_SEL_C0) | + R300_RS_SEL_T(R300_RS_SEL_K0) | + R300_RS_SEL_R(R300_RS_SEL_K0) | + R300_RS_SEL_Q(R300_RS_SEL_K1); + } else { + rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) | + R300_RS_SEL_S(R300_RS_SEL_C0) | + R300_RS_SEL_T(R300_RS_SEL_C1) | + R300_RS_SEL_R(R300_RS_SEL_C2) | + R300_RS_SEL_Q(R300_RS_SEL_C3); } + rs->inst[id] |= R300_RS_INST_TEX_ID(id); +} - assert(info->num_inputs <= 16); - for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - tab[i] = INTERP_LINEAR; - cols++; - break; - case TGSI_SEMANTIC_POSITION: - case TGSI_SEMANTIC_PSIZE: - debug_printf("r300: Implementation error: Can't use " - "pos attribs in fragshader yet!\n"); - /* Pass through for now */ - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - tab[i] = INTERP_PERSPECTIVE; - break; - default: - debug_printf("r300: Unknown vertex input %d\n", - info->input_semantic_name[i]); - break; - } +static void r300_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset) +{ + rs->inst[id] |= R300_RS_INST_TEX_CN_WRITE | + R300_RS_INST_TEX_ADDR(fp_offset); +} + +static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr, + boolean swizzle_0001) +{ + rs->ip[id] |= R500_RS_COL_PTR(ptr); + if (swizzle_0001) { + rs->ip[id] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001); + } else { + rs->ip[id] |= R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA); } + rs->inst[id] |= R500_RS_INST_COL_ID(id); +} - /* Now that we know where everything is... */ - DBG(r300, DBG_DRAW, "r300: fp input count: %d\n", info->num_inputs); - for (i = 0; i < info->num_inputs; i++) { - switch (tab[i]) { - case INTERP_LINEAR: - DBG(r300, DBG_DRAW, "r300: attrib: " - "stack offset %d, color, tab %d\n", - i, cols_emitted); - tab[i] = cols_emitted; - cols_emitted++; - break; - case INTERP_PERSPECTIVE: - DBG(r300, DBG_DRAW, "r300: attrib: " - "stack offset %d, texcoord, tab %d\n", - i, cols + texs); - tab[i] = cols + texs; - texs++; - break; - case -1: - debug_printf("r300: Implementation error: Bad fp interp!\n"); - default: - break; - } +static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset) +{ + rs->inst[id] |= R500_RS_INST_COL_CN_WRITE | + R500_RS_INST_COL_ADDR(fp_offset); +} + +static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr, + boolean swizzle_X001) +{ + int rs_tex_comp = ptr*4; + + if (swizzle_X001) { + rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | + R500_RS_SEL_T(R500_RS_IP_PTR_K0) | + R500_RS_SEL_R(R500_RS_IP_PTR_K0) | + R500_RS_SEL_Q(R500_RS_IP_PTR_K1); + } else { + rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) | + R500_RS_SEL_T(rs_tex_comp + 1) | + R500_RS_SEL_R(rs_tex_comp + 2) | + R500_RS_SEL_Q(rs_tex_comp + 3); } + rs->inst[id] |= R500_RS_INST_TEX_ID(id); +} +static void r500_rs_tex_write(struct r300_rs_block* rs, int id, int fp_offset) +{ + rs->inst[id] |= R500_RS_INST_TEX_CN_WRITE | + R500_RS_INST_TEX_ADDR(fp_offset); } -/* Set up the RS block. This is the part of the chipset that actually does - * the rasterization of vertices into fragments. This is also the part of the - * chipset that locks up if any part of it is even slightly wrong. */ +/* Set up the RS block. + * + * This is the part of the chipset that actually does the rasterization + * of vertices into fragments. This is also the part of the chipset that + * locks up if any part of it is even slightly wrong. */ static void r300_update_rs_block(struct r300_context* r300, - struct r300_rs_block* rs) + struct r300_shader_semantics* vs_outputs, + struct r300_shader_semantics* fs_inputs) { - struct tgsi_shader_info* info = &r300->fs->info; - int col_count = 0, fp_offset = 0, i, tex_count = 0; - int rs_tex_comp = 0; + struct r300_rs_block* rs = r300->rs_block; + int i, col_count = 0, tex_count = 0, fp_offset = 0; + void (*rX00_rs_col)(struct r300_rs_block*, int, int, boolean); + void (*rX00_rs_col_write)(struct r300_rs_block*, int, int); + void (*rX00_rs_tex)(struct r300_rs_block*, int, int, boolean); + void (*rX00_rs_tex_write)(struct r300_rs_block*, int, int); + boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || + vs_outputs->bcolor[1] != ATTR_UNUSED; if (r300_screen(r300->context.screen)->caps->is_r500) { - for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - rs->ip[col_count] |= - R500_RS_COL_PTR(col_count) | - R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA); - col_count++; - break; - case TGSI_SEMANTIC_GENERIC: - rs->ip[tex_count] |= - R500_RS_SEL_S(rs_tex_comp) | - R500_RS_SEL_T(rs_tex_comp + 1) | - R500_RS_SEL_R(rs_tex_comp + 2) | - R500_RS_SEL_Q(rs_tex_comp + 3); - tex_count++; - rs_tex_comp += 4; - break; - default: - break; - } - } + rX00_rs_col = r500_rs_col; + rX00_rs_col_write = r500_rs_col_write; + rX00_rs_tex = r500_rs_tex; + rX00_rs_tex_write = r500_rs_tex_write; + } else { + rX00_rs_col = r300_rs_col; + rX00_rs_col_write = r300_rs_col_write; + rX00_rs_tex = r300_rs_tex; + rX00_rs_tex_write = r300_rs_tex_write; + } - /* Rasterize at least one color, or bad things happen. */ - if ((col_count == 0) && (tex_count == 0)) { - rs->ip[0] |= R500_RS_COL_FMT(R300_RS_COL_FMT_0001); + /* Rasterize colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) { + /* Always rasterize if it's written by the VS, + * otherwise it locks up. */ + rX00_rs_col(rs, col_count, i, FALSE); + + /* Write it to the FS input register if it's used by the FS. */ + if (fs_inputs->color[i] != ATTR_UNUSED) { + rX00_rs_col_write(rs, col_count, fp_offset); + fp_offset++; + } col_count++; + } else { + /* Skip the FS input register, leave it uninitialized. */ + /* If we try to set it to (0,0,0,1), it will lock up. */ + if (fs_inputs->color[i] != ATTR_UNUSED) { + fp_offset++; + } } + } - for (i = 0; i < tex_count; i++) { - rs->inst[i] |= R500_RS_INST_TEX_ID(i) | - R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_offset); - fp_offset++; + /* Rasterize texture coordinates. */ + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (vs_outputs->generic[i] != ATTR_UNUSED) { + /* Always rasterize if it's written by the VS, + * otherwise it locks up. */ + rX00_rs_tex(rs, tex_count, tex_count, FALSE); + + /* Write it to the FS input register if it's used by the FS. */ + if (fs_inputs->generic[i] != ATTR_UNUSED) { + rX00_rs_tex_write(rs, tex_count, fp_offset); + fp_offset++; + } + tex_count++; + } else { + /* Skip the FS input register, leave it uninitialized. */ + /* If we try to set it to (0,0,0,1), it will lock up. */ + if (fs_inputs->generic[i] != ATTR_UNUSED) { + fp_offset++; + } } + } - for (i = 0; i < col_count; i++) { - rs->inst[i] |= R500_RS_INST_COL_ID(i) | - R500_RS_INST_COL_CN_WRITE | R500_RS_INST_COL_ADDR(fp_offset); + /* Rasterize fog coordinates. */ + if (vs_outputs->fog != ATTR_UNUSED) { + /* Always rasterize if it's written by the VS, + * otherwise it locks up. */ + rX00_rs_tex(rs, tex_count, tex_count, TRUE); + + /* Write it to the FS input register if it's used by the FS. */ + if (fs_inputs->fog != ATTR_UNUSED) { + rX00_rs_tex_write(rs, tex_count, fp_offset); fp_offset++; } + tex_count++; } else { - for (i = 0; i < info->num_inputs; i++) { - switch (info->input_semantic_name[i]) { - case TGSI_SEMANTIC_COLOR: - rs->ip[col_count] |= - R300_RS_COL_PTR(col_count) | - R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA); - col_count++; - break; - case TGSI_SEMANTIC_GENERIC: - rs->ip[tex_count] |= - R300_RS_TEX_PTR(rs_tex_comp) | - R300_RS_SEL_S(R300_RS_SEL_C0) | - R300_RS_SEL_T(R300_RS_SEL_C1) | - R300_RS_SEL_R(R300_RS_SEL_C2) | - R300_RS_SEL_Q(R300_RS_SEL_C3); - tex_count++; - rs_tex_comp+=4; - break; - default: - break; - } - } - - if (col_count == 0) { - rs->ip[0] |= R300_RS_COL_FMT(R300_RS_COL_FMT_0001); - } - - if (tex_count == 0) { - rs->ip[0] |= - R300_RS_SEL_S(R300_RS_SEL_K0) | - R300_RS_SEL_T(R300_RS_SEL_K0) | - R300_RS_SEL_R(R300_RS_SEL_K0) | - R300_RS_SEL_Q(R300_RS_SEL_K1); + /* Skip the FS input register, leave it uninitialized. */ + /* If we try to set it to (0,0,0,1), it will lock up. */ + if (fs_inputs->fog != ATTR_UNUSED) { + fp_offset++; } + } - /* Rasterize at least one color, or bad things happen. */ - if ((col_count == 0) && (tex_count == 0)) { - col_count++; - } + /* Rasterize WPOS. */ + /* If the FS doesn't need it, it's not written by the VS. */ + if (fs_inputs->wpos != ATTR_UNUSED) { + rX00_rs_tex(rs, tex_count, tex_count, FALSE); + rX00_rs_tex_write(rs, tex_count, fp_offset); - for (i = 0; i < tex_count; i++) { - rs->inst[i] |= R300_RS_INST_TEX_ID(i) | - R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_offset); - fp_offset++; - } + fp_offset++; + tex_count++; + } - for (i = 0; i < col_count; i++) { - rs->inst[i] |= R300_RS_INST_COL_ID(i) | - R300_RS_INST_COL_CN_WRITE | R300_RS_INST_COL_ADDR(fp_offset); - fp_offset++; - } + /* Rasterize at least one color, or bad things happen. */ + if (col_count == 0 && tex_count == 0) { + rX00_rs_col(rs, 0, 0, TRUE); + col_count++; } - rs->count = (rs_tex_comp) | (col_count << R300_IC_COUNT_SHIFT) | + rs->count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) | R300_HIRES_EN; - rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0); + rs->inst_count = MAX3(col_count - 1, tex_count - 1, 0); } /* Update the vertex format. */ static void r300_update_derived_shader_state(struct r300_context* r300) { - /* struct r300_screen* r300screen = r300_screen(r300->context.screen); */ - struct r300_vertex_info* vformat; - struct r300_rs_block* rs_block; - int i; + struct r300_screen* r300screen = r300_screen(r300->context.screen); /* struct r300_shader_key* key; @@ -495,32 +465,51 @@ static void r300_update_derived_shader_state(struct r300_context* r300) (void*)key, (void*)value); } */ - /* XXX This will be refactored ASAP. */ - vformat = CALLOC_STRUCT(r300_vertex_info); - rs_block = CALLOC_STRUCT(r300_rs_block); + /* Reset structures */ + memset(r300->rs_block, 0, sizeof(struct r300_rs_block)); + memset(r300->vertex_info, 0, sizeof(struct r300_vertex_info)); + memcpy(r300->vertex_info->vinfo.hwfmt, r300->vs->hwfmt, sizeof(uint)*4); + + r300_update_rs_block(r300, &r300->vs->outputs, &r300->fs->inputs); - for (i = 0; i < 16; i++) { - vformat->vs_tab[i] = -1; - vformat->fs_tab[i] = -1; + if (r300screen->caps->has_tcl) { + r300_vertex_psc(r300); + } else { + r300_draw_emit_all_attribs(r300); + draw_compute_vertex_size(&r300->vertex_info->vinfo); + r300_swtcl_vertex_psc(r300); } - r300_vs_tab_routes(r300, vformat); - r300_vertex_psc(r300, vformat); - r300_update_fs_tab(r300, vformat); + r300->dirty_state |= R300_NEW_RS_BLOCK; +} - r300_update_rs_block(r300, rs_block); +static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa) +{ + /* We are interested only in the cases when a new depth or stencil value + * can be written and changed. */ + + /* We might optionally check for [Z func: never] and inspect the stencil + * state in a similar fashion, but it's not terribly important. */ + return (dsa->z_buffer_control & R300_Z_WRITE_ENABLE) || + (dsa->stencil_ref_mask & R300_STENCILWRITEMASK_MASK) || + ((dsa->z_buffer_control & R500_STENCIL_REFMASK_FRONT_BACK) && + (dsa->stencil_ref_bf & R300_STENCILWRITEMASK_MASK)); +} - FREE(r300->vertex_info); - FREE(r300->rs_block); +static boolean r300_dsa_alpha_test_enabled(struct r300_dsa_state* dsa) +{ + /* We are interested only in the cases when alpha testing can kill + * a fragment. */ + uint32_t af = dsa->alpha_function; - r300->vertex_info = vformat; - r300->rs_block = rs_block; - r300->dirty_state |= (R300_NEW_VERTEX_FORMAT | R300_NEW_RS_BLOCK); + return (af & R300_FG_ALPHA_FUNC_ENABLE) && + (af & R300_FG_ALPHA_FUNC_ALWAYS) != R300_FG_ALPHA_FUNC_ALWAYS; } static void r300_update_ztop(struct r300_context* r300) { - r300->ztop_state.z_buffer_top = R300_ZTOP_ENABLE; + struct r300_ztop_state* ztop_state = + (struct r300_ztop_state*)r300->ztop_state.state; /* This is important enough that I felt it warranted a comment. * @@ -534,33 +523,45 @@ static void r300_update_ztop(struct r300_context* r300) * The docs claim that for the first three cases, if no ZS writes happen, * then ZTOP can be used. * + * (3) will never apply since we do not support chroma-keyed operations. + * (4) will need to be re-examined (and this comment updated) if/when + * Hyper-Z becomes supported. + * * Additionally, the following conditions require disabled ZTOP: - * ~) Depth writes in fragment shader - * ~) Outstanding occlusion queries + * 5) Depth writes in fragment shader + * 6) Outstanding occlusion queries + * + * This register causes stalls all the way from SC to CB when changed, + * but it is buffered on-chip so it does not hurt to write it if it has + * not changed. * * ~C. */ - if (r300->dsa_state->alpha_function) { - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; - } else if (r300->fs->info.uses_kill) { - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; - } else if (r300_fragment_shader_writes_depth(r300->fs)) { - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; - } else if (r300->query_current) { - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; + + /* ZS writes */ + if (r300_dsa_writes_depth_stencil(r300->dsa_state.state) && + (r300_dsa_alpha_test_enabled(r300->dsa_state.state) ||/* (1) */ + r300->fs->info.uses_kill)) { /* (2) */ + ztop_state->z_buffer_top = R300_ZTOP_DISABLE; + } else if (r300_fragment_shader_writes_depth(r300->fs)) { /* (5) */ + ztop_state->z_buffer_top = R300_ZTOP_DISABLE; + } else