diff options
| author | José Fonseca <[email protected]> | 2010-01-16 23:21:06 +0000 |
|---|---|---|
| committer | José Fonseca <[email protected]> | 2010-01-16 23:21:06 +0000 |
| commit | ba5d600c90f7075cf2c33a0c5c679ef822e8746e (patch) | |
| tree | ec6794fba9f2fe563a8bbc03ca06e288b7f51c28 /src/gallium/drivers | |
| parent | fdfe06ad804ea13e6e436d66c1bcafe0bde2f545 (diff) | |
| parent | 164fd16cfbc09970676c2e6866e062a5c9b410db (diff) | |
Merge remote branch 'origin/master' into lp-binning
Conflicts:
src/gallium/drivers/llvmpipe/lp_quad.h
src/gallium/drivers/llvmpipe/lp_setup.c
Diffstat (limited to 'src/gallium/drivers')
74 files changed, 1822 insertions, 838 deletions
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h index d5f5c7bbba8..aa29dcb3947 100644 --- a/src/gallium/drivers/cell/common.h +++ b/src/gallium/drivers/cell/common.h @@ -358,6 +358,7 @@ struct cell_spu_function_info /** This is the object passed to spe_create_thread() */ +PIPE_ALIGN_TYPE(16, struct cell_init_info { unsigned id; @@ -370,7 +371,7 @@ struct cell_init_info uint *buffer_status; /**< points at cell_context->buffer_status */ struct cell_spu_function_info *spu_functions; -} ALIGN16_ATTRIB; +}); #endif /* CELL_COMMON_H */ diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h index 5c3188e7f9d..e402ed29220 100644 --- a/src/gallium/drivers/cell/ppu/cell_context.h +++ b/src/gallium/drivers/cell/ppu/cell_context.h @@ -89,7 +89,7 @@ struct cell_buffer_node; */ struct cell_buffer_list { - struct cell_fence fence ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) struct cell_fence fence; struct cell_buffer_node *head; }; @@ -115,7 +115,7 @@ struct cell_context struct pipe_blend_color blend_color; struct pipe_clip_state clip; - struct pipe_constant_buffer constants[2]; + struct pipe_buffer *constants[2]; struct pipe_framebuffer_state framebuffer; struct pipe_poly_stipple poly_stipple; struct pipe_scissor_state scissor; @@ -150,18 +150,18 @@ struct cell_context /** Mapped constant buffers */ void *mapped_constants[PIPE_SHADER_TYPES]; - struct cell_spu_function_info spu_functions ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) struct cell_spu_function_info spu_functions; uint num_cells, num_spus; /** Buffers for command batches, vertex/index data */ uint buffer_size[CELL_NUM_BUFFERS]; - ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE] ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE]; int cur_batch; /**< which buffer is being filled w/ commands */ /** [4] to ensure 16-byte alignment for each status word */ - uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4] ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4]; /** Associated with each command/batch buffer is a list of pipe_buffers diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c index 3fa8b975d39..0a4da8ecc85 100644 --- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c +++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c @@ -51,17 +51,17 @@ cell_map_constant_buffers(struct cell_context *sp) struct pipe_winsys *ws = sp->pipe.winsys; uint i; for (i = 0; i < 2; i++) { - if (sp->constants[i].buffer && sp->constants[i].buffer->size) { - sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer, + if (sp->constants[i] && sp->constants[i]->size) { + sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i], PIPE_BUFFER_USAGE_CPU_READ); cell_flush_buffer_range(sp, sp->mapped_constants[i], - sp->constants[i].buffer->size); + sp->constants[i]->size); } } draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX, sp->mapped_constants[PIPE_SHADER_VERTEX], - sp->constants[PIPE_SHADER_VERTEX].buffer->size); + sp->constants[PIPE_SHADER_VERTEX]->size); } static void @@ -70,8 +70,8 @@ cell_unmap_constant_buffers(struct cell_context *sp) struct pipe_winsys *ws = sp->pipe.winsys; uint i; for (i = 0; i < 2; i++) { - if (sp->constants[i].buffer && sp->constants[i].buffer->size) - ws->buffer_unmap(ws, sp->constants[i].buffer); + if (sp->constants[i] && sp->constants[i]->size) + ws->buffer_unmap(ws, sp->constants[i]); sp->mapped_constants[i] = NULL; } } diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c index 5b87286d4c5..f1e1dcb9eb0 100644 --- a/src/gallium/drivers/cell/ppu/cell_state_emit.c +++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c @@ -240,12 +240,12 @@ cell_emit_state(struct cell_context *cell) if (cell->dirty & (CELL_NEW_FS_CONSTANTS)) { const uint shader = PIPE_SHADER_FRAGMENT; - const uint num_const = cell->constants[shader].buffer->size / sizeof(float); + const uint num_const = cell->constants[shader]->size / sizeof(float); uint i, j; float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float))); uint32_t *ibuf = (uint32_t *) buf; const float *constants = pipe_buffer_map(cell->pipe.screen, - cell->constants[shader].buffer, + cell->constants[shader], PIPE_BUFFER_USAGE_CPU_READ); ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS; ibuf[4] = num_const; @@ -253,7 +253,7 @@ cell_emit_state(struct cell_context *cell) for (i = 0; i < num_const; i++) { buf[j++] = constants[i]; } - pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader].buffer); + pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader]); } if (cell->dirty & (CELL_NEW_FRAMEBUFFER | diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c index 6568c784fec..1b09cf7f7d7 100644 --- a/src/gallium/drivers/cell/ppu/cell_state_shader.c +++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c @@ -183,7 +183,7 @@ cell_delete_vs_state(struct pipe_context *pipe, void *vs) static void cell_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct cell_context *cell = cell_context(pipe); @@ -193,7 +193,7 @@ cell_set_constant_buffer(struct pipe_context *pipe, draw_flush(cell->draw); /* note: reference counting */ - pipe_buffer_reference(&cell->constants[shader].buffer, buf->buffer); + pipe_buffer_reference(&cell->constants[shader], buf); if (shader == PIPE_SHADER_VERTEX) cell->dirty |= CELL_NEW_VS_CONSTANTS; diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c index 12b855a3db2..55bd85bde2b 100644 --- a/src/gallium/drivers/cell/spu/spu_command.c +++ b/src/gallium/drivers/cell/spu/spu_command.c @@ -53,8 +53,7 @@ struct spu_vs_context draw; /** * Buffers containing dynamically generated SPU code: */ -static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS] - ALIGN16_ATTRIB; +PIPE_ALIGN_VAR(16) static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS]; @@ -543,7 +542,7 @@ cmd_batch(uint opcode) { const uint buf = (opcode >> 8) & 0xff; uint size = (opcode >> 16); - qword buffer[CELL_BUFFER_SIZE / 16] ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) qword buffer[CELL_BUFFER_SIZE / 16]; const unsigned usize = ROUNDUP16(size) / sizeof(buffer[0]); uint pos; diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c index d86d8e09a51..d2166a49016 100644 --- a/src/gallium/drivers/cell/spu/spu_exec.c +++ b/src/gallium/drivers/cell/spu/spu_exec.c @@ -1839,10 +1839,11 @@ spu_exec_machine_run( struct spu_exec_machine *mach ) /* execute declarations (interpolants) */ if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) { for (i = 0; i < mach->NumDeclarations; i++) { + PIPE_ALIGN_VAR(16) union { struct tgsi_full_declaration decl; qword buffer[ROUNDUP16(sizeof(struct tgsi_full_declaration)) / 16]; - } d ALIGN16_ATTRIB; + } d; unsigned ea = (unsigned) (mach->Declarations + pc); spu_dcache_fetch_unaligned(d.buffer, ea, sizeof(d.decl)); @@ -1853,10 +1854,11 @@ spu_exec_machine_run( struct spu_exec_machine *mach ) /* execute instructions, until pc is set to -1 */ while (pc != -1) { + PIPE_ALIGN_VAR(16) union { struct tgsi_full_instruction inst; qword buffer[ROUNDUP16(sizeof(struct tgsi_full_instruction)) / 16]; - } i ALIGN16_ATTRIB; + } i; unsigned ea = (unsigned) (mach->Instructions + pc); spu_dcache_fetch_unaligned(i.buffer, ea, sizeof(i.inst)); diff --git a/src/gallium/drivers/cell/spu/spu_exec.h b/src/gallium/drivers/cell/spu/spu_exec.h index 86056799405..0ca92af248d 100644 --- a/src/gallium/drivers/cell/spu/spu_exec.h +++ b/src/gallium/drivers/cell/spu/spu_exec.h @@ -98,9 +98,9 @@ struct spu_exec_machine * 4 internal temporaries * 1 address */ + PIPE_ALIGN_VAR(16) struct spu_exec_vector Temps[TGSI_EXEC_NUM_TEMPS - + TGSI_EXEC_NUM_TEMP_EXTRAS + 1] - ALIGN16_ATTRIB; + + TGSI_EXEC_NUM_TEMP_EXTRAS + 1]; struct spu_exec_vector *Addrs; diff --git a/src/gallium/drivers/cell/spu/spu_funcs.c b/src/gallium/drivers/cell/spu/spu_funcs.c index ff3d609d258..98919c43ffc 100644 --- a/src/gallium/drivers/cell/spu/spu_funcs.c +++ b/src/gallium/drivers/cell/spu/spu_funcs.c @@ -144,7 +144,7 @@ export_func(struct cell_spu_function_info *spu_functions, void return_function_info(void) { - struct cell_spu_function_info funcs ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) struct cell_spu_function_info funcs; int tag = TAG_MISC; ASSERT(sizeof(funcs) == 256); /* must be multiple of 16 bytes */ diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h index 33767e7c51d..b18f4c22ef1 100644 --- a/src/gallium/drivers/cell/spu/spu_main.h +++ b/src/gallium/drivers/cell/spu/spu_main.h @@ -93,6 +93,7 @@ typedef vector unsigned int (*spu_fragment_program_func)(vector float *inputs, vector float *constants); +PIPE_ALIGN_TYPE(16, struct spu_framebuffer { void *color_start; /**< addr of color surface in main memory */ @@ -107,10 +108,11 @@ struct spu_framebuffer uint zsize; /**< 0, 2 or 4 bytes per Z */ float zscale; /**< 65535.0, 2^24-1 or 2^32-1 */ -} ALIGN16_ATTRIB; +}); /** per-texture level info */ +PIPE_ALIGN_TYPE(16, struct spu_texture_level { void *start; @@ -123,20 +125,22 @@ struct spu_texture_level vector signed int mask_s, mask_t, mask_r; /** texcoord clamp limits */ vector signed int max_s, max_t, max_r; -} ALIGN16_ATTRIB; +}); +PIPE_ALIGN_TYPE(16, struct spu_texture { struct spu_texture_level level[CELL_MAX_TEXTURE_LEVELS]; uint max_level; uint target; /**< PIPE_TEXTURE_x */ -} ALIGN16_ATTRIB; +}); /** * All SPU global/context state will be in a singleton object of this type: */ +PIPE_ALIGN_TYPE(16, struct spu_global { /** One-time init/constant info */ @@ -155,8 +159,8 @@ struct spu_global struct vertex_info vertex_info; /** Current color and Z tiles */ - tile_t ctile ALIGN16_ATTRIB; - tile_t ztile ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) tile_t ctile; + PIPE_ALIGN_VAR(16) tile_t ztile; /** Read depth/stencil tiles? */ boolean read_depth_stencil; @@ -165,8 +169,8 @@ struct spu_global ubyte cur_ctile_status, cur_ztile_status; /** Status of all tiles in framebuffer */ - ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB; - ubyte ztile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE]; + PIPE_ALIGN_VAR(16) ubyte ztile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE]; /** Current fragment ops machine code, at 8-byte boundary */ uint *fragment_ops_code; @@ -175,7 +179,7 @@ struct spu_global spu_fragment_ops_func fragment_ops[2]; /** Current fragment program machine code, at 8-byte boundary */ - uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS] ALIGN8_ATTRIB; + PIPE_ALIGN_VAR(8) uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS]; /** Current fragment ops function */ spu_fragment_program_func fragment_program; @@ -187,7 +191,7 @@ struct spu_global /** Fragment program constants */ vector float constants[4 * CELL_MAX_CONSTANTS]; -} ALIGN16_ATTRIB; +}); extern struct spu_global spu; diff --git a/src/gallium/drivers/cell/spu/spu_render.c b/src/gallium/drivers/cell/spu/spu_render.c index 5ffb7073abf..14987e3c3a2 100644 --- a/src/gallium/drivers/cell/spu/spu_render.c +++ b/src/gallium/drivers/cell/spu/spu_render.c @@ -169,7 +169,7 @@ void cmd_render(const struct cell_command_render *render, uint *pos_incr) { /* we'll DMA into these buffers */ - ubyte vertex_data[CELL_BUFFER_SIZE] ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) ubyte vertex_data[CELL_BUFFER_SIZE]; const uint vertex_size = render->vertex_size; /* in bytes */ /*const*/ uint total_vertex_bytes = render->num_verts * vertex_size; uint index_bytes; diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c index 03375d84a57..087963960df 100644 --- a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c +++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c @@ -43,7 +43,8 @@ typedef void (*spu_fetch_func)(qword *out, const qword *in, const qword *shuffle_data); -static const qword fetch_shuffle_data[5] ALIGN16_ATTRIB = { +PIPE_ALIGN_VAR(16) static const qword +fetch_shuffle_data[5] = { /* Shuffle used by CVT_64_FLOAT */ { @@ -110,7 +111,7 @@ static void generic_vertex_fetch(struct spu_vs_context *draw, unsigned idx; const unsigned bytes_per_entry = draw->vertex_fetch.size[attr]; const unsigned quads_per_entry = (bytes_per_entry + 15) / 16; - qword in[2 * 4] ALIGN16_ATTRIB; + PIPE_ALIGN_VAR(16) qword in[2 * 4]; /* Fetch four attributes for four vertices. diff --git a/src/gallium/drivers/cell/spu/spu_vertex_shader.c b/src/gallium/drivers/cell/spu/spu_vertex_shader.c index fbe5b34d397..3e9804bf8ee 100644 --- a/src/gallium/drivers/cell/spu/spu_vertex_shader.c +++ b/src/gallium/drivers/cell/spu/spu_vertex_shader.c @@ -107,8 +107,8 @@ run_vertex_program(struct spu_vs_context *draw, struct spu_exec_machine *machine = &draw->machine; unsigned int j; - ALIGN16_DECL(struct spu_exec_vector, inputs, PIPE_MAX_ATTRIBS); - ALIGN16_DECL(struct spu_exec_vector, outputs, PIPE_MAX_ATTRIBS); + PIPE_ALIGN_VAR(16) struct spu_exec_vector inputs[PIPE_MAX_ATTRIBS]; + PIPE_ALIGN_VAR(16) struct spu_exec_vector outputs[PIPE_MAX_ATTRIBS]; const float *scale = draw->viewport.scale; const float *trans = draw->viewport.translate; @@ -119,8 +119,8 @@ run_vertex_program(struct spu_vs_context *draw, ASSERT_ALIGN16(draw->constants); machine->Consts = (float (*)[4]) draw->constants; - machine->Inputs = ALIGN16_ASSIGN(inputs); - machine->Outputs = ALIGN16_ASSIGN(outputs); + machine->Inputs = inputs; + machine->Outputs = outputs; spu_vertex_fetch( draw, machine, elts, count ); @@ -132,8 +132,9 @@ run_vertex_program(struct spu_vs_context *draw, for (j = 0; j < count; j++) { unsigned slot; float x, y, z, w; + PIPE_ALIGN_VAR(16) unsigned char buffer[sizeof(struct vertex_header) - + MAX_VERTEX_SIZE] ALIGN16_ATTRIB; + + MAX_VERTEX_SIZE]; struct vertex_header *const tmpOut = (struct vertex_header *) buffer; const unsigned vert_size = ROUNDUP16(sizeof(struct vertex_header) @@ -186,8 +187,8 @@ run_vertex_program(struct spu_vs_context *draw, } -unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32] - ALIGN16_ATTRIB; +PIPE_ALIGN_VAR(16) unsigned char +immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]); void diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h index 149393712a3..191a44c3dfc 100644 --- a/src/gallium/drivers/failover/fo_context.h +++ b/src/gallium/drivers/failover/fo_context.h @@ -125,7 +125,7 @@ failover_context( struct pipe_context *pipe ) void failover_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf); + struct pipe_buffer *buf); #endif /* FO_CONTEXT_H */ diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c index 3f5f5560323..d6ec4d13136 100644 --- a/src/gallium/drivers/failover/fo_state.c +++ b/src/gallium/drivers/failover/fo_state.c @@ -495,7 +495,7 @@ failover_set_vertex_elements(struct pipe_context *pipe, void failover_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct failover_context *failover = failover_context(pipe); diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h index 234b441ce6e..37cbd56036b 100644 --- a/src/gallium/drivers/i915/i915_context.h +++ b/src/gallium/drivers/i915/i915_context.h @@ -233,7 +233,8 @@ struct i915_context struct pipe_blend_color blend_color; struct pipe_clip_state clip; - struct pipe_constant_buffer constants[PIPE_SHADER_TYPES]; + /* XXX unneded */ + struct pipe_buffer *constants[PIPE_SHADER_TYPES]; struct pipe_framebuffer_state framebuffer; struct pipe_poly_stipple poly_stipple; struct pipe_scissor_state scissor; diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c index 5f5b6f8e185..0fab6e1bc36 100644 --- a/src/gallium/drivers/i915/i915_state.c +++ b/src/gallium/drivers/i915/i915_state.c @@ -518,7 +518,7 @@ static void i915_delete_vs_state(struct pipe_context *pipe, void *shader) static void i915_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct i915_context *i915 = i915_context(pipe); struct pipe_screen *screen = pipe->screen; @@ -538,13 +538,13 @@ static void i915_set_constant_buffer(struct pipe_context *pipe, */ if (buf) { void *mapped; - if (buf->buffer && buf->buffer->size && - (mapped = pipe_buffer_map(screen, buf->buffer, + if (buf->size && + (mapped = pipe_buffer_map(screen, buf, PIPE_BUFFER_USAGE_CPU_READ))) { - memcpy(i915->current.constants[shader], mapped, buf->buffer->size); - pipe_buffer_unmap(screen, buf->buffer); + memcpy(i915->current.constants[shader], mapped, buf->size); + pipe_buffer_unmap(screen, buf); i915->current.num_user_constants[shader] - = buf->buffer->size / (4 * sizeof(float)); + = buf->size / (4 * sizeof(float)); } else { i915->current.num_user_constants[shader] = 0; diff --git a/src/gallium/drivers/i965/brw_pipe_shader.c b/src/gallium/drivers/i965/brw_pipe_shader.c index bb32d90e331..e389587f3e1 100644 --- a/src/gallium/drivers/i965/brw_pipe_shader.c +++ b/src/gallium/drivers/i965/brw_pipe_shader.c @@ -262,7 +262,7 @@ static void brw_delete_vs_state( struct pipe_context *pipe, void *prog ) static void brw_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct brw_context *brw = brw_context(pipe); @@ -270,13 +270,13 @@ static void brw_set_constant_buffer(struct pipe_context *pipe, if (shader == PIPE_SHADER_FRAGMENT) { pipe_buffer_reference( &brw->curr.fragment_constants, - buf->buffer ); + buf ); brw->state.dirty.mesa |= PIPE_NEW_FRAGMENT_CONSTANTS; } else { pipe_buffer_reference( &brw->curr.vertex_constants, - buf->buffer ); + buf ); brw->state.dirty.mesa |= PIPE_NEW_VERTEX_CONSTANTS; } diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c index 9f5b4e63236..f9063d90fb1 100644 --- a/src/gallium/drivers/identity/id_context.c +++ b/src/gallium/drivers/identity/id_context.c @@ -404,17 +404,17 @@ static void identity_set_constant_buffer(struct pipe_context *_pipe, uint shader, uint index, - const struct pipe_constant_buffer *_buffer) + struct pipe_buffer *_buffer) { struct identity_context *id_pipe = identity_context(_pipe); struct pipe_context *pipe = id_pipe->pipe; - struct pipe_constant_buffer unwrapped_buffer; - struct pipe_constant_buffer *buffer = NULL; + struct pipe_buffer *unwrapped_buffer; + struct pipe_buffer *buffer = NULL; - /* unwrap the input state */ + /* XXX hmm? unwrap the input state */ if (_buffer) { - unwrapped_buffer.buffer = identity_buffer_unwrap(_buffer->buffer); - buffer = &unwrapped_buffer; + unwrapped_buffer = identity_buffer_unwrap(_buffer); + buffer = unwrapped_buffer; } pipe->set_constant_buffer(pipe, diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README index 0c3f00fd58f..72d9f39658f 100644 --- a/src/gallium/drivers/llvmpipe/README +++ b/src/gallium/drivers/llvmpipe/README @@ -59,27 +59,16 @@ Requirements See /proc/cpuinfo to know what your CPU supports. - - LLVM 2.5 or greater. LLVM 2.6 is preferred. + - LLVM 2.6. - On Debian based distributions do: + For Linux, on a recent Debian based distribution do: aptitude install llvm-dev - There is a typo in one of the llvm 2.5 headers, that may cause compilation - errors. To fix it apply the change: - - --- /usr/include/llvm-c/Core.h.orig 2009-08-10 15:38:54.000000000 +0100 - +++ /usr/include/llvm-c/Core.h 2009-08-10 15:38:25.000000000 +0100 - @@ -831,7 +831,7 @@ - template<typename T> - inline T **unwrap(LLVMValueRef *Vals, unsigned Length) { - #if DEBUG - - for (LLVMValueRef *I = Vals, E = Vals + Length; I != E; ++I) - + for (LLVMValueRef *I = Vals, *E = Vals + Length; I != E; ++I) - cast<T>(*I); - #endif - return reinterpret_cast<T**>(Vals); - + For Windows download pre-built MSVC 9.0 or MinGW binaries from + http://people.freedesktop.org/~jrfonseca/llvm/ and set the LLVM environment + variable to the extracted path. + - scons (optional) - udis86, http://udis86.sourceforge.net/ (optional): @@ -95,9 +84,9 @@ Requirements Building ======== -To build everything invoke scons as: +To build everything on Linux invoke scons as: - scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false -k + scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=xlib dri=false Alternatively, you can build it with GNU make, if you prefer, by invoking it as @@ -105,12 +94,15 @@ Alternatively, you can build it with GNU make, if you prefer, by invoking it as but the rest of these instructions assume that scons is used. +For windows is everything the except except the winsys: + + scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=gdi dri=false Using ===== -Building will create a drop-in alternative for libGL.so. To use it set the -environment variables: +On Linux, building will create a drop-in alternative for libGL.so. To use it +set the environment variables: export LD_LIBRARY_PATH=$PWD/build/linux-x86_64-debug/lib:$LD_LIBRARY_PATH @@ -121,6 +113,11 @@ or For performance evaluation pass debug=no to scons, and use the corresponding lib directory without the "-debug" suffix. +On Windows, building will create a drop-in alternative for opengl32.dll. To use +it put it in the same directory as the application. It can also be used by +replacing the native ICD driver, but it's quite an advanced usage, so if you +need to ask, don't even try it. + Unit testing ============ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp b/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp index d3f78c06d92..6e79438ead0 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp +++ b/src/gallium/drivers/llvmpipe/lp_bld_misc.cpp @@ -59,3 +59,17 @@ LLVMInitializeNativeTarget(void) #endif + + +/* + * Hack to allow the linking of release LLVM static libraries on a debug build. + * + * See also: + * - http://social.msdn.microsoft.com/Forums/en-US/vclanguage/thread/7234ea2b-0042-42ed-b4e2-5d8644dfb57d + */ +#if defined(_MSC_VER) && defined(_DEBUG) +#include <crtdefs.h> +extern "C" { + _CRTIMP void __cdecl _invalid_parameter_noinfo(void) {} +} +#endif diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c index bd549d4028b..c5b00f8e23f 100644 --- a/src/gallium/drivers/llvmpipe/lp_context.c +++ b/src/gallium/drivers/llvmpipe/lp_context.c @@ -74,8 +74,8 @@ static void llvmpipe_destroy( struct pipe_context *pipe ) } for (i = 0; i < Elements(llvmpipe->constants); i++) { - if (llvmpipe->constants[i].buffer) { - pipe_buffer_reference(&llvmpipe->constants[i].buffer, NULL); + if (llvmpipe->constants[i]) { + pipe_buffer_reference(&llvmpipe->constants[i], NULL); } } diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h index 5390a713699..3af7b62a53b 100644 --- a/src/gallium/drivers/llvmpipe/lp_context.h +++ b/src/gallium/drivers/llvmpipe/lp_context.h @@ -62,7 +62,7 @@ struct llvmpipe_context { /** Other rendering state */ struct pipe_blend_color blend_color; struct pipe_clip_state clip; - struct pipe_constant_buffer constants[PIPE_SHADER_TYPES]; + struct pipe_buffer *constants[PIPE_SHADER_TYPES]; struct pipe_framebuffer_state framebuffer; struct pipe_poly_stipple poly_stipple; struct pipe_scissor_state scissor; diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h index 21ebfa7ca91..e4c56f153fc 100644 --- a/src/gallium/drivers/llvmpipe/lp_rast.h +++ b/src/gallium/drivers/llvmpipe/lp_rast.h @@ -86,7 +86,7 @@ struct lp_rast_shader_inputs { float (*dady)[4]; /* edge/step info for 3 edges and 4x4 block of pixels */ - int ALIGN16_ATTRIB step[3][16]; + PIPE_ALIGN_VAR(16) int step[3][16]; }; @@ -121,7 +121,7 @@ struct lp_rast_triangle { int c1, c2, c3; /* inputs for the shader */ - struct lp_rast_shader_inputs ALIGN16_ATTRIB inputs; + PIPE_ALIGN_VAR(16) struct lp_rast_shader_inputs inputs; }; diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h index 224b6e523c3..4242653c909 100644 --- a/src/gallium/drivers/llvmpipe/lp_state.h +++ b/src/gallium/drivers/llvmpipe/lp_state.h @@ -162,7 +162,7 @@ void llvmpipe_set_clip_state( struct pipe_context *, void llvmpipe_set_constant_buffer(struct pipe_context *, uint shader, uint index, - const struct pipe_constant_buffer *buf); + struct pipe_buffer *buf); void *llvmpipe_create_fs_state(struct pipe_context *, const struct pipe_shader_state *); diff --git a/src/gallium/drivers/llvmpipe/lp_state_derived.c b/src/gallium/drivers/llvmpipe/lp_state_derived.c index 28af477914c..bdd906e1a73 100644 --- a/src/gallium/drivers/llvmpipe/lp_state_derived.c +++ b/src/gallium/drivers/llvmpipe/lp_state_derived.c @@ -180,7 +180,7 @@ void llvmpipe_update_derived( struct llvmpipe_context *llvmpipe ) if (llvmpipe->dirty & LP_NEW_CONSTANTS) lp_setup_set_fs_constants(llvmpipe->setup, - llvmpipe->constants[PIPE_SHADER_FRAGMENT].buffer); + llvmpipe->constants[PIPE_SHADER_FRAGMENT]); if (llvmpipe->dirty & LP_NEW_TEXTURE) lp_setup_set_sampler_textures(llvmpipe->setup, diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c index a8f4a4ed463..4b51d6b964f 100644 --- a/src/gallium/drivers/llvmpipe/lp_state_fs.c +++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c @@ -998,23 +998,22 @@ llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs) void llvmpipe_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *constants) + struct pipe_buffer *constants) { struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe); - struct pipe_buffer *buffer = constants ? constants->buffer : NULL; - unsigned size = buffer ? buffer->size : 0; - const void *data = buffer ? llvmpipe_buffer(buffer)->data : NULL; + unsigned size = constants ? constants->size : 0; + const void *data = constants ? llvmpipe_buffer(constants)->data : NULL; assert(shader < PIPE_SHADER_TYPES); assert(index == 0); - if(llvmpipe->constants[shader].buffer == buffer) + if(llvmpipe->constants[shader] == constants) return; draw_flush(llvmpipe->draw); /* note: reference counting */ - pipe_buffer_reference(&llvmpipe->constants[shader].buffer, buffer); + pipe_buffer_reference(&llvmpipe->constants[shader], constants); if(shader == PIPE_SHADER_VERTEX) { draw_set_mapped_constant_buffer(llvmpipe->draw, PIPE_SHADER_VERTEX, diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c index 29fff91981a..6c29e8d8ace 100644 --- a/src/gallium/drivers/llvmpipe/lp_test_blend.c +++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c @@ -462,7 +462,7 @@ compute_blend_ref(const struct pipe_blend_state *blend, } -ALIGN_STACK +PIPE_ALIGN_STACK static boolean test_one(unsigned verbose, FILE *fp, @@ -531,11 +531,11 @@ test_one(unsigned verbose, success = TRUE; for(i = 0; i < n && success; ++i) { if(mode == AoS) { - ALIGN16_ATTRIB uint8_t src[LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t dst[LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t con[LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t res[LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t ref[LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t src[LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t dst[LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t con[LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t res[LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t ref[LP_NATIVE_VECTOR_WIDTH/8]; int64_t start_counter = 0; int64_t end_counter = 0; @@ -596,11 +596,11 @@ test_one(unsigned verbose, if(mode == SoA) { const unsigned stride = type.length*type.width/8; - ALIGN16_ATTRIB uint8_t src[4*LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t dst[4*LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t con[4*LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t res[4*LP_NATIVE_VECTOR_WIDTH/8]; - ALIGN16_ATTRIB uint8_t ref[4*LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t src[4*LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t dst[4*LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t con[4*LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t res[4*LP_NATIVE_VECTOR_WIDTH/8]; + PIPE_ALIGN_VAR(16) uint8_t ref[4*LP_NATIVE_VECTOR_WIDTH/8]; int64_t start_counter = 0; int64_t end_counter = 0; boolean mismatch; diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c index faddfb96779..c1abee424c9 100644 --- a/src/gallium/drivers/llvmpipe/lp_test_conv.c +++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c @@ -142,7 +142,7 @@ add_conv_test(LLVMModuleRef module, } -ALIGN_STACK +PIPE_ALIGN_STACK static boolean test_one(unsigned verbose, FILE *fp, @@ -230,8 +230,8 @@ test_one(unsigned verbose, for(i = 0; i < n && success; ++i) { unsigned src_stride = src_type.length*src_type.width/8; unsigned dst_stride = dst_type.length*dst_type.width/8; - ALIGN16_ATTRIB uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; - ALIGN16_ATTRIB uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; + PIPE_ALIGN_VAR(16) uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; + PIPE_ALIGN_VAR(16) uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]; int64_t start_counter = 0; diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c index 23ea9ebbe7d..2b258f1052e 100644 --- a/src/gallium/drivers/llvmpipe/lp_test_format.c +++ b/src/gallium/drivers/llvmpipe/lp_test_format.c @@ -199,7 +199,7 @@ add_store_rgba_test(LLVMModuleRef module, } -ALIGN_STACK +PIPE_ALIGN_STACK static boolean test_format(unsigned verbose, FILE *fp, const struct pixel_test_case *test) { diff --git a/src/gallium/drivers/nv04/nv04_state.c b/src/gallium/drivers/nv04/nv04_state.c index e3dc4c5bf44..b67f1e16b1c 100644 --- a/src/gallium/drivers/nv04/nv04_state.c +++ b/src/gallium/drivers/nv04/nv04_state.c @@ -332,7 +332,7 @@ nv04_set_clip_state(struct pipe_context *pipe, static void nv04_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf ) + struct pipe_buffer *buf ) { struct nv04_context *nv04 = nv04_context(pipe); struct pipe_screen *pscreen = pipe->screen; @@ -342,13 +342,13 @@ nv04_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf) { void *mapped; - if (buf->buffer && buf->buffer->size && - (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) + if (buf && buf->size && + (mapped = pipe_buffer_map(pscreen, buf, PIPE_BUFFER_USAGE_CPU_READ))) { - memcpy(nv04->constbuf[shader], mapped, buf->buffer->size); + memcpy(nv04->constbuf[shader], mapped, buf->size); nv04->constbuf_nr[shader] = - buf->buffer->size / (4 * sizeof(float)); - pipe_buffer_unmap(pscreen, buf->buffer); + buf->size / (4 * sizeof(float)); + pipe_buffer_unmap(pscreen, buf); } } } diff --git a/src/gallium/drivers/nv10/nv10_state.c b/src/gallium/drivers/nv10/nv10_state.c index ffc6be3c401..ad7def53b12 100644 --- a/src/gallium/drivers/nv10/nv10_state.c +++ b/src/gallium/drivers/nv10/nv10_state.c @@ -458,7 +458,7 @@ nv10_set_clip_state(struct pipe_context *pipe, static void nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf ) + struct pipe_buffer *buf ) { struct nv10_context *nv10 = nv10_context(pipe); struct pipe_screen *pscreen = pipe->screen; @@ -468,13 +468,13 @@ nv10_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf) { void *mapped; - if (buf->buffer && buf->buffer->size && - (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) + if (buf->size && + (mapped = pipe_buffer_map(pscreen, buf, PIPE_BUFFER_USAGE_CPU_READ))) { - memcpy(nv10->constbuf[shader], mapped, buf->buffer->size); + memcpy(nv10->constbuf[shader], mapped, buf->size); nv10->constbuf_nr[shader] = - buf->buffer->size / (4 * sizeof(float)); - pipe_buffer_unmap(pscreen, buf->buffer); + buf->size / (4 * sizeof(float)); + pipe_buffer_unmap(pscreen, buf); } } } diff --git a/src/gallium/drivers/nv20/nv20_state.c b/src/gallium/drivers/nv20/nv20_state.c index 3a82e63423d..45697a60efd 100644 --- a/src/gallium/drivers/nv20/nv20_state.c +++ b/src/gallium/drivers/nv20/nv20_state.c @@ -451,7 +451,7 @@ nv20_set_clip_state(struct pipe_context *pipe, static void nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf ) + struct pipe_buffer *buf ) { struct nv20_context *nv20 = nv20_context(pipe); struct pipe_screen *pscreen = pipe->screen; @@ -461,13 +461,13 @@ nv20_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, if (buf) { void *mapped; - if (buf->buffer && buf->buffer->size && - (mapped = pipe_buffer_map(pscreen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ))) + if (buf->size && + (mapped = pipe_buffer_map(pscreen, buf, PIPE_BUFFER_USAGE_CPU_READ))) { - memcpy(nv20->constbuf[shader], mapped, buf->buffer->size); + memcpy(nv20->constbuf[shader], mapped, buf->size); nv20->constbuf_nr[shader] = - buf->buffer->size / (4 * sizeof(float)); - pipe_buffer_unmap(pscreen, buf->buffer); + buf->size / (4 * sizeof(float)); + pipe_buffer_unmap(pscreen, buf); } } } diff --git a/src/gallium/drivers/nv30/nv30_fragtex.c b/src/gallium/drivers/nv30/nv30_fragtex.c index 98935678911..0cc3172dcd5 100644 --- a/src/gallium/drivers/nv30/nv30_fragtex.c +++ b/src/gallium/drivers/nv30/nv30_fragtex.c @@ -43,7 +43,6 @@ static struct nv30_texture_format * nv30_fragtex_format(uint pipe_format) { struct nv30_texture_format *tf = nv30_texture_formats; - char fs[128]; while (tf->defined) { if (tf->pipe == pipe_format) @@ -65,7 +64,7 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit) struct nouveau_bo *bo = nouveau_bo(nv30mt->buffer); struct nv30_texture_format *tf; struct nouveau_stateobj *so; - uint32_t txf, txs , txp; + uint32_t txf, txs; unsigned tex_flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD; tf = nv30_fragtex_format(pt->format); @@ -97,13 +96,6 @@ nv30_fragtex_build(struct nv30_context *nv30, int unit) return NULL; } - if (!