if (r300->query_current) { /* (6) */ + ztop_state->z_buffer_top = R300_ZTOP_DISABLE; + } else { + ztop_state->z_buffer_top = R300_ZTOP_ENABLE; } + + r300->ztop_state.dirty = TRUE; } void r300_update_derived_state(struct r300_context* r300) { /* XXX */ - if (TRUE || r300->dirty_state & - (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER)) { + if (r300->dirty_state & + (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER | + R300_NEW_VERTEX_FORMAT) || r300->rs_state.dirty) { r300_update_derived_shader_state(r300); } - if (r300->dirty_state & - (R300_NEW_DSA | R300_NEW_FRAGMENT_SHADER | R300_NEW_QUERY)) { - r300_update_ztop(r300); - } + r300_update_ztop(r300); } diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h index e6c1cb54da..35be00e1b0 100644 --- a/src/gallium/drivers/r300/r300_state_inlines.h +++ b/src/gallium/drivers/r300/r300_state_inlines.h @@ -28,6 +28,8 @@ #include "pipe/p_format.h" +#include "util/u_format.h" + #include "r300_reg.h" /* Some maths. These should probably find their way to u_math, if needed. */ @@ -255,38 +257,37 @@ static INLINE uint32_t r300_translate_wrap(int wrap) } } -static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip) +static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip, + int is_anisotropic) { uint32_t retval = 0; - switch (min) { + if (is_anisotropic) + retval |= R300_TX_MIN_FILTER_ANISO | R300_TX_MAG_FILTER_ANISO; + else { + switch (min) { case PIPE_TEX_FILTER_NEAREST: retval |= R300_TX_MIN_FILTER_NEAREST; break; case PIPE_TEX_FILTER_LINEAR: retval |= R300_TX_MIN_FILTER_LINEAR; break; - case PIPE_TEX_FILTER_ANISO: - retval |= R300_TX_MIN_FILTER_ANISO; - break; default: debug_printf("r300: Unknown texture filter %d\n", min); assert(0); break; - } - switch (mag) { + } + switch (mag) { case PIPE_TEX_FILTER_NEAREST: retval |= R300_TX_MAG_FILTER_NEAREST; break; case PIPE_TEX_FILTER_LINEAR: retval |= R300_TX_MAG_FILTER_LINEAR; break; - case PIPE_TEX_FILTER_ANISO: - retval |= R300_TX_MAG_FILTER_ANISO; - break; default: debug_printf("r300: Unknown texture filter %d\n", mag); assert(0); break; + } } switch (mip) { case PIPE_TEX_MIPFILTER_NONE: @@ -443,20 +444,22 @@ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count) static INLINE unsigned pf_component_count(enum pipe_format format) { unsigned count = 0; - if (pf_layout(format) != PIPE_FORMAT_LAYOUT_RGBAZS) { - return count; + if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) { + count++; } - - if (pf_size_x(format)) { + if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 1)) { count++; } - if (pf_size_y(format)) { + if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 2)) { count++; } - if (pf_size_z(format)) { + if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 3)) { count++; } - if (pf_size_w(format)) { + if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 0)) { + count++; + } + if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) { count++; } @@ -467,19 +470,23 @@ static INLINE unsigned pf_component_count(enum pipe_format format) { static INLINE uint16_t r300_translate_vertex_data_type(enum pipe_format format) { uint32_t result = 0; + const struct util_format_description *desc; unsigned components = pf_component_count(format); - if (pf_layout(format) != PIPE_FORMAT_LAYOUT_RGBAZS) { + desc = util_format_description(format); + + if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH && + desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) { debug_printf("r300: Bad format %s in %s:%d\n", pf_name(format), __FUNCTION__, __LINE__); assert(0); } - switch (pf_type(format)) { + switch (desc->channel[0].type) { /* Half-floats, floats, doubles */ - case PIPE_FORMAT_TYPE_FLOAT: - switch (pf_size_x(format)) { - case 4: + case UTIL_FORMAT_TYPE_FLOAT: + switch (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) { + case 32: result = R300_DATA_TYPE_FLOAT_1 + (components - 1); break; default: @@ -488,19 +495,15 @@ r300_translate_vertex_data_type(enum pipe_format format) { assert(0); } break; - /* Normalized unsigned ints */ - case PIPE_FORMAT_TYPE_UNORM: - /* Normalized signed ints */ - case PIPE_FORMAT_TYPE_SNORM: - /* Non-normalized unsigned ints */ - case PIPE_FORMAT_TYPE_USCALED: - /* Non-normalized signed ints */ - case PIPE_FORMAT_TYPE_SSCALED: - switch (pf_size_x(format)) { - case 1: + /* Unsigned ints */ + case UTIL_FORMAT_TYPE_UNSIGNED: + /* Signed ints */ + case UTIL_FORMAT_TYPE_SIGNED: + switch (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) { + case 8: result = R300_DATA_TYPE_BYTE; break; - case 2: + case 16: if (components > 2) { result = R300_DATA_TYPE_SHORT_4; } else { @@ -510,8 +513,8 @@ r300_translate_vertex_data_type(enum pipe_format format) { default: debug_printf("r300: Bad format %s in %s:%d\n", pf_name(format), __FUNCTION__, __LINE__); - debug_printf("r300: pf_size_x(format) == %d\n", - pf_size_x(format)); + debug_printf("r300: util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0) == %d\n", + util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)); assert(0); } break; @@ -521,12 +524,11 @@ r300_translate_vertex_data_type(enum pipe_format format) { assert(0); } - if (pf_type(format) == PIPE_FORMAT_TYPE_SSCALED) { + if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) { result |= R300_SIGNED; - } else if (pf_type(format) == PIPE_FORMAT_TYPE_UNORM) { + } + if (desc->channel[0].normalized) { result |= R300_NORMALIZE; - } else if (pf_type(format) == PIPE_FORMAT_TYPE_SNORM) { - result |= (R300_SIGNED | R300_NORMALIZE); } return result; @@ -534,17 +536,21 @@ r300_translate_vertex_data_type(enum pipe_format format) { static INLINE uint16_t r300_translate_vertex_data_swizzle(enum pipe_format format) { + const struct util_format_description *desc = util_format_description(format); + + assert(format); - if (pf_layout(format) != PIPE_FORMAT_LAYOUT_RGBAZS) { + if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH && + desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) { debug_printf("r300: Bad format %s in %s:%d\n", pf_name(format), __FUNCTION__, __LINE__); return 0; } - return ((pf_swizzle_x(format) << R300_SWIZZLE_SELECT_X_SHIFT) | - (pf_swizzle_y(format) << R300_SWIZZLE_SELECT_Y_SHIFT) | - (pf_swizzle_z(format) << R300_SWIZZLE_SELECT_Z_SHIFT) | - (pf_swizzle_w(format) << R300_SWIZZLE_SELECT_W_SHIFT) | + return ((desc->swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) | + (desc->swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) | + (desc->swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) | + (desc->swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) | (0xf << R300_WRITE_ENA_SHIFT)); } diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c index c07e6ae676..b0f309695c 100644 --- a/src/gallium/drivers/r300/r300_state_invariant.c +++ b/src/gallium/drivers/r300/r300_state_invariant.c @@ -43,7 +43,7 @@ void r300_emit_invariant_state(struct r300_context* r300) struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps; CS_LOCALS(r300); - BEGIN_CS(24 + (caps->has_tcl ? 2: 0)); + BEGIN_CS(16 + (caps->has_tcl ? 2: 0)); /*** Graphics Backend (GB) ***/ /* Various GB enables */ @@ -66,13 +66,8 @@ void r300_emit_invariant_state(struct r300_context* r300) OUT_CS_REG(R300_FG_FOG_COLOR_R, 0x0); OUT_CS_REG(R300_FG_FOG_COLOR_G, 0x0); OUT_CS_REG(R300_FG_FOG_COLOR_B, 0x0); - OUT_CS_REG(R300_FG_DEPTH_SRC, 0x0); - OUT_CS_REG(R300_US_W_FMT, 0x0); /*** VAP ***/ - /* Max and min vertex index clamp. */ - OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0x0); - OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, 0xffffff); /* Sign/normalize control */ OUT_CS_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO); /* TCL-only stuff */ @@ -84,15 +79,11 @@ void r300_emit_invariant_state(struct r300_context* r300) END_CS; /* XXX unsorted stuff from surface_fill */ - BEGIN_CS(60 + (caps->has_tcl ? 5 : 0) + (caps->is_r500 ? 4 : 0)); - /* Flush PVS. */ - OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0); + BEGIN_CS(44 + (caps->has_tcl ? 7 : 0) + (caps->is_r500 ? 4 : 0)); - OUT_CS_REG(R300_SE_VTE_CNTL, R300_VPORT_X_SCALE_ENA | - R300_VPORT_X_OFFSET_ENA | R300_VPORT_Y_SCALE_ENA | - R300_VPORT_Y_OFFSET_ENA | R300_VPORT_Z_SCALE_ENA | - R300_VPORT_Z_OFFSET_ENA | R300_VTX_W0_FMT); if (caps->has_tcl) { + /*Flushing PVS is required before the VAP_GB registers can be changed*/ + OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0); OUT_CS_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4); OUT_CS_32F(1.0); OUT_CS_32F(1.0); @@ -123,21 +114,17 @@ void r300_emit_invariant_state(struct r300_context* r300) OUT_CS_REG(R300_SU_DEPTH_OFFSET, 0x00000000); OUT_CS_REG(R300_SC_HYPERZ, 0x0000001C); OUT_CS_REG(R300_SC_EDGERULE, 0x2DA49525); - OUT_CS_REG(R300_RB3D_CCTL, 0x00000000); OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0x00000000); - if (caps->is_r500) { - OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x00000000); - OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFFFFFFFF); + + if (caps->family >= CHIP_FAMILY_RV350) { + OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101); + OUT_CS_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE); } - OUT_CS_REG(R300_ZB_FORMAT, 0x00000002); - OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT, 0x00000003); + OUT_CS_REG(R300_ZB_BW_CNTL, 0x00000000); OUT_CS_REG(R300_ZB_DEPTHCLEARVALUE, 0x00000000); OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0x00000000); OUT_CS_REG(R300_ZB_HIZ_PITCH, 0x00000000); - OUT_CS_REG(R300_VAP_VTX_STATE_CNTL, 0x1); - OUT_CS_REG(R300_VAP_VSM_VTX_ASSM, 0x405); - OUT_CS_REG(R300_SE_VTE_CNTL, 0x0000043F); /* XXX */ OUT_CS_REG(R300_SC_CLIP_RULE, 0xaaaa); diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index aea25cf71d..a9bbdd56d8 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -22,6 +22,7 @@ #include "pipe/p_screen.h" +#include "util/u_format.h" #include "util/u_math.h" #include "util/u_memory.h" @@ -29,13 +30,25 @@ #include "r300_texture.h" #include "r300_screen.h" +#define TILE_WIDTH 0 +#define TILE_HEIGHT 1 + +static const unsigned microblock_table[5][3][2] = { + /*linear tiled square-tiled */ + {{32, 1}, {8, 4}, {0, 0}}, /* 8 bits per pixel */ + {{16, 1}, {8, 2}, {4, 4}}, /* 16 bits per pixel */ + {{ 8, 1}, {4, 2}, {0, 0}}, /* 32 bits per pixel */ + {{ 4, 1}, {0, 0}, {2, 2}}, /* 64 bits per pixel */ + {{ 2, 1}, {0, 0}, {0, 0}} /* 128 bits per pixel */ +}; + static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500) { struct r300_texture_state* state = &tex->state; struct pipe_texture *pt = &tex->tex; - state->format0 = R300_TX_WIDTH((pt->width[0] - 1) & 0x7ff) | - R300_TX_HEIGHT((pt->height[0] - 1) & 0x7ff); + state->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) | + R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff); if (tex->is_npot) { /* rectangles love this */ @@ -43,8 +56,7 @@ static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500) state->format2 = (tex->pitch[0] - 1) & 0x1fff; } else { /* power of two textures (3D, mipmaps, and no pitch) */ - state->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth[0]) & 0xf) | - R300_TX_NUM_LEVELS(pt->last_level & 0xf); + state->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf); } state->format1 = r300_translate_texformat(pt->format); @@ -58,17 +70,17 @@ static void r300_setup_texture_state(struct r300_texture* tex, boolean is_r500) /* large textures on r500 */ if (is_r500) { - if (pt->width[0] > 2048) { + if (pt->width0 > 2048) { state->format2 |= R500_TXWIDTH_BIT11; } - if (pt->height[0] > 2048) { + if (pt->height0 > 2048) { state->format2 |= R500_TXHEIGHT_BIT11; } } - assert(is_r500 || (pt->width[0] <= 2048 && pt->height[0] <= 2048)); + assert(is_r500 || (pt->width0 <= 2048 && pt->height0 <= 2048)); debug_printf("r300: Set texture state (%dx%d, %d levels)\n", - pt->width[0], pt->height[0], pt->last_level); + pt->width0, pt->height0, pt->last_level); } unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level, @@ -92,63 +104,90 @@ unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level, } /** + * Return the width (dim==TILE_WIDTH) or height (dim==TILE_HEIGHT) of one tile + * of the given texture. + */ +static unsigned r300_texture_get_tile_size(struct r300_texture* tex, int dim) +{ + unsigned pixsize, tile_size; + + pixsize = util_format_get_blocksize(tex->tex.format); + tile_size = microblock_table[util_logbase2(pixsize)][tex->microtile][dim] * + (tex->macrotile == R300_BUFFER_TILED ? 