(pt->tex_usage & NOUVEAU_TEXTURE_USAGE_LINEAR)) { - txp = 0; - } else { - txp = nv30mt->level[0].pitch; - txf |= (1<<13) /*FIXME: NV34TCL_TX_FORMAT_LINEAR ? */; - } - txs = tf->swizzle; so = so_new(1, 8, 2); diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c index a80dfb04880..065c927a10d 100644 --- a/src/gallium/drivers/nv30/nv30_state.c +++ b/src/gallium/drivers/nv30/nv30_state.c @@ -590,12 +590,12 @@ nv30_set_clip_state(struct pipe_context *pipe, static void nv30_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf ) + struct pipe_buffer *buf ) { struct nv30_context *nv30 = nv30_context(pipe); - nv30->constbuf[shader] = buf->buffer; - nv30->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float)); + nv30->constbuf[shader] = buf; + nv30->constbuf_nr[shader] = buf->size / (4 * sizeof(float)); if (shader == PIPE_SHADER_VERTEX) { nv30->dirty |= NV30_NEW_VERTPROG; diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c index ed0ca9e02c3..7d990f7d567 100644 --- a/src/gallium/drivers/nv40/nv40_state.c +++ b/src/gallium/drivers/nv40/nv40_state.c @@ -605,12 +605,12 @@ nv40_set_clip_state(struct pipe_context *pipe, static void nv40_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf ) + struct pipe_buffer *buf ) { struct nv40_context *nv40 = nv40_context(pipe); - nv40->constbuf[shader] = buf->buffer; - nv40->constbuf_nr[shader] = buf->buffer->size / (4 * sizeof(float)); + nv40->constbuf[shader] = buf; + nv40->constbuf_nr[shader] = buf->size / (4 * sizeof(float)); if (shader == PIPE_SHADER_VERTEX) { nv40->dirty |= NV40_NEW_VERTPROG; diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c index 5997456e4c9..22a8498c69b 100644 --- a/src/gallium/drivers/nv50/nv50_context.c +++ b/src/gallium/drivers/nv50/nv50_context.c @@ -67,8 +67,12 @@ nv50_destroy(struct pipe_context *pipe) so_ref(NULL, &nv50->state.vertprog); if (nv50->state.fragprog) so_ref(NULL, &nv50->state.fragprog); - if (nv50->state.programs) - so_ref(NULL, &nv50->state.programs); + if (nv50->state.geomprog) + so_ref(NULL, &nv50->state.geomprog); + if (nv50->state.fp_linkage) + so_ref(NULL, &nv50->state.fp_linkage); + if (nv50->state.gp_linkage) + so_ref(NULL, &nv50->state.gp_linkage); if (nv50->state.vtxfmt) so_ref(NULL, &nv50->state.vtxfmt); if (nv50->state.vtxbuf) diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h index cbd4c3ff86d..d024be6ea95 100644 --- a/src/gallium/drivers/nv50/nv50_context.h +++ b/src/gallium/drivers/nv50/nv50_context.h @@ -45,9 +45,11 @@ #define NV50_NEW_VERTPROG_CB (1 << 9) #define NV50_NEW_FRAGPROG (1 << 10) #define NV50_NEW_FRAGPROG_CB (1 << 11) -#define NV50_NEW_ARRAYS (1 << 12) -#define NV50_NEW_SAMPLER (1 << 13) -#define NV50_NEW_TEXTURE (1 << 14) +#define NV50_NEW_GEOMPROG (1 << 12) +#define NV50_NEW_GEOMPROG_CB (1 << 13) +#define NV50_NEW_ARRAYS (1 << 14) +#define NV50_NEW_SAMPLER (1 << 15) +#define NV50_NEW_TEXTURE (1 << 16) struct nv50_blend_stateobj { struct pipe_blend_state pipe; @@ -129,7 +131,9 @@ struct nv50_state { unsigned miptree_nr[PIPE_SHADER_TYPES]; struct nouveau_stateobj *vertprog; struct nouveau_stateobj *fragprog; - struct nouveau_stateobj *programs; + struct nouveau_stateobj *geomprog; + struct nouveau_stateobj *fp_linkage; + struct nouveau_stateobj *gp_linkage; struct nouveau_stateobj *vtxfmt; struct nouveau_stateobj *vtxbuf; struct nouveau_stateobj *vtxattr; @@ -157,6 +161,7 @@ struct nv50_context { struct pipe_framebuffer_state framebuffer; struct nv50_program *vertprog; struct nv50_program *fragprog; + struct nv50_program *geomprog; struct pipe_buffer *constbuf[PIPE_SHADER_TYPES]; struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS]; unsigned vtxbuf_nr; @@ -207,7 +212,9 @@ extern void nv50_clear(struct pipe_context *pipe, unsigned buffers, /* nv50_program.c */ extern void nv50_vertprog_validate(struct nv50_context *nv50); extern void nv50_fragprog_validate(struct nv50_context *nv50); -extern void nv50_linkage_validate(struct nv50_context *nv50); +extern void nv50_geomprog_validate(struct nv50_context *nv50); +extern void nv50_fp_linkage_validate(struct nv50_context *nv50); +extern void nv50_gp_linkage_validate(struct nv50_context *nv50); extern void nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p); diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c index 3f1edf0a139..dc8364ced7e 100644 --- a/src/gallium/drivers/nv50/nv50_miptree.c +++ b/src/gallium/drivers/nv50/nv50_miptree.c @@ -92,12 +92,23 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) case PIPE_FORMAT_Z24S8_UNORM: tile_flags = 0x1800; break; + case PIPE_FORMAT_Z16_UNORM: + tile_flags = 0x6c00; + break; case PIPE_FORMAT_X8Z24_UNORM: case PIPE_FORMAT_S8Z24_UNORM: tile_flags = 0x2800; break; + case PIPE_FORMAT_R32G32B32A32_FLOAT: + case PIPE_FORMAT_R32G32B32_FLOAT: + tile_flags = 0x7400; + break; default: - tile_flags = 0x7000; + if ((pt->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) && + util_format_get_blocksizebits(pt->format) == 32) + tile_flags = 0x7a00; + else + tile_flags = 0x7000; break; } @@ -145,7 +156,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp) mt->level[0].tile_mode, tile_flags, &mt->base.bo); if (ret) { - for (l = 0; l < pt->last_level; ++l) + for (l = 0; l <= pt->last_level; ++l) FREE(mt->level[l].image_offset); FREE(mt); return NULL; @@ -188,7 +199,7 @@ nv50_miptree_destroy(struct pipe_texture *pt) struct nv50_miptree *mt = nv50_miptree(pt); unsigned l; - for (l = 0; l < pt->last_level; ++l) + for (l = 0; l <= pt->last_level; ++l) FREE(mt->level[l].image_offset); nouveau_bo_ref(NULL, &mt->base.bo); diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 53f9f0adf31..acd695b39bc 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -92,6 +92,9 @@ struct nv50_reg { int rhw; /* result hw for FP outputs, or interpolant index */ int acc; /* instruction where this reg is last read (first insn == 1) */ + + int vtx; /* vertex index, for GP inputs (TGSI Dimension.Index) */ + int indirect[2]; /* index into pc->addr, or -1 */ }; #define NV50_MOD_NEG 1 @@ -135,9 +138,11 @@ struct nv50_pc { int immd_nr; struct nv50_reg **addr; int addr_nr; - uint8_t addr_alloc; /* set bit indicates used for TGSI_FILE_ADDRESS */ + struct nv50_reg *sysval; + int sysval_nr; struct nv50_reg *temp_temp[16]; + struct nv50_program_exec *temp_temp_exec[16]; unsigned temp_temp_nr; /* broadcast and destination replacement regs */ @@ -170,6 +175,8 @@ struct nv50_pc { uint8_t edgeflag_out; }; +static struct nv50_reg *get_address_reg(struct nv50_pc *, struct nv50_reg *); + static INLINE void ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw) { @@ -178,7 +185,9 @@ ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw) reg->hw = hw; reg->mod = 0; reg->rhw = -1; + reg->vtx = -1; reg->acc = 0; + reg->indirect[0] = reg->indirect[1] = -1; } static INLINE unsigned @@ -196,7 +205,8 @@ terminate_mbb(struct nv50_pc *pc) /* remove records of temporary address register values */ for (i = 0; i < NV50_SU_MAX_ADDR; ++i) - pc->r_addr[i].rhw = -1; + if (pc->r_addr[i].index < 0) + pc->r_addr[i].acc = 0; } static void @@ -245,7 +255,8 @@ alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg) } } - assert(0); + NOUVEAU_ERR("out of registers\n"); + abort(); } static INLINE struct nv50_reg * @@ -258,6 +269,7 @@ reg_instance(struct nv50_pc *pc, struct nv50_reg *reg) if (reg) { alloc_reg(pc, reg); *ri = *reg; + reg->indirect[0] = reg->indirect[1] = -1; reg->mod = 0; } return ri; @@ -285,7 +297,8 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst) } } - assert(0); + NOUVEAU_ERR("out of registers\n"); + abort(); return NULL; } @@ -347,23 +360,29 @@ free_temp4(struct nv50_pc *pc, struct nv50_reg *reg[4]) } static struct nv50_reg * -temp_temp(struct nv50_pc *pc) +temp_temp(struct nv50_pc *pc, struct nv50_program_exec *e) { if (pc->temp_temp_nr >= 16) assert(0); pc->temp_temp[pc->temp_temp_nr] = alloc_temp(pc, NULL); + pc->temp_temp_exec[pc->temp_temp_nr] = e; return pc->temp_temp[pc->temp_temp_nr++]; } +/* This *must* be called for all nv50_program_exec that have been + * given as argument to temp_temp, or the temps will be leaked ! + */ static void -kill_temp_temp(struct nv50_pc *pc) +kill_temp_temp(struct nv50_pc *pc, struct nv50_program_exec *e) { int i; for (i = 0; i < pc->temp_temp_nr; i++) - free_temp(pc, pc->temp_temp[i]); - pc->temp_temp_nr = 0; + if (pc->temp_temp_exec[i] == e) + free_temp(pc, pc->temp_temp[i]); + if (!e) + pc->temp_temp_nr = 0; } static int @@ -425,6 +444,8 @@ emit(struct nv50_pc *pc, struct nv50_program_exec *e) p->exec_head = e; p->exec_tail = e; p->exec_size += (e->inst[0] & 1) ? 2 : 1; + + kill_temp_temp(pc, e); } static INLINE void set_long(struct nv50_pc *, struct nv50_program_exec *); @@ -514,11 +535,33 @@ set_immd(struct nv50_pc *pc, struct nv50_reg *imm, struct nv50_program_exec *e) static INLINE void set_addr(struct nv50_program_exec *e, struct nv50_reg *a) { + assert(a->type == P_ADDR); + assert(!(e->inst[0] & 0x0c000000)); assert(!(e->inst[1] & 0x00000004)); e->inst[0] |= (a->hw & 3) << 26; - e->inst[1] |= (a->hw >> 2) << 2; + e->inst[1] |= a->hw & 4; +} + +static void +emit_arl(struct nv50_pc *, struct nv50_reg *, struct nv50_reg *, uint8_t); + +static void +emit_shl_imm(struct nv50_pc *, struct nv50_reg *, struct nv50_reg *, int); + +static void +emit_mov_from_addr(struct nv50_pc *pc, struct nv50_reg *dst, + struct nv50_reg *src) +{ + struct nv50_program_exec *e = exec(pc); + + e->inst[1] = 0x40000000; + set_long(pc, e); + set_dst(pc, dst, e); + set_addr(e, src); + + emit(pc, e); } static void @@ -537,72 +580,6 @@ emit_add_addr_imm(struct nv50_pc *pc, struct nv50_reg *dst, emit(pc, e); } -static struct nv50_reg * -alloc_addr(struct nv50_pc *pc, struct nv50_reg *ref) -{ - struct nv50_reg *a_tgsi = NULL, *a = NULL; - int i; - uint8_t avail = ~pc->addr_alloc; - - if (!ref) { - /* allocate for TGSI_FILE_ADDRESS */ - while (avail) { - i = ffs(avail) - 1; - - if (pc->r_addr[i].rhw < 0 || - pc->r_addr[i].acc != pc->insn_cur) { - pc->addr_alloc |= (1 << i); - - pc->r_addr[i].rhw = -1; - pc->r_addr[i].index = i; - return &pc->r_addr[i]; - } - avail &= ~(1 << i); - } - assert(0); - return NULL; - } - - /* Allocate and set an address reg so we can access 'ref'. - * - * If and r_addr->index will be -1 or the hw index the value - * value in rhw is relative to. If rhw < 0, the reg has not - * been initialized or is in use for TGSI_FILE_ADDRESS. - */ - while (avail) { /* only consider regs that are not TGSI */ - i = ffs(avail) - 1; - avail &= ~(1 << i); - - if ((!a || a->rhw >= 0) && pc->r_addr[i].rhw < 0) { - /* prefer an usused reg with low hw index */ - a = &pc->r_addr[i]; - continue; - } - if (!a && pc->r_addr[i].acc != pc->insn_cur) - a = &pc->r_addr[i]; - - if (ref->hw - pc->r_addr[i].rhw >= 128) - continue; - - if ((ref->acc >= 0 && pc->r_addr[i].index < 0) || - (ref->acc < 0 && pc->r_addr[i].index == ref->index)) { - pc->r_addr[i].acc = pc->insn_cur; - return &pc->r_addr[i]; - } - } - assert(a); - - if (ref->acc < 0) - a_tgsi = pc->addr[ref->index]; - - emit_add_addr_imm(pc, a, a_tgsi, (ref->hw & ~0x7f) * 4); - - a->rhw = ref->hw & ~0x7f; - a->acc = pc->insn_cur; - a->index = a_tgsi ? ref->index : -1; - return a; -} - #define INTERP_LINEAR 0 #define INTERP_FLAT 1 #define INTERP_PERSPECTIVE 2 @@ -646,12 +623,12 @@ set_data(struct nv50_pc *pc, struct nv50_reg *src, unsigned m, unsigned s, e->param.shift = s; e->param.mask = m << (s % 32); - if (src->hw > 127) - set_addr(e, alloc_addr(pc, src)); + if (src->hw < 0 || src->hw > 127) /* need (additional) address reg */ + set_addr(e, get_address_reg(pc, src)); else if (src->acc < 0) { assert(src->type == P_CONST); - set_addr(e, pc->addr[src->index]); + set_addr(e, pc->addr[src->indirect[0]]); } e->inst[1] |= (((src->type == P_IMMD) ? 0 : 1) << 22); @@ -683,6 +660,12 @@ emit_mov(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) if (src->type == P_ATTR) { set_long(pc, e); e->inst[1] |= 0x00200000; + + if (src->vtx >= 0) { + /* indirect (vertex base + c) load from p[] */ + e->inst[0] |= 0x01800000; + set_addr(e, get_address_reg(pc, src)); + } } alloc_reg(pc, src); @@ -780,7 +763,7 @@ set_src_0_restricted(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_reg *temp; if (src->type != P_TEMP) { - temp = temp_temp(pc); + temp = temp_temp(pc, e); emit_mov(pc, temp, src); src = temp; } @@ -797,9 +780,14 @@ set_src_0(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e) if (src->type == P_ATTR) { set_long(pc, e); e->inst[1] |= 0x00200000; + + if (src->vtx >= 0) { + e->inst[0] |= 0x01800000; /* src from p[] */ + set_addr(e, get_address_reg(pc, src)); + } } else if (src->type == P_CONST || src->type == P_IMMD) { - struct nv50_reg *temp = temp_temp(pc); + struct nv50_reg *temp = temp_temp(pc, e); emit_mov(pc, temp, src); src = temp; @@ -815,19 +803,19 @@ static void set_src_1(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e) { if (src->type == P_ATTR) { - struct nv50_reg *temp = temp_temp(pc); + struct nv50_reg *temp = temp_temp(pc, e); emit_mov(pc, temp, src); src = temp; } else if (src->type == P_CONST || src->type == P_IMMD) { - assert(!(e->inst[0] & 0x00800000)); - if (e->inst[0] & 0x01000000) { - struct nv50_reg *temp = temp_temp(pc); + if (e->inst[0] & 0x01800000) { + struct nv50_reg *temp = temp_temp(pc, e); emit_mov(pc, temp, src); src = temp; } else { + assert(!