8 : 1); + + assert(tile_size); + return tile_size; +} + +/** * Return the stride, in bytes, of the texture images of the given texture * at the given level. */ unsigned r300_texture_get_stride(struct r300_texture* tex, unsigned level) { + unsigned tile_width, width; + if (tex->stride_override) return tex->stride_override; + /* Check the level. */ if (level > tex->tex.last_level) { debug_printf("%s: level (%u) > last_level (%u)\n", __FUNCTION__, level, tex->tex.last_level); return 0; } - return align(pf_get_stride(&tex->tex.block, tex->tex.width[level]), 32); + tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH); + width = align(u_minify(tex->tex.width0, level), tile_width); + + /* Should already be aligned except for S3TC. */ + return align(util_format_get_stride(tex->tex.format, width), 32); +} + +static unsigned r300_texture_get_nblocksy(struct r300_texture* tex, + unsigned level) +{ + unsigned height, tile_height; + + tile_height = r300_texture_get_tile_size(tex, TILE_HEIGHT); + height = align(u_minify(tex->tex.height0, level), tile_height); + + return util_format_get_nblocksy(tex->tex.format, height); } static void r300_setup_miptree(struct r300_texture* tex) { struct pipe_texture* base = &tex->tex; - int stride, size, layer_size; - int i; + unsigned stride, size, layer_size, nblocksy, i; - for (i = 0; i <= base->last_level; i++) { - if (i > 0) { - base->width[i] = minify(base->width[i-1]); - base->height[i] = minify(base->height[i-1]); - base->depth[i] = minify(base->depth[i-1]); - } - - base->nblocksx[i] = pf_get_nblocksx(&base->block, base->width[i]); - base->nblocksy[i] = pf_get_nblocksy(&base->block, base->height[i]); + debug_printf("r300: Making miptree for texture, format %s\n", pf_name(base->format)); + for (i = 0; i <= base->last_level; i++) { stride = r300_texture_get_stride(tex, i); - layer_size = stride * base->nblocksy[i]; + nblocksy = r300_texture_get_nblocksy(tex, i); + layer_size = stride * nblocksy; if (base->target == PIPE_TEXTURE_CUBE) size = layer_size * 6; else - size = layer_size * base->depth[i]; + size = layer_size * u_minify(base->depth0, i); tex->offset[i] = align(tex->size, 32); tex->size = tex->offset[i] + size; tex->layer_size[i] = layer_size; - tex->pitch[i] = stride / base->block.size; + tex->pitch[i] = stride / util_format_get_blocksize(base->format); debug_printf("r300: Texture miptree: Level %d " - "(%dx%dx%d px, pitch %d bytes)\n", - i, base->width[i], base->height[i], base->depth[i], - stride); + "(%dx%dx%d px, pitch %d bytes) %d bytes total\n", + i, u_minify(base->width0, i), u_minify(base->height0, i), + u_minify(base->depth0, i), stride, tex->size); } } static void r300_setup_flags(struct r300_texture* tex) { - tex->is_npot = !util_is_power_of_two(tex->tex.width[0]) || - !util_is_power_of_two(tex->tex.height[0]); + tex->is_npot = !util_is_power_of_two(tex->tex.width0) || + !util_is_power_of_two(tex->tex.height0); } /* Create a new texture. */ @@ -170,7 +209,7 @@ static struct pipe_texture* r300_setup_miptree(tex); r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500); - tex->buffer = screen->buffer_create(screen, 1024, + tex->buffer = screen->buffer_create(screen, 2048, PIPE_BUFFER_USAGE_PIXEL, tex->size); @@ -208,8 +247,8 @@ static struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen, pipe_reference_init(&surface->reference, 1); pipe_texture_reference(&surface->texture, texture); surface->format = texture->format; - surface->width = texture->width[level]; - surface->height = texture->height[level]; + surface->width = u_minify(texture->width0, level); + surface->height = u_minify(texture->height0, level); surface->offset = offset; surface->usage = flags; surface->zslice = zslice; @@ -237,7 +276,7 @@ static struct pipe_texture* /* Support only 2D textures without mipmaps */ if (base->target != PIPE_TEXTURE_2D || - base->depth[0] != 1 || + base->depth0 != 1 || base->last_level != 0) { return NULL; } @@ -252,7 +291,7 @@ static struct pipe_texture* tex->tex.screen = screen; tex->stride_override = *stride; - tex->pitch[0] = *stride / base->block.size; + tex->pitch[0] = *stride / util_format_get_blocksize(base->format); r300_setup_flags(tex); r300_setup_texture_state(tex, r300_screen(screen)->caps->is_r500); @@ -287,10 +326,9 @@ r300_video_surface_create(struct pipe_screen *screen, template.target = PIPE_TEXTURE_2D; template.format = PIPE_FORMAT_X8R8G8B8_UNORM; template.last_level = 0; - template.width[0] = util_next_power_of_two(width); - template.height[0] = util_next_power_of_two(height); - template.depth[0] = 1; - pf_get_block(template.format, &template.block); + template.width0 = util_next_power_of_two(width); + template.height0 = util_next_power_of_two(height); + template.depth0 = 1; template.tex_usage = PIPE_TEXTURE_USAGE_SAMPLER | PIPE_TEXTURE_USAGE_RENDER_TARGET; diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c index 589f1984ee..a792c2cf98 100644 --- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c +++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c @@ -120,7 +120,7 @@ static unsigned translate_opcode(unsigned opcode) /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */ /* case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC; */ /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */ - /* case TGSI_OPCODE_SHR: return RC_OPCODE_SHR; */ + /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */ /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */ /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */ /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */ @@ -190,10 +190,10 @@ static void transform_dstreg( struct rc_dst_register * dst, struct tgsi_full_dst_register * src) { - dst->File = translate_register_file(src->DstRegister.File); - dst->Index = translate_register_index(ttr, src->DstRegister.File, src->DstRegister.Index); - dst->WriteMask = src->DstRegister.WriteMask; - dst->RelAddr = src->DstRegister.Indirect; + dst->File = translate_register_file(src->Register.File); + dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index); + dst->WriteMask = src->Register.WriteMask; + dst->RelAddr = src->Register.Indirect; } static void transform_srcreg( @@ -201,18 +201,19 @@ static void transform_srcreg( struct rc_src_register * dst, struct tgsi_full_src_register * src) { - dst->File = translate_register_file(src->SrcRegister.File); - dst->Index = translate_register_index(ttr, src->SrcRegister.File, src->SrcRegister.Index); - dst->RelAddr = src->SrcRegister.Indirect; + dst->File = translate_register_file(src->Register.File); + dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index); + dst->RelAddr = src->Register.Indirect; dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0); dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3; dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6; dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9; - dst->Abs = src->SrcRegisterExtMod.Absolute; - dst->Negate = src->SrcRegister.Negate ? RC_MASK_XYZW : 0; + dst->Abs = src->Register.Absolute; + dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0; } -static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_ext_texture src) +static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src, + uint32_t *shadowSamplers) { switch(src.Texture) { case TGSI_TEXTURE_1D: @@ -233,14 +234,17 @@ static void transform_texture(struct rc_instruction * dst, struct tgsi_instructi case TGSI_TEXTURE_SHADOW1D: dst->U.I.TexSrcTarget = RC_TEXTURE_1D; dst->U.I.TexShadow = 1; + *shadowSamplers |= 1 << dst->U.I.TexSrcUnit; break; case TGSI_TEXTURE_SHADOW2D: dst->U.I.TexSrcTarget = RC_TEXTURE_2D; dst->U.I.TexShadow = 1; + *shadowSamplers |= 1 << dst->U.I.TexSrcUnit; break; case TGSI_TEXTURE_SHADOWRECT: dst->U.I.TexSrcTarget = RC_TEXTURE_RECT; dst->U.I.TexShadow = 1; + *shadowSamplers |= 1 << dst->U.I.TexSrcUnit; break; } } @@ -258,17 +262,19 @@ static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_inst dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate); if (src->Instruction.NumDstRegs) - transform_dstreg(ttr, &dst->U.I.DstReg, &src->FullDstRegisters[0]); + transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]); for(i = 0; i < src->Instruction.NumSrcRegs; ++i) { - if (src->FullSrcRegisters[i].SrcRegister.File == TGSI_FILE_SAMPLER) - dst->U.I.TexSrcUnit = src->FullSrcRegisters[i].SrcRegister.Index; + if (src->Src[i].Register.File == TGSI_FILE_SAMPLER) + dst->U.I.TexSrcUnit = src->Src[i].Register.Index; else - transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->FullSrcRegisters[i]); + transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]); } /* Texturing. */ - transform_texture(dst, src->InstructionExtTexture); + if (src->Instruction.Texture) + transform_texture(dst, src->Texture, + &ttr->compiler->Program.ShadowSamplers); } static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm) diff --git a/src/gallium/drivers/r300/r300_vbo.c b/src/gallium/drivers/r300/r300_vbo.c deleted file mode 100644 index a6a159667a..0000000000 --- a/src/gallium/drivers/r300/r300_vbo.c +++ /dev/null @@ -1,132 +0,0 @@ -/* - * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> - * - * 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 - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR 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. - */ - -/* r300_vbo: Various helpers for emitting vertex buffers. Needs cleanup, - * refactoring, etc. */ - -#include "r300_vbo.h" - -#include "pipe/p_format.h" - -#include "r300_cs.h" -#include "r300_context.h" -#include "r300_state_inlines.h" -#include "r300_reg.h" -#include "r300_winsys.h" - -static INLINE void setup_vertex_attribute(struct r300_vertex_info *vinfo, - struct pipe_vertex_element *vert_elem, - unsigned attr_num) -{ - uint16_t hw_fmt1, hw_fmt2; - - hw_fmt1 = r300_translate_vertex_data_type(vert_elem->src_format) | - (attr_num << R300_DST_VEC_LOC_SHIFT); - hw_fmt2 = r300_translate_vertex_data_swizzle(vert_elem->src_format); - - if (attr_num % 2 == 0) - { - vinfo->vap_prog_stream_cntl[attr_num >> 1] = hw_fmt1; - vinfo->vap_prog_stream_cntl_ext[attr_num >> 1] = hw_fmt2; - } - else - { - vinfo->vap_prog_stream_cntl[attr_num >> 1] |= hw_fmt1 << 16; - vinfo->vap_prog_stream_cntl_ext[attr_num >> 1] |= hw_fmt2 << 16; - } -} - -static void finish_vertex_attribs_setup(struct r300_vertex_info *vinfo, - unsigned attribs_num) -{ - uint32_t last_vec_bit = (attribs_num % 2 == 0) ? - (R300_LAST_VEC << 16) : R300_LAST_VEC; - - assert(attribs_num > 0 && attribs_num <= 16); - vinfo->vap_prog_stream_cntl[(attribs_num - 1) >> 1] |= last_vec_bit; -} - -void setup_vertex_attributes(struct r300_context *r300) -{ - struct pipe_vertex_element *vert_elem; - int i; - - for (i = 0; i < r300->vertex_element_count; i++) { - vert_elem = &r300->vertex_element[i]; - setup_vertex_attribute(r300->vertex_info, vert_elem, i); - } - - finish_vertex_attribs_setup(r300->vertex_info, - r300->vertex_element_count); -} - -static INLINE int get_buffer_offset(struct r300_context *r300, - unsigned int buf_nr, - unsigned int elem_offset) -{ - return r300->vertex_buffer[buf_nr].buffer_offset + elem_offset; -} -#if 0 -/* XXX not called at all */ -static void setup_vertex_buffers(struct r300_context *r300) -{ - struct pipe_vertex_element *vert_elem; - int i; - - for (i = 0; i < r300->aos_count; i++) - { - vert_elem = &r300->vertex_element[i]; - /* XXX use translate module to convert the data */ - if (!format_is_supported(vert_elem->src_format, - vert_elem->nr_components)) { - assert(0); - /* - struct pipe_buffer *buf; - const unsigned int max_index = r300->vertex_buffers[vert_elem->vertex_buffer_index].max_index; - buf = pipe_buffer_create(r300->context.screen, 4, usage, vert_elem->nr_components * max_index * sizeof(float)); - */ - } - - if (get_buffer_offset(r300, - vert_elem->vertex_buffer_index, - vert_elem->src_offset) % 4) { - /* XXX need to align buffer */ - assert(0); - } - } -} -#endif -/* XXX these shouldn't be asserts since we can work around bad indexbufs */ -void setup_index_buffer(struct r300_context *r300, - struct pipe_buffer* indexBuffer, - unsigned indexSize) -{ - if (!r300->winsys->add_buffer(r300->winsys, indexBuffer, - RADEON_GEM_DOMAIN_GTT, 0)) { - assert(0); - } - - if (!r300->winsys->validate(r300->winsys)) { - assert(0); - } -} diff --git a/src/gallium/drivers/r300/r300_vbo.h b/src/gallium/drivers/r300/r300_vbo.h deleted file mode 100644 index 7afa75899c..0000000000 --- a/src/gallium/drivers/r300/r300_vbo.h +++ /dev/null @@ -1,36 +0,0 @@ -/* - * Copyright 2009 Maciej Cencora <m.cencora@gmail.com> - * - * 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 - * on the rights to use, copy, modify, merge, publish, distribute, sub - * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL - * THE AUTHOR(S) AND/OR THEIR 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. - */ - -#ifndef R300_VBO_H -#define R300_VBO_H - -struct r300_context; -struct pipe_buffer; - -void setup_vertex_attributes(struct r300_context *r300); - -void setup_index_buffer(struct r300_context *r300, - struct pipe_buffer* indexBuffer, - unsigned indexSize); - -#endif diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c index 74ef416dc1..68aef70872 100644 --- a/src/gallium/drivers/r300/r300_vs.c +++ b/src/gallium/drivers/r300/r300_vs.c @@ -1,5 +1,6 @@ /* * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -21,91 +22,296 @@ * USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include "r300_vs.h" +#include "r300_fs.h" #include "r300_context.h" +#include "r300_screen.h" #include "r300_tgsi_to_rc.h" +#include "r300_reg.h" #include "tgsi/tgsi_dump.h" #include "tgsi/tgsi_parse.h" #include "radeon_compiler.h" +#include "util/u_math.h" -static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c) +/* Convert info about VS output semantics into r300_shader_semantics. */ +static void r300_shader_read_vs_outputs( + struct tgsi_shader_info* info, + struct r300_shader_semantics* vs_outputs) { - struct r300_vertex_shader * vs = c->UserData; - struct tgsi_shader_info* info = &vs->info; - struct tgsi_parse_context parser; - struct tgsi_full_declaration * decl; - boolean pointsize = FALSE; - int out_colors = 0; - int colors = 0; - int out_generic = 0; - int