(e->inst[0] & 0x00800000)); set_data(pc, src, 0x7f, 16, e); e->inst[0] |= 0x00800000; } @@ -845,19 +833,19 @@ set_src_2(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e) set_long(pc, e); if (src->type == P_ATTR) { - struct nv50_reg *temp = temp_temp(pc); + struct nv50_reg *temp = temp_temp(pc, e); emit_mov(pc, temp, src); src = temp; } else if (src->type == P_CONST || src->type == P_IMMD) { - assert(!(e->inst[0] & 0x01000000)); - if (e->inst[0] & 0x00800000) { - struct nv50_reg *temp = temp_temp(pc); + if (e->inst[0] & 0x01800000) { + struct nv50_reg *temp = temp_temp(pc, e); emit_mov(pc, temp, src); src = temp; } else { + assert(!(e->inst[0] & 0x01000000)); set_data(pc, src, 0x7f, 32+14, e); e->inst[0] |= 0x01000000; } @@ -868,6 +856,26 @@ set_src_2(struct nv50_pc *pc, struct nv50_reg *src, struct nv50_program_exec *e) } static void +set_half_src(struct nv50_pc *pc, struct nv50_reg *src, int lh, + struct nv50_program_exec *e, int pos) +{ + struct nv50_reg *r = src; + + alloc_reg(pc, r); + if (r->type != P_TEMP) { + r = temp_temp(pc, e); + emit_mov(pc, r, src); + } + + if (r->hw > (NV50_SU_MAX_TEMP / 2)) { + NOUVEAU_ERR("out of low GPRs\n"); + abort(); + } + + e->inst[pos / 32] |= ((src->hw * 2) + lh) << (pos % 32); +} + +static void emit_mov_from_pred(struct nv50_pc *pc, struct nv50_reg *dst, int pred) { struct nv50_program_exec *e = exec(pc); @@ -966,11 +974,125 @@ emit_arl(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src, e->inst[0] |= dst->hw << 2; e->inst[0] |= s << 16; /* shift left */ - set_src_0_restricted(pc, src, e); + set_src_0(pc, src, e); emit(pc, e); } +static boolean +address_reg_suitable(struct nv50_reg *a, struct nv50_reg *r) +{ + if (!r) + return FALSE; + + if (r->vtx != a->vtx) + return FALSE; + if (r->vtx >= 0) + return (r->indirect[1] == a->indirect[1]); + + if (r->hw < a->rhw || (r->hw - a->rhw) >= 128) + return FALSE; + + if (a->index >= 0) + return (a->index == r->indirect[0]); + return (a->indirect[0] == r->indirect[0]); +} + +static void +load_vertex_base(struct nv50_pc *pc, struct nv50_reg *dst, + struct nv50_reg *a, int shift) +{ + struct nv50_reg mem, *temp; + + ctor_reg(&mem, P_ATTR, -1, dst->vtx); + + assert(dst->type == P_ADDR); + if (!a) { + emit_arl(pc, dst, &mem, 0); + return; + } + temp = alloc_temp(pc, NULL); + + if (shift) { + emit_mov_from_addr(pc, temp, a); + if (shift < 0) + emit_shl_imm(pc, temp, temp, shift); + emit_arl(pc, dst, temp, MAX2(shift, 0)); + } + emit_mov(pc, temp, &mem); + set_addr(pc->p->exec_tail, dst); + + emit_arl(pc, dst, temp, 0); + free_temp(pc, temp); +} + +/* case (ref == NULL): allocate address register for TGSI_FILE_ADDRESS + * case (vtx >= 0, acc >= 0): load vertex base from a[vtx * 4] to $aX + * case (vtx >= 0, acc < 0): load vertex base from s[$aY + vtx * 4] to $aX + * case (vtx < 0, acc >= 0): memory address too high to encode + * case (vtx < 0, acc < 0): get source register for TGSI_FILE_ADDRESS + */ +static struct nv50_reg * +get_address_reg(struct nv50_pc *pc, struct nv50_reg *ref) +{ + int i; + struct nv50_reg *a_ref, *a = NULL; + + for (i = 0; i < NV50_SU_MAX_ADDR; ++i) { + if (pc->r_addr[i].acc == 0) + a = &pc->r_addr[i]; /* an unused address reg */ + else + if (address_reg_suitable(&pc->r_addr[i], ref)) { + pc->r_addr[i].acc = pc->insn_cur; + return &pc->r_addr[i]; + } else + if (!a && pc->r_addr[i].index < 0 && + pc->r_addr[i].acc < pc->insn_cur) + a = &pc->r_addr[i]; + } + if (!a) { + /* We'll be able to spill address regs when this + * mess is replaced with a proper compiler ... + */ + NOUVEAU_ERR("out of address regs\n"); + abort(); + return NULL; + } + + /* initialize and reserve for this TGSI instruction */ + a->rhw = 0; + a->index = a->indirect[0] = a->indirect[1] = -1; + a->acc = pc->insn_cur; + + if (!ref) { + a->vtx = -1; + return a; + } + a->vtx = ref->vtx; + + /* now put in the correct value ... */ + + if (ref->vtx >= 0) { + a->indirect[1] = ref->indirect[1]; + + /* For an indirect vertex index, we need to shift address right + * by 2, the address register will contain vtx * 16, we need to + * load from a[vtx * 4]. + */ + load_vertex_base(pc, a, (ref->acc < 0) ? + pc->addr[ref->indirect[1]] : NULL, -2); + } else { + assert(ref->acc < 0 || ref->indirect[0] < 0); + + a->rhw = ref->hw & ~0x7f; + a->indirect[0] = ref->indirect[0]; + a_ref = (ref->acc < 0) ? pc->addr[ref->indirect[0]] : NULL; + + emit_add_addr_imm(pc, a, a_ref, a->rhw * 4); + } + return a; +} + #define NV50_MAX_F32 0x880 #define NV50_MAX_S32 0x08c #define NV50_MAX_U32 0x084 @@ -1050,6 +1172,20 @@ emit_bitop2(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, } static void +emit_not(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) +{ + struct nv50_program_exec *e = exec(pc); + + e->inst[0] = 0xd0000000; + e->inst[1] = 0x0402c000; + set_long(pc, e); + set_dst(pc, dst, e); + set_src_1(pc, src, e); + + emit(pc, e); +} + +static void emit_shift(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1, unsigned dir) { @@ -1078,6 +1214,27 @@ emit_shift(struct nv50_pc *pc, struct nv50_reg *dst, } static void +emit_shl_imm(struct nv50_pc *pc, struct nv50_reg *dst, + struct nv50_reg *src, int s) +{ + struct nv50_program_exec *e = exec(pc); + + e->inst[0] = 0x30000000; + e->inst[1] = 0xc4100000; + if (s < 0) { + e->inst[1] |= 1 << 29; + s = -s; + } + e->inst[1] |= ((s & 0x7f) << 16); + + set_long(pc, e); + set_dst(pc, dst, e); + set_src_0(pc, src, e); + + emit(pc, e); +} + +static void emit_mad(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src0, struct nv50_reg *src1, struct nv50_reg *src2) { @@ -1321,7 +1478,7 @@ emit_add_b32(struct nv50_pc *pc, struct nv50_reg *dst, src1 = rsrc1; if (src0->mod & rsrc1->mod & NV50_MOD_NEG) { - src1 = alloc_temp(pc, NULL); + src1 = temp_temp(pc, e); emit_cvt(pc, src1, rsrc1, -1, CVT_S32_S32); } @@ -1350,9 +1507,43 @@ emit_add_b32(struct nv50_pc *pc, struct nv50_reg *dst, e->inst[0] |= 1 << 22; emit(pc, e); +} + +static void +emit_mad_u16(struct nv50_pc *pc, struct nv50_reg *dst, + struct nv50_reg *src0, int lh_0, struct nv50_reg *src1, int lh_1, + struct nv50_reg *src2) +{ + struct nv50_program_exec *e = exec(pc); + + e->inst[0] = 0x60000000; + if (!pc->allow32) + set_long(pc, e); + set_dst(pc, dst, e); + + set_half_src(pc, src0, lh_0, e, 9); + set_half_src(pc, src1, lh_1, e, 16); + alloc_reg(pc, src2); + if (is_long(e) || (src2->type != P_TEMP) || (src2->hw != dst->hw)) + set_src_2(pc, src2, e); + + emit(pc, e); +} + +static void +emit_mul_u16(struct nv50_pc *pc, struct nv50_reg *dst, + struct nv50_reg *src0, int lh_0, struct nv50_reg *src1, int lh_1) +{ + struct nv50_program_exec *e = exec(pc); + + e->inst[0] = 0x40000000; + set_long(pc, e); + set_dst(pc, dst, e); + + set_half_src(pc, src0, lh_0, e, 9); + set_half_src(pc, src1, lh_1, e, 16); - if (src1 != rsrc1) - free_temp(pc, src1); + emit(pc, e); } static void @@ -1362,6 +1553,9 @@ emit_sad(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_program_exec *e = exec(pc); e->inst[0] = 0x50000000; + if (!pc->allow32) + set_long(pc, e); + check_swap_src_0_1(pc, &src0, &src1); set_dst(pc, dst, e); set_src_0(pc, src0, e); set_src_1(pc, src1, e); @@ -1373,6 +1567,8 @@ emit_sad(struct nv50_pc *pc, struct nv50_reg *dst, e->inst[1] |= 0x0c << 24; else e->inst[0] |= 0x81 << 8; + + emit(pc, e); } static INLINE void @@ -1422,10 +1618,10 @@ emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, if (mask & (1 << 2)) { set_pred_wr(pc, 1, 0, pc->p->exec_tail); - tmp[1] = temp_temp(pc); + tmp[1] = temp_temp(pc, NULL); emit_minmax(pc, NV50_MAX_F32, tmp[1], src[1], zero); - tmp[3] = temp_temp(pc); + tmp[3] = temp_temp(pc, NULL); emit_minmax(pc, NV50_MAX_F32, tmp[3], src[3], neg128); emit_minmax(pc, NV50_MIN_F32, tmp[3], tmp[3], pos128); @@ -1524,6 +1720,18 @@ emit_ret(struct nv50_pc *pc, int pred, unsigned cc) emit_control_flow(pc, 0x3, pred, cc); } +static void +emit_prim_cmd(struct nv50_pc *pc, unsigned cmd) +{ + struct nv50_program_exec *e = exec(pc); + + e->inst[0] = 0xf0000000 | (cmd << 9); + e->inst[1] = 0xc0000000; + set_long(pc, e); + + emit(pc, e); +} + #define QOP_ADD 0 #define QOP_SUBR 1 #define QOP_SUB 2 @@ -1884,7 +2092,11 @@ convert_to_long(struct nv50_pc *pc, struct nv50_program_exec *e) case 0x5: /* SAD */ m = ~(0x81 << 8); - q = 0x0c << 24; + q = (0x0c << 24) | ((e->inst[0] & (0x7f << 2)) << 12); + break; + case 0x6: + /* MAD u16 */ + q = (e->inst[0] & (0x7f << 2)) << 12; break; case 0x8: /* INTERP (move centroid, perspective and flat bits) */ @@ -1964,8 +2176,11 @@ get_supported_mods(const struct tgsi_full_instruction *insn, int i) case TGSI_OPCODE_IMAX: case TGSI_OPCODE_IMIN: case TGSI_OPCODE_ISHR: + case TGSI_OPCODE_NOT: + case TGSI_OPCODE_UMAD: case TGSI_OPCODE_UMAX: case TGSI_OPCODE_UMIN: + case TGSI_OPCODE_UMUL: case TGSI_OPCODE_USHR: return NV50_MOD_I32; default: @@ -2059,14 +2274,19 @@ tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst) { struct nv50_reg *r = pc->addr[dst->Register.Index * 4 + c]; if (!r) { - r = alloc_addr(pc, NULL); - pc->addr[dst->Register.Index * 4 + c] = r; + r = get_address_reg(pc, NULL); + r->index = dst->Register.Index * 4 + c; + pc->addr[r->index] = r; } assert(r); return r; } case TGSI_FILE_NULL: return NULL; + case TGSI_FILE_SYSTEM_VALUE: + assert(pc->sysval[dst->Register.Index].type == P_RESULT); + assert(c == 0); + return &pc->sysval[dst->Register.Index]; default: break; } @@ -2096,6 +2316,18 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, switch (src->Register.File) { case TGSI_FILE_INPUT: r = &pc->attr[src->Register.Index * 4 + c]; + + if (!src->Dimension.Dimension) + break; + r = reg_instance(pc, r); + r->vtx = src->Dimension.Index; + + if (!src->Dimension.Indirect) + break; + swz = tgsi_util_get_src_register_swizzle( + &src->DimIndirect, 0); + r->acc = -1; + r->indirect[1] = src->DimIndirect.Index * 4 + swz; break; case TGSI_FILE_TEMPORARY: r = &pc->temp[src->Register.Index * 4 + c]; @@ -2109,12 +2341,12 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, * use the index field to select the address reg. */ r = reg_instance(pc, NULL); + ctor_reg(r, P_CONST, -1, src->Register.Index * 4 + c); + swz = tgsi_util_get_src_register_swizzle( - &src->Indirect, 0); - ctor_reg(r, P_CONST, - src->Indirect.Index * 4 + swz, - src->Register.Index * 4 + c); + &src->Indirect, 0); r->acc = -1; + r->indirect[0] = src->Indirect.Index * 4 + swz; break; case TGSI_FILE_IMMEDIATE: r = &pc->immd[src->Register.Index * 4 + c]; @@ -2125,6 +2357,10 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, r = pc->addr[src->Register.Index * 4 + c]; assert(r); break; + case TGSI_FILE_SYSTEM_VALUE: + assert(c == 0); + r = &pc->sysval[src->Register.Index]; + break; default: assert(0); break; @@ -2153,7 +2389,7 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, } if ((r->mod & mod) != r->mod) { - temp = temp_temp(pc); + temp = temp_temp(pc, NULL); emit_cvt(pc, temp, r, -1, cvn); r->mod = 0; r = temp; @@ -2161,7 +2397,7 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, r->mod |= mod & NV50_MOD_I32; assert(r); - if (r->acc >= 0 && r != temp) + if (r->acc >= 0 && r->vtx < 0 && r != temp) return reg_instance(pc, r); /* will clear r->mod */ return r; } @@ -2341,7 +2577,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, brdc = temp = pc->r_brdc; if (brdc && brdc->type != P_TEMP) { - temp = temp_temp(pc); + temp = temp_temp(pc, NULL); if (sat) brdc = temp; } else @@ -2350,7 +2586,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, if (!