generic = 0; int i; + unsigned index; - /* Fill in the input mapping */ - for (i = 0; i < info->num_inputs; i++) - c->code->inputs[i] = i; + r300_shader_semantics_reset(vs_outputs); - /* Fill in the output mapping */ for (i = 0; i < info->num_outputs; i++) { + index = info->output_semantic_index[i]; + switch (info->output_semantic_name[i]) { + case TGSI_SEMANTIC_POSITION: + assert(index == 0); + vs_outputs->pos = i; + break; + case TGSI_SEMANTIC_PSIZE: - pointsize = TRUE; + assert(index == 0); + vs_outputs->psize = i; break; + case TGSI_SEMANTIC_COLOR: - out_colors++; + assert(index <= ATTR_COLOR_COUNT); + vs_outputs->color[index] = i; break; - case TGSI_SEMANTIC_FOG: + + case TGSI_SEMANTIC_BCOLOR: + assert(index <= ATTR_COLOR_COUNT); + vs_outputs->bcolor[index] = i; + break; + case TGSI_SEMANTIC_GENERIC: - out_generic++; + assert(index <= ATTR_GENERIC_COUNT); + vs_outputs->generic[index] = i; + break; + + case TGSI_SEMANTIC_FOG: + assert(index == 0); + vs_outputs->fog = i; break; + + case TGSI_SEMANTIC_EDGEFLAG: + assert(index == 0); + fprintf(stderr, "r300 VP: cannot handle edgeflag output\n"); + assert(0); + break; + default: + assert(0); } } +} - tgsi_parse_init(&parser, vs->state.tokens); +static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs) +{ + struct r300_shader_semantics* vs_outputs = &vs->outputs; + uint32_t* hwfmt = vs->hwfmt; + int i, gen_count; + boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || + vs_outputs->bcolor[1] != ATTR_UNUSED; - while (!tgsi_parse_end_of_tokens(&parser)) { - tgsi_parse_token(&parser); + /* Do the actual vertex_info setup. + * + * vertex_info has four uints of hardware-specific data in it. + * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL + * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM + * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0 + * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */ - if (parser.FullToken.Token.Type != TGSI_TOKEN_TYPE_DECLARATION) - continue; + hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */ - decl = &parser.FullToken.FullDeclaration; + /* Position. */ + if (vs_outputs->pos != ATTR_UNUSED) { + hwfmt[1] |= R300_INPUT_CNTL_POS; + hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT; + } else { + assert(0); + } - if (decl->Declaration.File != TGSI_FILE_OUTPUT) - continue; + /* Point size. */ + if (vs_outputs->psize != ATTR_UNUSED) { + hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT; + } - switch (decl->Semantic.SemanticName) { - case TGSI_SEMANTIC_POSITION: - c->code->outputs[decl->DeclarationRange.First] = 0; - break; - case TGSI_SEMANTIC_PSIZE: - c->code->outputs[decl->DeclarationRange.First] = 1; - break; - case TGSI_SEMANTIC_COLOR: - c->code->outputs[decl->DeclarationRange.First] = 1 + - (pointsize ? 1 : 0) + - colors++; - break; - case TGSI_SEMANTIC_FOG: - case TGSI_SEMANTIC_GENERIC: - c->code->outputs[decl->DeclarationRange.First] = 1 + - (pointsize ? 1 : 0) + - out_colors + - generic++; - break; - default: - debug_printf("r300: vs: Bad semantic declaration %d\n", - decl->Semantic.SemanticName); - break; + /* Colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) { + hwfmt[1] |= R300_INPUT_CNTL_COLOR; + hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i; + } + } + + /* Back-face colors. */ + if (any_bcolor_used) { + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + hwfmt[1] |= R300_INPUT_CNTL_COLOR; + hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i); + } + } + + /* Texture coordinates. */ + gen_count = 0; + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (vs_outputs->generic[i] != ATTR_UNUSED) { + hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count); + hwfmt[3] |= (4 << (3 * gen_count)); + gen_count++; } } - tgsi_parse_free(&parser); + /* Fog coordinates. */ + if (vs_outputs->fog != ATTR_UNUSED) { + hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count); + hwfmt[3] |= (4 << (3 * gen_count)); + gen_count++; + } + + /* XXX magic */ + assert(gen_count <= 8); + + /* WPOS. */ + vs->wpos_tex_output = gen_count; } +/* Sets up stream mapping to equivalent VS outputs if TCL is bypassed + * or isn't present. */ +static void r300_stream_locations_notcl( + struct r300_shader_semantics* vs_outputs, + int* stream_loc) +{ + int i, tabi = 0, gen_count; + boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED || + vs_outputs->bcolor[1] != ATTR_UNUSED; + + /* Position. */ + stream_loc[tabi++] = 0; + + /* Point size. */ + if (vs_outputs->psize != ATTR_UNUSED) { + stream_loc[tabi++] = 1; + } + + /* Colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used) { + stream_loc[tabi++] = 2 + i; + } + } + + /* Back-face colors. */ + if (any_bcolor_used) { + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + stream_loc[tabi++] = 4 + i; + } + } + + /* Texture coordinates. */ + gen_count = 0; + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (vs_outputs->generic[i] != ATTR_UNUSED) { + assert(tabi < 16); + stream_loc[tabi++] = 6 + gen_count; + gen_count++; + } + } + + /* Fog coordinates. */ + if (vs_outputs->fog != ATTR_UNUSED) { + assert(tabi < 16); + stream_loc[tabi++] = 6 + gen_count; + gen_count++; + } + + /* WPOS. */ + if (vs_outputs->wpos != ATTR_UNUSED) { + assert(tabi < 16); + stream_loc[tabi++] = 6 + gen_count; + gen_count++; + } + + for (; tabi < 16;) { + stream_loc[tabi++] = -1; + } +} + +static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c) +{ + struct r300_vertex_shader * vs = c->UserData; + struct r300_shader_semantics* outputs = &vs->outputs; + struct tgsi_shader_info* info = &vs->info; + int i, reg = 0; + boolean any_bcolor_used = outputs->bcolor[0] != ATTR_UNUSED || + outputs->bcolor[1] != ATTR_UNUSED; + + /* Fill in the input mapping */ + for (i = 0; i < info->num_inputs; i++) + c->code->inputs[i] = i; + + /* Position. */ + if (outputs->pos != ATTR_UNUSED) { + c->code->outputs[outputs->pos] = reg++; + } else { + assert(0); + } + + /* Point size. */ + if (outputs->psize != ATTR_UNUSED) { + c->code->outputs[outputs->psize] = reg++; + } + + /* If we're writing back facing colors we need to send + * four colors to make front/back face colors selection work. + * If the vertex program doesn't write all 4 colors, lets + * pretend it does by skipping output index reg so the colors + * get written into appropriate output vectors. + */ + + /* Colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (outputs->color[i] != ATTR_UNUSED) { + c->code->outputs[outputs->color[i]] = reg++; + } else if (any_bcolor_used) { + reg++; + } + } + + /* Back-face colors. */ + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + if (outputs->bcolor[i] != ATTR_UNUSED) { + c->code->outputs[outputs->bcolor[i]] = reg++; + } else if (any_bcolor_used) { + reg++; + } + } + + /* Texture coordinates. */ + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + if (outputs->generic[i] != ATTR_UNUSED) { + c->code->outputs[outputs->generic[i]] = reg++; + } + } + + /* Fog coordinates. */ + if (outputs->fog != ATTR_UNUSED) { + c->code->outputs[outputs->fog] = reg++; + } + + /* WPOS. */ + if (outputs->wpos != ATTR_UNUSED) { + c->code->outputs[outputs->wpos] = reg++; + } +} + +static void r300_insert_wpos(struct r300_vertex_program_compiler* c, + struct r300_shader_semantics* outputs) +{ + int i, lastOutput = 0; + + /* Find the max output index. */ + lastOutput = MAX2(lastOutput, outputs->psize); + for (i = 0; i < ATTR_COLOR_COUNT; i++) { + lastOutput = MAX2(lastOutput, outputs->color[i]); + lastOutput = MAX2(lastOutput, outputs->bcolor[i]); + } + for (i = 0; i < ATTR_GENERIC_COUNT; i++) { + lastOutput = MAX2(lastOutput, outputs->generic[i]); + } + lastOutput = MAX2(lastOutput, outputs->fog); + + /* Set WPOS after the last output. */ + lastOutput++; + rc_copy_output(&c->Base, 0, lastOutput); /* out[lastOutput] = out[0]; */ + outputs->wpos = lastOutput; +} void r300_translate_vertex_shader(struct r300_context* r300, struct r300_vertex_shader* vs) @@ -113,6 +319,9 @@ void r300_translate_vertex_shader(struct r300_context* r300, struct r300_vertex_program_compiler compiler; struct tgsi_to_rc ttr; + /* Initialize. */ + r300_shader_read_vs_outputs(&vs->info, &vs->outputs); + /* Setup the compiler */ rc_init(&compiler.Base); @@ -131,13 +340,19 @@ void r300_translate_vertex_shader(struct r300_context* r300, r300_tgsi_to_rc(&ttr, vs->state.tokens); - compiler.RequiredOutputs = ~(~0 << vs->info.num_outputs); + compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs+1)); compiler.SetHwInputOutput = &set_vertex_inputs_outputs; + /* Insert the WPOS output. */ + r300_insert_wpos(&compiler, &vs->outputs); + + r300_shader_vap_output_fmt(vs); + r300_stream_locations_notcl(&vs->outputs, vs->stream_loc_notcl); + /* Invoke the compiler */ r3xx_compile_vertex_program(&compiler); if (compiler.Base.Error) { - /* Todo: Fail gracefully */ + /* XXX We should fallback using Draw. */ fprintf(stderr, "r300 VP: Compiler error\n"); abort(); } @@ -146,3 +361,30 @@ void r300_translate_vertex_shader(struct r300_context* r300, rc_destroy(&compiler.Base); vs->translated = TRUE; } + +boolean r300_vertex_shader_setup_wpos(struct r300_context* r300) +{ + struct r300_vertex_shader* vs = r300->vs; + int tex_output = r300->vs->wpos_tex_output; + uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output; + uint32_t* hwfmt = vs->hwfmt; + + if (r300->fs->inputs.wpos != ATTR_UNUSED) { + /* Enable WPOS in VAP. */ + if (!(hwfmt[1] & tex_fmt)) { + hwfmt[1] |= tex_fmt; + hwfmt[3] |= (4 << (3 * tex_output)); + + assert(tex_output < 8); + return TRUE; + } + } else { + /* Disable WPOS in VAP. */ + if (hwfmt[1] & tex_fmt) { + hwfmt[1] &= ~tex_fmt; + hwfmt[3] &= ~(4 << (3 * tex_output)); + return TRUE; + } + } + return FALSE; +} diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h index 2a4ce315e3..18cfeee3cd 100644 --- a/src/gallium/drivers/r300/r300_vs.h +++ b/src/gallium/drivers/r300/r300_vs.h @@ -1,5 +1,6 @@ /* * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com> + * Copyright 2009 Marek Olšák <maraeo@gmail.com> * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), @@ -25,18 +26,25 @@ #include "pipe/p_state.h" #include "tgsi/tgsi_scan.h" - #include "radeon_code.h" +#include "r300_shader_semantics.h" + struct r300_context; struct r300_vertex_shader { /* Parent class */ struct pipe_shader_state state; + struct tgsi_shader_info info; + struct r300_shader_semantics outputs; + uint hwfmt[4]; + + /* Stream locations for SWTCL or if TCL is bypassed. */ + int stream_loc_notcl[16]; - /* Fallback shader, because Draw has issues */ - struct draw_vertex_shader* draw; + /* Output stream location for WPOS. */ + int wpos_tex_output; /* Has this shader been translated yet? */ boolean translated; @@ -45,10 +53,10 @@ struct r300_vertex_shader { struct r300_vertex_program_code code; }; - -extern struct r300_vertex_program_code r300_passthrough_vertex_shader; - void r300_translate_vertex_shader(struct r300_context* r300, struct r300_vertex_shader* vs); +/* Return TRUE if VAP (hwfmt) needs to be re-emitted. */ +boolean r300_vertex_shader_setup_wpos(struct r300_context* r300); + #endif /* R300_VS_H */ diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h index 864a6146b2..1ae6de70fe 100644 --- a/src/gallium/drivers/r300/r300_winsys.h +++ b/src/gallium/drivers/r300/r300_winsys.h @@ -35,76 +35,10 @@ extern "C" { #include "pipe/p_state.h" #include "pipe/internal/p_winsys_screen.h" -struct r300_winsys { - /* Parent class */ - struct pipe_winsys base; - - /* Opaque Radeon-specific winsys object. */ - void* radeon_winsys; - - /* PCI ID */ - uint32_t pci_id; - - /* GB pipe count */ - uint32_t gb_pipes; - - /* Z pipe count (rv530 only) */ - uint32_t z_pipes; - - /* GART size. */ - uint32_t gart_size; - - /* VRAM size. */ - uint32_t vram_size; - - /* Add a pipe_buffer to the list of buffer objects to validate. */ - boolean (*add_buffer)(struct r300_winsys* winsys, - struct pipe_buffer* pbuffer, - uint32_t rd, - uint32_t wd); - - /* Revalidate all currently setup pipe_buffers. - * Returns TRUE if a flush is required. */ - boolean (*validate)(struct r300_winsys* winsys); - - /* Check to see if there's room for commands. */ - boolean (*check_cs)(struct r300_winsys* winsys, int size); - - /* Start a command emit. */ - void (*begin_cs)(struct r300_winsys* winsys, - int size, - const char* file, - const char* function, - int line); - - /* Write a dword to the command buffer. */ - void (*write_cs_dword)(struct r300_winsys* winsys, uint32_t dword); - - /* Write a relocated dword to the command buffer. */ - void (*write_cs_reloc)(struct r300_winsys* winsys, - struct pipe_buffer* bo, - uint32_t rd, - uint32_t wd, - uint32_t flags); - - /* Finish a command emit. */ - void (*end_cs)(struct r300_winsys* winsys, - const char* file, - const char* function, - int line); - - /* Flush the CS. */ - void (*flush_cs)(struct r300_winsys* winsys); - - /* winsys flush - callback from winsys when flush required */ - void (*set_flush_cb)(struct r300_winsys *winsys, - void (*flush_cb)(void *), void *data); - - void (*reset_bos)(struct r300_winsys *winsys); -}; +#include "radeon_winsys.h" struct pipe_context* r300_create_context(struct pipe_screen* screen, - struct r300_winsys* r300_winsys); + struct radeon_winsys* radeon_winsys); boolean r300_get_texture_buffer(struct pipe_texture* texture, struct pipe_buffer** buffer, |