(mask & (1 << c)) || dst[c]->type == P_TEMP) continue; /* rdst[c] = dst[c]; */ /* done above */ - dst[c] = temp_temp(pc); + dst[c] = temp_temp(pc, NULL); } } @@ -2383,10 +2619,14 @@ nv50_program_tx_insn(struct nv50_pc *pc, } break; case TGSI_OPCODE_ARL: - assert(src[0][0]); - temp = temp_temp(pc); - emit_cvt(pc, temp, src[0][0], -1, CVT_FLOOR | CVT_S32_F32); - emit_arl(pc, dst[0], temp, 4); + temp = temp_temp(pc, NULL); + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + emit_cvt(pc, temp, src[0][c], -1, + CVT_FLOOR | CVT_S32_F32); + emit_arl(pc, dst[c], temp, 4); + } break; case TGSI_OPCODE_BGNLOOP: pc->loop_brka[pc->loop_lvl] = emit_breakaddr(pc); @@ -2441,7 +2681,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, if (!(mask &= 7)) break; if (temp == dst[3]) - temp = brdc = temp_temp(pc); + temp = brdc = temp_temp(pc, NULL); } emit_precossin(pc, temp, src[0][0]); emit_flop(pc, NV50_FLOP_COS, brdc, temp); @@ -2493,6 +2733,9 @@ nv50_program_tx_insn(struct nv50_pc *pc, pc->if_insn[pc->if_lvl++] = pc->p->exec_tail; terminate_mbb(pc); break; + case TGSI_OPCODE_EMIT: + emit_prim_cmd(pc, 1); + break; case TGSI_OPCODE_ENDIF: pc->if_insn[--pc->if_lvl]->param.index = pc->p->exec_size; @@ -2516,8 +2759,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, pc->loop_brka[pc->loop_lvl]->param.index = pc->p->exec_size; terminate_mbb(pc); break; + case TGSI_OPCODE_ENDPRIM: + emit_prim_cmd(pc, 2); + break; case TGSI_OPCODE_ENDSUB: assert(pc->in_subroutine); + terminate_mbb(pc); pc->in_subroutine = FALSE; break; case TGSI_OPCODE_EX2: @@ -2529,8 +2776,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, struct nv50_reg *t[2]; assert(!temp); - t[0] = temp_temp(pc); - t[1] = temp_temp(pc); + t[0] = temp_temp(pc, NULL); + t[1] = temp_temp(pc, NULL); if (mask & 0x6) emit_mov(pc, t[0], src[0][0]); @@ -2575,7 +2822,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, } break; case TGSI_OPCODE_FRC: - temp = temp_temp(pc); + temp = temp_temp(pc, NULL); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; @@ -2639,9 +2886,9 @@ nv50_program_tx_insn(struct nv50_pc *pc, { struct nv50_reg *t[2]; - t[0] = temp_temp(pc); + t[0] = temp_temp(pc, NULL); if (mask & (1 << 1)) - t[1] = temp_temp(pc); + t[1] = temp_temp(pc, NULL); else t[1] = t[0]; @@ -2664,7 +2911,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, } break; case TGSI_OPCODE_LRP: - temp = temp_temp(pc); + temp = temp_temp(pc, NULL); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; @@ -2707,10 +2954,19 @@ nv50_program_tx_insn(struct nv50_pc *pc, emit_mul(pc, dst[c], src[0][c], src[1][c]); } break; + case TGSI_OPCODE_NOT: + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + emit_not(pc, dst[c], src[0][c]); + } + break; case TGSI_OPCODE_POW: emit_pow(pc, brdc, src[0][0], src[1][0]); break; case TGSI_OPCODE_RCP: + if (!sat && popcnt4(mask) == 1) + brdc = dst[ffs(mask) - 1]; emit_flop(pc, NV50_FLOP_RCP, brdc, src[0][0]); break; case TGSI_OPCODE_RET: @@ -2719,6 +2975,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, emit_ret(pc, -1, 0); break; case TGSI_OPCODE_RSQ: + if (!sat && popcnt4(mask) == 1) + brdc = dst[ffs(mask) - 1]; src[0][0]->mod |= NV50_MOD_ABS; emit_flop(pc, NV50_FLOP_RSQ, brdc, src[0][0]); break; @@ -2730,7 +2988,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, } break; case TGSI_OPCODE_SCS: - temp = temp_temp(pc); + temp = temp_temp(pc, NULL); if (mask & 3) emit_precossin(pc, temp, src[0][0]); if (mask & (1 << 0)) @@ -2759,7 +3017,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, if (!(mask &= 7)) break; if (temp == dst[3]) - temp = brdc = temp_temp(pc); + temp = brdc = temp_temp(pc, NULL); } emit_precossin(pc, temp, src[0][0]); emit_flop(pc, NV50_FLOP_SIN, brdc, temp); @@ -2847,8 +3105,41 @@ nv50_program_tx_insn(struct nv50_pc *pc, emit_minmax(pc, 0x0a4, dst[c], src[0][c], src[1][c]); } break; + case TGSI_OPCODE_UMAD: + { + assert(!temp); + temp = temp_temp(pc, NULL); + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + emit_mul_u16(pc, temp, src[0][c], 0, src[1][c], 1); + emit_mad_u16(pc, temp, src[0][c], 1, src[1][c], 0, + temp); + emit_shl_imm(pc, temp, temp, 16); + emit_mad_u16(pc, temp, src[0][c], 0, src[1][c], 0, + temp); + emit_add_b32(pc, dst[c], temp, src[2][c]); + } + } + break; + case TGSI_OPCODE_UMUL: + { + assert(!temp); + temp = temp_temp(pc, NULL); + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + emit_mul_u16(pc, temp, src[0][c], 0, src[1][c], 1); + emit_mad_u16(pc, temp, src[0][c], 1, src[1][c], 0, + temp); + emit_shl_imm(pc, temp, temp, 16); + emit_mad_u16(pc, dst[c], src[0][c], 0, src[1][c], 0, + temp); + } + } + break; case TGSI_OPCODE_XPD: - temp = temp_temp(pc); + temp = temp_temp(pc, NULL); if (mask & (1 << 0)) { emit_mul(pc, temp, src[0][2], src[1][1]); emit_msb(pc, dst[0], src[0][1], src[1][2], temp); @@ -2876,6 +3167,8 @@ nv50_program_tx_insn(struct nv50_pc *pc, emit_nop(pc); pc->p->exec_tail->inst[1] |= 1; /* set exit bit */ + + terminate_mbb(pc); break; default: NOUVEAU_ERR("invalid opcode %d\n", inst->Instruction.Opcode); @@ -2902,7 +3195,7 @@ nv50_program_tx_insn(struct nv50_pc *pc, } } - kill_temp_temp(pc); + kill_temp_temp(pc, NULL); pc->reg_instance_nr = 0; return TRUE; @@ -3101,7 +3394,7 @@ nv50_tgsi_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) if (is_scalar_op(insn.Instruction.Opcode)) { pc->r_brdc = tgsi_broadcast_dst(pc, fd, deqs); if (!pc->r_brdc) - pc->r_brdc = temp_temp(pc); + pc->r_brdc = temp_temp(pc, NULL); return nv50_program_tx_insn(pc, &insn); } pc->r_brdc = NULL; @@ -3171,17 +3464,40 @@ load_interpolant(struct nv50_pc *pc, struct nv50_reg *reg) * value of 0 for back-facing, and 0xffffffff for front-facing. */ static void -load_frontfacing(struct nv50_pc *pc, struct nv50_reg *a) +load_frontfacing(struct nv50_pc *pc, struct nv50_reg *sv) { - struct nv50_reg *one = alloc_immd(pc, 1.0f); + struct nv50_reg *temp = alloc_temp(pc, NULL); + int r_pred = 0; - assert(a->rhw == -1); - alloc_reg(pc, a); /* do this before rhw is set */ - a->rhw = 255; - load_interpolant(pc, a); - emit_bitop2(pc, a, a, one, TGSI_OPCODE_AND); + temp->rhw = 255; + emit_interp(pc, temp, NULL, INTERP_FLAT); - FREE(one); + emit_cvt(pc, sv, temp, r_pred, CVT_ABS | CVT_F32_S32); + + emit_not(pc, temp, temp); + set_pred(pc, 0x2, r_pred, pc->p->exec_tail); + emit_cvt(pc, sv, temp, -1, CVT_F32_S32); + set_pred(pc, 0x2, r_pred, pc->p->exec_tail); + + free_temp(pc, temp); +} + +static void +copy_semantic_info(struct nv50_program *p) +{ + unsigned i, id; + + for (i = 0; i < p->cfg.in_nr; ++i) { + id = p->cfg.in[i].id; + p->cfg.in[i].sn = p->info.input_semantic_name[id]; + p->cfg.in[i].si = p->info.input_semantic_index[id]; + } + + for (i = 0; i < p->cfg.out_nr; ++i) { + id = p->cfg.out[i].id; + p->cfg.out[i].sn = p->info.output_semantic_name[id]; + p->cfg.out[i].si = p->info.output_semantic_index[id]; + } } static boolean @@ -3190,7 +3506,7 @@ nv50_program_tx_prep(struct nv50_pc *pc) struct tgsi_parse_context tp; struct nv50_program *p = pc->p; boolean ret = FALSE; - unsigned i, c, flat_nr = 0; + unsigned i, c, instance_id, vertex_id, flat_nr = 0; tgsi_parse_init(&tp, pc->p->pipe.tokens); while (!tgsi_parse_end_of_tokens(&tp)) { @@ -3230,13 +3546,13 @@ nv50_program_tx_prep(struct nv50_pc *pc) switch (d->Semantic.Name) { case TGSI_SEMANTIC_BCOLOR: p->cfg.two_side[si].hw = first; - if (p->cfg.io_nr > first) - p->cfg.io_nr = first; + if (p->cfg.out_nr > first) + p->cfg.out_nr = first; break; case TGSI_SEMANTIC_PSIZE: p->cfg.psiz = first; - if (p->cfg.io_nr > first) - p->cfg.io_nr = first; + if (p->cfg.out_nr > first) + p->cfg.out_nr = first; break; case TGSI_SEMANTIC_EDGEFLAG: pc->edgeflag_out = first; @@ -3276,6 +3592,37 @@ nv50_program_tx_prep(struct nv50_pc *pc) pc->interp_mode[i] = mode; } break; + case TGSI_FILE_SYSTEM_VALUE: + assert(d->Declaration.Semantic); + switch (d->Semantic.Name) { + case TGSI_SEMANTIC_FACE: + assert(p->type == PIPE_SHADER_FRAGMENT); + load_frontfacing(pc, + &pc->sysval[first]); + break; + case TGSI_SEMANTIC_INSTANCEID: + assert(p->type == PIPE_SHADER_VERTEX); + instance_id = first; + p->cfg.regs[0] |= (1 << 4); + break; + case TGSI_SEMANTIC_PRIMID: + assert(p->type != PIPE_SHADER_VERTEX); + p->cfg.prim_id = first; + break; + /* + case TGSI_SEMANTIC_PRIMIDIN: + assert(p->type == PIPE_SHADER_GEOMETRY); + pc->sysval[first].hw = 6; + p->cfg.regs[0] |= (1 << 8); + break; + case TGSI_SEMANTIC_VERTEXID: + assert(p->type == PIPE_SHADER_VERTEX); + vertex_id = first; + p->cfg.regs[0] |= (1 << 12) | (1 << 0); + break; + */ + } + break; case TGSI_FILE_ADDRESS: case TGSI_FILE_CONSTANT: case TGSI_FILE_SAMPLER: @@ -3296,68 +3643,96 @@ nv50_program_tx_prep(struct nv50_pc *pc) } } - if (p->type == PIPE_SHADER_VERTEX) { + if (p->type == PIPE_SHADER_VERTEX || p->type == PIPE_SHADER_GEOMETRY) { int rid = 0; - for (i = 0; i < pc->attr_nr * 4; ++i) { - if (pc->attr[i].acc) { - pc->attr[i].hw = rid++; - p->cfg.attr[i / 32] |= 1 << (i % 32); + if (p->type == PIPE_SHADER_GEOMETRY) { + for (i = 0; i < pc->attr_nr; ++i) { + p->cfg.in[i].hw = rid; + p->cfg.in[i].id = i; + + for (c = 0; c < 4; ++c) { + int n = i * 4 + c; + if (!pc->attr[n].acc) + continue; + pc->attr[n].hw = rid++; + p->cfg.in[i].mask |= 1 << c; + } + } + } else { + for (i = 0; i < pc->attr_nr * 4; ++i) { + if (pc->attr[i].acc) { + pc->attr[i].hw = rid++; + p->cfg.attr[i / 32] |= 1 << (i % 32); + } } + if (p->cfg.regs[0] & (1 << 0)) + pc->sysval[vertex_id].hw = rid++; + if (p->cfg.regs[0] & (1 << 4)) + pc->sysval[instance_id].hw = rid++; } for (i = 0, rid = 0; i < pc->result_nr; ++i) { - p->cfg.io[i].hw = rid; - p->cfg.io[i].id = i; + p->cfg.out[i].hw = rid; + p->cfg.out[i].id = i; for (c = 0; c < 4; ++c) { int n = i * 4 + c; if (!pc->result[n].acc) continue; pc->result[n].hw = rid++; - p->cfg.io[i].mask |= 1 << c; + p->cfg.out[i].mask |= 1 << c; } } + if (p->cfg.prim_id < 0x40) { + /* GP has to write to PrimitiveID */ + ctor_reg(&pc->sysval[p->cfg.prim_id], + P_RESULT, p->cfg.prim_id, rid); + p->cfg.prim_id = rid++; + } for (c = 0; c < 2; ++c) if (p->cfg.two_side[c].hw < 0x40) - p->cfg.two_side[c] = p->cfg.io[ + p->cfg.two_side[c] = p->cfg.out[ p->cfg.two_side[c].hw]; if (p->cfg.psiz < 0x40) - p->cfg.psiz = p->cfg.io[p->cfg.psiz].hw; + p->cfg.psiz = p->cfg.out[p->cfg.psiz].hw; + + copy_semantic_info(p); } else if (p->type == PIPE_SHADER_FRAGMENT) { - int rid, aid; + int rid, aid, base; unsigned n = 0, m = pc->attr_nr - flat_nr; pc->allow32 = TRUE; - int base = (TGSI_SEMANTIC_POSITION == - p->info.input_semantic_name[0]) ? 0 : 1; + base = (TGSI_SEMANTIC_POSITION == + p->info.input_semantic_name[0]) ? 0 : 1; /* non-flat interpolants have to be mapped to * the lower hardware IDs, so sort them: */ for (i = 0; i < pc->attr_nr; i++) { if (pc->interp_mode[i] == INTERP_FLAT) - p->cfg.io[m++].id = i; + p->cfg.in[m++].id = i; else { if (!(pc->interp_mode[i] & INTERP_PERSPECTIVE)) - p->cfg.io[n].linear = TRUE; - p->cfg.io[n++].id = i; + p->cfg.in[n].linear = TRUE; + p->cfg.in[n++].id = i; } } + copy_semantic_info(p); if (!base) /* set w-coordinate mask from perspective interp */ - p->cfg.io[0].mask |= p->cfg.regs[1] >> 24; + p->cfg.in[0].mask |= p->cfg.regs[1] >> 24; aid = popcnt4( /* if fcrd isn't contained in cfg.io */ - base ? (p->cfg.regs[1] >> 24) : p->cfg.io[0].mask); + base ? (p->cfg.regs[1] >> 24) : p->cfg.in[0].mask); for (n = 0; n < pc->attr_nr; ++n) { - p->cfg.io[n].hw = rid = aid; - i = p->cfg.io[n].id; + p->cfg.in[n].hw = rid = aid; + i = p->cfg.in[n].id; if (p->info.input_semantic_name[n] == TGSI_SEMANTIC_FACE) { @@ -3369,15 +3744,15 @@ nv50_program_tx_prep(struct nv50_pc *pc) if (!pc->attr[i * 4 + c].acc) continue; pc->attr[i * 4 + c].rhw = rid++; - p->cfg.io[n].mask |= 1 << c; + p->cfg.in[n].mask |= 1 << c; load_interpolant(pc, &pc->attr[i * 4 + c]); } - aid += popcnt4(p->cfg.io[n].mask); + aid += popcnt4(p->cfg.in[n].mask); } if (!base) - p->cfg.regs[1] |= p->cfg.io[0].mask << 24; + p->cfg.regs[1] |= p->cfg.in[0].mask << 24; m = popcnt4(p->cfg.regs[1] >> 24); @@ -3387,32 +3762,33 @@ nv50_program_tx_prep(struct nv50_pc *pc) p->cfg.regs[1] |= aid - m; if (flat_nr) { - i = p->cfg.io[pc->attr_nr - flat_nr].hw; + i = p->cfg.in[pc->attr_nr - flat_nr].hw; p->cfg.regs[1] |= (i - m) << 16; } else p->cfg.regs[1] |= p->cfg.regs[1] << 16; /* mark color semantic for light-twoside */ - n = 0x40; - for (i = 0; i < pc->attr_nr; i++) { - ubyte si, sn; - - sn = p->info.input_semantic_name[p->cfg.io[i].id]; - si = p->info.input_semantic_index[p->cfg.io[i].id]; - - if (sn == TGSI_SEMANTIC_COLOR) { - p->cfg.two_side[si] = p->cfg.io[i]; - - /* increase colour count */ - p->cfg.regs[0] += popcnt4( - p->cfg.two_side[si].mask) << 16; - - n = MIN2(n, p->cfg.io[i].hw - m); + n = 0x80; + for (i = 0; i < p->cfg.in_nr; i++) { + if (p->cfg.in[i].sn == TGSI_SEMANTIC_COLOR) { + n = MIN2(n, p->cfg.in[i].hw - m); + p->cfg.two_side[p->cfg.in[i].si] = p->cfg.in[i]; + + p->cfg.regs[0] += /* increase colour count */ + popcnt4(p->cfg.in[i].mask) << 16; } } - if (n < 0x40) + if (n < 0x80) p->cfg.regs[0] += n; + if (p->cfg.prim_id < 0x40) { + pc->sysval[p->cfg.prim_id].rhw = rid++; + emit_interp(pc, &pc->sysval[p->cfg.prim_id], NULL, + INTERP_FLAT); + /* increase FP_INTERPOLANT_CTRL_COUNT */ + p->cfg.regs[1] += 1; + } + /* Initialize FP results: * FragDepth is always first TGSI and last hw output */ @@ -3466,10 +3842,31 @@ free_nv50_pc(struct nv50_pc *pc) FREE(pc->attr); if (pc->temp) FREE(pc->temp); + if (pc->sysval) + FREE(pc->sysval); + if (pc->insn_pos) + FREE(pc->insn_pos); FREE(pc); } +static INLINE uint32_t +nv50_map_gs_output_prim(unsigned pprim) +{ + switch (pprim) { + case PIPE_PRIM_POINTS: + return NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE_POINTS; + case PIPE_PRIM_LINE_STRIP: + return NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE_LINE_STRIP; + case PIPE_PRIM_TRIANGLE_STRIP: + return NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE_TRIANGLE_STRIP; + default: + NOUVEAU_ERR("invalid GS_OUTPUT_PRIMITIVE: %u\n", pprim); + abort(); + return 0; + } +} + static boolean ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p) { @@ -3483,25 +3880,55 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p) pc->param_nr = p->info.file_max[TGSI_FILE_CONSTANT] + 1; pc->addr_nr = p->info.file_max[TGSI_FILE_ADDRESS] + 1; assert(pc->addr_nr <= 2); + pc->sysval_nr = p->info.file_max[TGSI_FILE_SYSTEM_VALUE] + 1; p->cfg.high_temp = 4; p->cfg.two_side[0].hw = 0x40; p->cfg.two_side[1].hw = 0x40; + p->cfg.prim_id = 0x40; p->cfg.edgeflag_in = pc->edgeflag_out = 0xff; + for (i = 0; i < p->info.num_properties; ++i) { + unsigned *data = &p->info.properties[i].data[0]; + + switch (p->info.properties[i].name) { + case TGSI_PROPERTY_GS_OUTPUT_PRIM: + p->cfg.prim_type = nv50_map_gs_output_prim(data[0]); + break; + case TGSI_PROPERTY_GS_MAX_VERTICES: + p->cfg.vert_count = data[0]; + break; + default: + break; + } + } + switch (p->type) { case PIPE_SHADER_VERTEX: p->cfg.psiz = 0x40; p->cfg.clpd = 0x40; - p->cfg.io_nr = pc->result_nr; + p->cfg.out_nr = pc->result_nr; + break; + case PIPE_SHADER_GEOMETRY: + assert(p->cfg.prim_type); + assert(p->cfg.vert_count); + + p->cfg.psiz = 0x80; + p->cfg.clpd = 0x80; + p->cfg.prim_id = 0x80; + p->cfg.out_nr = pc->result_nr; + p->cfg.in_nr = pc->attr_nr; + + p->cfg.two_side[0].hw = 0x80; + p->cfg.two_side[1].hw = 0x80; break; case PIPE_SHADER_FRAGMENT: rtype[0] = rtype[1] = P_TEMP; p->cfg.regs[0] = 0x01000004; - p->cfg.io_nr = pc->attr_nr; + p->cfg.in_nr = pc->attr_nr; if (p->info.writes_z) { p->cfg.regs[2] |= 0x00000100; @@ -3559,7 +3986,16 @@ ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p) return FALSE; } for (i = 0; i < NV50_SU_MAX_ADDR; ++i) - ctor_reg(&pc->r_addr[i], P_ADDR, -256, i + 1); + ctor_reg(&pc->r_addr[i], P_ADDR, -1, i + 1); + + if (pc->sysval_nr) { + pc->sysval = CALLOC(pc->sysval_nr, sizeof(struct nv50_reg *)); + if (!pc->sysval) + return FALSE; + /* will only ever use SYSTEM_VALUE[i].x (hopefully) */ + for (i = 0; i < pc->sysval_nr; ++i) + ctor_reg(&pc->sysval[i], rtype[0], i, -1); + } return TRUE; } @@ -3721,13 +4157,17 @@ nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p) if (p->param_nr) { unsigned cb; - uint32_t *map = pipe_buffer_map(pscreen, nv50->constbuf[p->type], + uint32_t *map = pipe_buffer_map(pscreen, + nv50->constbuf[p->type], PIPE_BUFFER_USAGE_CPU_READ); - - if (p->type == PIPE_SHADER_VERTEX) + switch (p->type) { + case PIPE_SHADER_GEOMETRY: cb = NV50_CB_PGP; break; + case PIPE_SHADER_FRAGMENT: cb = NV50_CB_PFP; break; + default: cb = NV50_CB_PVP; - else - cb = NV50_CB_PFP; + assert(p->type == PIPE_SHADER_VERTEX); + break; + } nv50_program_upload_data(nv50, map, 0, p->param_nr, cb); pipe_buffer_unmap(pscreen, nv50->constbuf[p->type]); @@ -3821,19 +4261,18 @@ nv50_vertprog_validate(struct nv50_context *nv50) nv50_program_validate_data(nv50, p); nv50_program_validate_code(nv50, p); - so = so_new(5, 8, 2); + so = so_new(5, 7, 2); so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2); so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_HIGH, 0, 0); + NOUVEAU_BO_HIGH, 0, 0); so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | - NOUVEAU_BO_LOW, 0, 0); + NOUVEAU_BO_LOW, 0, 0); so_method(so, tesla, NV50TCL_VP_ATTR_EN_0, 2); so_data (so, p->cfg.attr[0]); so_data (so, p->cfg.attr[1]); so_method(so, tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1); so_data (so, p->cfg.high_result); - so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 2); - so_data (so, p->cfg.high_result); //8); + so_method(so, tesla, NV50TCL_VP_REG_ALLOC_TEMP, 1); so_data (so, p->cfg.high_temp); so_method(so, tesla, NV50TCL_VP_START_ID, 1); so_data (so, 0); /* program start offset */ @@ -3877,12 +4316,51 @@ nv50_fragprog_validate(struct nv50_context *nv50) so_ref(NULL, &so); } -static void +void +nv50_geomprog_validate(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nv50_program *p = nv50->geomprog; + struct nouveau_stateobj *so; + + if (!p->translated) { + nv50_program_validate(nv50, p); + if (!p->translated) + assert(0); + } + + nv50_program_validate_data(nv50, p); + nv50_program_validate_code(nv50, p); + + so = so_new(6, 7, 2); + so_method(so, tesla, NV50TCL_GP_ADDRESS_HIGH, 2); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | + NOUVEAU_BO_LOW, 0, 0); + so_method(so, tesla, NV50TCL_GP_REG_ALLOC_TEMP, 1); + so_data (so, p->cfg.high_temp); + so_method(so, tesla, NV50TCL_GP_REG_ALLOC_RESULT, 1); + so_data (so, p->cfg.high_result); + so_method(so, tesla, NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE, 1); + so_data (so, p->cfg.prim_type); + so_method(so, tesla, NV50TCL_GP_VERTEX_OUTPUT_COUNT, 1); + so_data (so, p->cfg.vert_count); + so_method(so, tesla, NV50TCL_GP_START_ID, 1); + so_data (so, 0); + so_ref(so, &nv50->state.geomprog); + so_ref(NULL, &so); +} + +static uint32_t nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base) { + struct nv50_program *vp; struct nv50_program *fp = nv50->fragprog; - struct nv50_program *vp = nv50->vertprog; unsigned i, c, m = base; + uint32_t origin = 0x00000010; + + vp = nv50->geomprog ? nv50->geomprog : nv50->vertprog; /* XXX: this might not work correctly in all cases yet - we'll * just assume that an FP generic input that is not written in @@ -3890,58 +4368,59 @@ nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base) */ memset(pntc, 0, 8 * sizeof(uint32_t)); - for (i = 0; i < fp->cfg.io_nr; i++) { - uint8_t sn, si; - uint8_t j, k = fp->cfg.io[i].id; - unsigned n = popcnt4(fp->cfg.io[i].mask); + for (i = 0; i < fp->cfg.in_nr; i++) { + unsigned j, n = popcnt4(fp->cfg.in[i].mask); - if (fp->info.input_semantic_name[k] != TGSI_SEMANTIC_GENERIC) { + if (fp->cfg.in[i].sn != TGSI_SEMANTIC_GENERIC) { m += n; continue; } - for (j = 0; j < vp->info.num_outputs; ++j) { - sn = vp->info.output_semantic_name[j]; - si = vp->info.output_semantic_index[j]; - - if (sn == fp->info.input_semantic_name[k] && - si == fp->info.input_semantic_index[k]) + for (j = 0; j < vp->cfg.out_nr; ++j) + if (vp->cfg.out[j].sn == fp->cfg.in[i].sn && + vp->cfg.out[j].si == fp->cfg.in[i].si) break; - } - if (j < vp->info.num_outputs) { - ubyte mode = - nv50->rasterizer->pipe.sprite_coord_mode[si]; + if (j < vp->cfg.out_nr) { + ubyte mode = nv50->rasterizer->pipe.sprite_coord_mode[ + vp->cfg.out[j].si]; if (mode == PIPE_SPRITE_COORD_NONE) { m += n; continue; - } + } else + if (mode == PIPE_SPRITE_COORD_LOWER_LEFT) + origin = 0; } /* this is either PointCoord or replaced by sprite coords */ for (c = 0; c < 4; c++) { - if (!(fp->cfg.io[i].mask & (1 << c))) + if (!(fp->cfg.in[i].mask & (1 << c))) continue; pntc[m / 8] |= (c + 1) << ((m % 8) * 4); ++m; } } + return origin; } static int -nv50_sreg4_map(uint32_t *p_map, int mid, uint32_t lin[4], - struct nv50_sreg4 *fpi, struct nv50_sreg4 *vpo) +nv50_vec4_map(uint32_t *map32, int mid, uint8_t zval, uint32_t lin[4], + struct nv50_sreg4 *fpi, struct nv50_sreg4 *vpo) { int c; uint8_t mv = vpo->mask, mf = fpi->mask, oid = vpo->hw; - uint8_t *map = (uint8_t *)p_map; + uint8_t *map = (uint8_t *)map32; for (c = 0; c < 4; ++c) { if (mf & 1) { if (fpi->linear == TRUE) lin[mid / 32] |= 1 << (mid % 32); - map[mid++] = (mv & 1) ? oid : ((c == 3) ? 0x41 : 0x40); + if (mv & 1) + map[mid] = oid; + else + map[mid] = (c == 3) ? (zval + 1) : zval; + ++mid; } oid += mv & 1; @@ -3953,34 +4432,42 @@ nv50_sreg4_map(uint32_t *p_map, int mid, uint32_t lin[4], } void -nv50_linkage_validate(struct nv50_context *nv50) +nv50_fp_linkage_validate(struct nv50_context *nv50) { struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_program *vp = nv50->vertprog; struct nv50_program *fp = nv50->fragprog; struct nouveau_stateobj *so; - struct nv50_sreg4 dummy, *vpo; + struct nv50_sreg4 dummy; int i, n, c, m = 0; - uint32_t map[16], lin[4], reg[5], pcrd[8]; + uint32_t map[16], lin[4], reg[6], pcrd[8]; + uint8_t zval = 0x40; + if (nv50->geomprog) { + vp = nv50->geomprog; + zval = 0x80; + } memset(map, 0, sizeof(map)); memset(lin, 0, sizeof(lin)); reg[1] = 0x00000004; /* low and high clip distance map ids */ reg[2] = 0x00000000; /* layer index map id (disabled, GP only) */ reg[3] = 0x00000000; /* point size map id & enable */ + reg[5] = 0x00000000; /* primitive ID map slot */ reg[0] = fp->cfg.regs[0]; /* colour semantic reg */ reg[4] = fp->cfg.regs[1]; /* interpolant info */ dummy.linear = FALSE; dummy.mask = 0xf; /* map all components of HPOS */ - m = nv50_sreg4_map(map, m, lin, &dummy, &vp->cfg.io[0]); + m = nv50_vec4_map(map, m, zval, lin, &dummy, &vp->cfg.out[0]); dummy.mask = 0x0; if (vp->cfg.clpd < 0x40) { - for (c = 0; c < vp->cfg.clpd_nr; ++c) - map[m++] = vp->cfg.clpd + c; + for (c = 0; c < vp->cfg.clpd_nr; ++c) { + map[m / 4] |= (vp->cfg.clpd + c) << ((m % 4) * 8); + ++m; + } reg[1] = (m << 8); } @@ -3988,35 +4475,37 @@ nv50_linkage_validate(struct nv50_context *nv50) /* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */ if (nv50->rasterizer->pipe.light_twoside) { - vpo = &vp->cfg.two_side[0]; + struct nv50_sreg4 *vpo = &vp->cfg.two_side[0]; + struct nv50_sreg4 *fpi = &fp->cfg.two_side[0]; - m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[0], &vpo[0]); - m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[1], &vpo[1]); + m = nv50_vec4_map(map, m, zval, lin, &fpi[0], &vpo[0]); + m = nv50_vec4_map(map, m, zval, lin, &fpi[1], &vpo[1]); } reg[0] += m - 4; /* adjust FFC0 id */ reg[4] |= m << 8; /* set mid where 'normal' FP inputs start */ - for (i = 0; i < fp->cfg.io_nr; i++) { - ubyte sn = fp->info.input_semantic_name[fp->cfg.io[i].id]; - ubyte si = fp->info.input_semantic_index[fp->cfg.io[i].id]; - - /* position must be mapped first */ - assert(i == 0 || sn != TGSI_SEMANTIC_POSITION); - + for (i = 0; i < fp->cfg.in_nr; i++) { /* maybe even remove these from cfg.io */ - if (sn == TGSI_SEMANTIC_POSITION || sn == TGSI_SEMANTIC_FACE) + if (fp->cfg.in[i].sn == TGSI_SEMANTIC_POSITION || + fp->cfg.in[i].sn == TGSI_SEMANTIC_FACE) continue; - /* VP outputs and vp->cfg.io are in the same order */ - for (n = 0; n < vp->info.num_outputs; ++n) { - if (vp->info.output_semantic_name[n] == sn && - vp->info.output_semantic_index[n] == si) + for (n = 0; n < vp->cfg.out_nr; ++n) + if (vp->cfg.out[n].sn == fp->cfg.in[i].sn && + vp->cfg.out[n].si == fp->cfg.in[i].si) break; - } - vpo = (n < vp->info.num_outputs) ? &vp->cfg.io[n] : &dummy; - m = nv50_sreg4_map(map, m, lin, &fp->cfg.io[i], vpo); + m = nv50_vec4_map(map, m, zval, lin, &fp->cfg.in[i], + (n < vp->cfg.out_nr) ? + &vp->cfg.out[n] : &dummy); + } + /* PrimitiveID either is replaced by the system value, or + * written by the geometry shader into an output register + */ + if (fp->cfg.prim_id < 0x40) { + map[m / 4] |= vp->cfg.prim_id << ((m % 4) * 8); + reg[5] = m++; } if (nv50->rasterizer->pipe.point_size_per_vertex) { @@ -4024,14 +4513,28 @@ nv50_linkage_validate(struct nv50_context *nv50) reg[3] = (m++ << 4) | 1; } - /* now fill the stateobj */ - so = so_new(6, 58, 0); + /* now fill the stateobj (at most 28 so_data) */ + so = so_new(10, 54, 0); n = (m + 3) / 4; - so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1); - so_data (so, m); - so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n); - so_datap (so, map, n); + assert(m <= 32); + if (vp->type == PIPE_SHADER_GEOMETRY) { + so_method(so, tesla, NV50TCL_GP_RESULT_MAP_SIZE, 1); + so_data (so, m); + so_method(so, tesla, NV50TCL_GP_RESULT_MAP(0), n); + so_datap (so, map, n); + } else { + so_method(so, tesla, NV50TCL_VP_GP_BUILTIN_ATTR_EN, 1); + so_data (so, vp->cfg.regs[0]); + + so_method(so, tesla, NV50TCL_MAP_SEMANTIC_4, 1); + so_data (so, reg[5]); + + so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1); + so_data (so, m); + so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n); + so_datap (so, map, n); + } so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4); so_datap (so, reg, 4); @@ -4043,14 +4546,85 @@ nv50_linkage_validate(struct nv50_context *nv50) so_datap (so, lin, 4); if (nv50->rasterizer->pipe.point_sprite) { - nv50_pntc_replace(nv50, pcrd, (reg[4] >> 8) & 0xff); + so_method(so, tesla, NV50TCL_POINT_SPRITE_CTRL, 1); + so_data (so, + nv50_pntc_replace(nv50, pcrd, (reg[4] >> 8) & 0xff)); so_method(so, tesla, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8); so_datap (so, pcrd, 8); } - so_ref(so, &nv50->state.programs); - so_ref(NULL, &so); + so_method(so, tesla, NV50TCL_GP_ENABLE, 1); + so_data (so, (vp->type == PIPE_SHADER_GEOMETRY) ? 1 : 0); + + so_ref(so, &nv50->state.fp_linkage); + so_ref(NULL, &so); +} + +static int +construct_vp_gp_mapping(uint32_t *map32, int m, + struct nv50_program *vp, struct nv50_program *gp) +{ + uint8_t *map = (uint8_t *)map32; + int i, j, c; + + for (i = 0; i < gp->cfg.in_nr; ++i) { + uint8_t oid, mv = 0, mg = gp->cfg.in[i].mask; + + for (j = 0; j < vp->cfg.out_nr; ++j) { + if (vp->cfg.out[j].sn == gp->cfg.in[i].sn && + vp->cfg.out[j].si == gp->cfg.in[i].si) { + mv = vp->cfg.out[j].mask; + oid = vp->cfg.out[j].hw; + break; + } + } + + for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) { + if (mg & mv & 1) + map[m++] = oid; + else + if (mg & 1) + map[m++] = (c == 3) ? 0x41 : 0x40; + oid += mv & 1; + } + } + return m; +} + +void +nv50_gp_linkage_validate(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nouveau_stateobj *so; + struct nv50_program *vp = nv50->vertprog; + struct nv50_program *gp = nv50->geomprog; + uint32_t map[16]; + int m = 0; + + if (!gp) { + so_ref(NULL, &nv50->state.gp_linkage); + return; + } + memset(map, 0, sizeof(map)); + + m = construct_vp_gp_mapping(map, m, vp, gp); + + so = so_new(3, 24 - 3, 0); + + so_method(so, tesla, NV50TCL_VP_GP_BUILTIN_ATTR_EN, 1); + so_data (so, vp->cfg.regs[0] | gp->cfg.regs[0]); + + assert(m <= 32); + so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1); + so_data (so, m); + + m = (m + 3) / 4; + so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), m); + so_datap (so, map, m); + + so_ref(so, &nv50->state.gp_linkage); + so_ref(NULL, &so); } void diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h index 461fec1d89c..1e3ad6bff05 100644 --- a/src/gallium/drivers/nv50/nv50_program.h +++ b/src/gallium/drivers/nv50/nv50_program.h @@ -16,11 +16,13 @@ struct nv50_program_exec { }; struct nv50_sreg4 { - uint8_t hw; - uint8_t id; /* tgsi index, nv50 needs them sorted: flat ones last */ + uint8_t hw; /* hw index, nv50 wants flat FP inputs last */ + uint8_t id; /* tgsi index */ uint8_t mask; boolean linear; + + ubyte sn, si; /* semantic name & index */ }; struct nv50_program { @@ -49,16 +51,24 @@ struct nv50_program { uint32_t regs[4]; /* for VPs, io_nr doesn't count 'private' results (PSIZ etc.) */ - unsigned io_nr; - struct nv50_sreg4 io[PIPE_MAX_SHADER_OUTPUTS]; + unsigned in_nr, out_nr; + struct nv50_sreg4 in[PIPE_MAX_SHADER_INPUTS]; + struct nv50_sreg4 out[PIPE_MAX_SHADER_OUTPUTS]; /* FP colour inputs, VP/GP back colour outputs */ struct nv50_sreg4 two_side[2]; - /* VP only */ + /* GP only */ + unsigned vert_count; + uint8_t prim_type; + + /* VP & GP only */ uint8_t clpd, clpd_nr; uint8_t psiz; uint8_t edgeflag_in; + + /* FP & GP only */ + uint8_t prim_id; } cfg; }; diff --git a/src/gallium/drivers/nv50/nv50_query.c b/src/gallium/drivers/nv50/nv50_query.c index 5d9e18218ae..5a4ab3508b8 100644 --- a/src/gallium/drivers/nv50/nv50_query.c +++ b/src/gallium/drivers/nv50/nv50_query.c @@ -111,7 +111,7 @@ nv50_query_result(struct pipe_context *pipe, struct pipe_query *pq, if (!q->ready) { ret = nouveau_bo_map(q->bo, NOUVEAU_BO_RD | - wait ? 0 : NOUVEAU_BO_NOWAIT); + (wait ? 0 : NOUVEAU_BO_NOWAIT)); if (ret) return false; q->result = ((uint32_t *)q->bo->map)[1]; diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c index 28e2b35deaa..73205652cc4 100644 --- a/src/gallium/drivers/nv50/nv50_screen.c +++ b/src/gallium/drivers/nv50/nv50_screen.c @@ -167,7 +167,7 @@ nv50_screen_destroy(struct pipe_screen *pscreen) struct nv50_screen *screen = nv50_screen(pscreen); unsigned i; - for (i = 0; i < 2; i++) { + for (i = 0; i < 3; i++) { if (screen->constbuf_parm[i]) nouveau_bo_ref(NULL, &screen->constbuf_parm[i]); } @@ -329,7 +329,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) so_ref(NULL, &so); /* Static tesla init */ - so = so_new(40, 84, 20); + so = so_new(44, 90, 22); so_method(so, screen->tesla, NV50TCL_COND_MODE, 1); so_data (so, NV50TCL_COND_MODE_ALWAYS); @@ -352,10 +352,11 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) so_data (so, 0xf); /* max TIC (bits 4:8) & TSC (ignored) bindings, per program type */ - so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(0), 1); - so_data (so, 0x54); - so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(2), 1); - so_data (so, 0x54); + for (i = 0; i < 3; ++i) { + so_method(so, screen->tesla, NV50TCL_TEX_LIMITS(i), 1); + so_data (so, 0x54); + } + /* origin is top left (set to 1 for bottom left) */ so_method(so, screen->tesla, NV50TCL_Y_ORIGIN_BOTTOM, 1); so_data (so, 0); @@ -370,7 +371,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) return NULL; } - for (i = 0; i < 2; i++) { + for (i = 0; i < 3; i++) { ret = nouveau_bo_new(dev, NOUVEAU_BO_VRAM, 0, (128 * 4) * 4, &screen->constbuf_parm[i]); if (ret) { @@ -406,22 +407,33 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev) so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); so_data (so, 0x00000001 | (NV50_CB_PMISC << 12)); so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); + so_data (so, 0x00000021 | (NV50_CB_PMISC << 12)); + so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); so_data (so, 0x00000031 | (NV50_CB_PMISC << 12)); so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_parm[0], 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_parm[PIPE_SHADER_VERTEX], 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_data (so, (NV50_CB_PVP << 16) | 0x00000800); so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); so_data (so, 0x00000101 | (NV50_CB_PVP << 12)); so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); - so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); - so_reloc (so, screen->constbuf_parm[1], 0, NOUVEAU_BO_VRAM | - NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_parm[PIPE_SHADER_GEOMETRY], 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); + so_data (so, (NV50_CB_PGP << 16) | 0x00000800); + so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); + so_data (so, 0x00000121 | (NV50_CB_PGP << 12)); + + so_method(so, screen->tesla, NV50TCL_CB_DEF_ADDRESS_HIGH, 3); + so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_HIGH, 0, 0); + so_reloc (so, screen->constbuf_parm[PIPE_SHADER_FRAGMENT], 0, + NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_data (so, (NV50_CB_PFP << 16) | 0x00000800); so_method(so, screen->tesla, NV50TCL_SET_PROGRAM_CB, 1); so_data (so, 0x00000131 | (NV50_CB_PFP << 12)); diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h index a038a4e3c2a..7a155ca9c45 100644 --- a/src/gallium/drivers/nv50/nv50_screen.h +++ b/src/gallium/drivers/nv50/nv50_screen.h @@ -18,10 +18,10 @@ struct nv50_screen { struct nouveau_notifier *sync; struct nouveau_bo *constbuf_misc[1]; - struct nouveau_bo *constbuf_parm[2]; + struct nouveau_bo *constbuf_parm[PIPE_SHADER_TYPES]; struct nouveau_resource *immd_heap[1]; - struct nouveau_resource *parm_heap[2]; + struct nouveau_resource *parm_heap[PIPE_SHADER_TYPES]; struct nouveau_bo *tic; struct nouveau_bo *tsc; diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index 1f67df814b1..6ab33be663d 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -531,7 +531,7 @@ nv50_vp_state_delete(struct pipe_context *pipe, void *hwcso) struct nv50_program *p = hwcso; nv50_program_destroy(nv50, p); - FREE((void*)p->pipe.tokens); + FREE((void *)p->pipe.tokens); FREE(p); } @@ -563,7 +563,39 @@ nv50_fp_state_delete(struct pipe_context *pipe, void *hwcso) struct nv50_program *p = hwcso; nv50_program_destroy(nv50, p); - FREE((void*)p->pipe.tokens); + FREE((void *)p->pipe.tokens); + FREE(p); +} + +static void * +nv50_gp_state_create(struct pipe_context *pipe, + const struct pipe_shader_state *cso) +{ + struct nv50_program *p = CALLOC_STRUCT(nv50_program); + + p->pipe.tokens = tgsi_dup_tokens(cso->tokens); + p->type = PIPE_SHADER_GEOMETRY; + tgsi_scan_shader(p->pipe.tokens, &p->info); + return (void *)p; +} + +static void +nv50_gp_state_bind(struct pipe_context *pipe, void *hwcso) +{ + struct nv50_context *nv50 = nv50_context(pipe); + + nv50->fragprog = hwcso; + nv50->dirty |= NV50_NEW_GEOMPROG; +} + +static void +nv50_gp_state_delete(struct pipe_context *pipe, void *hwcso) +{ + struct nv50_context *nv50 = nv50_context(pipe); + struct nv50_program *p = hwcso; + + nv50_program_destroy(nv50, p); + FREE((void *)p->pipe.tokens); FREE(p); } @@ -585,17 +617,21 @@ nv50_set_clip_state(struct pipe_context *pipe, static void nv50_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf ) + struct pipe_buffer *buf ) { struct nv50_context *nv50 = nv50_context(pipe); if (shader == PIPE_SHADER_VERTEX) { - nv50->constbuf[PIPE_SHADER_VERTEX] = buf->buffer; + nv50->constbuf[PIPE_SHADER_VERTEX] = buf; nv50->dirty |= NV50_NEW_VERTPROG_CB; } else if (shader == PIPE_SHADER_FRAGMENT) { - nv50->constbuf[PIPE_SHADER_FRAGMENT] = buf->buffer; + nv50->constbuf[PIPE_SHADER_FRAGMENT] = buf; nv50->dirty |= NV50_NEW_FRAGPROG_CB; + } else + if (shader == PIPE_SHADER_GEOMETRY) { + nv50->constbuf[PIPE_SHADER_GEOMETRY] = buf; + nv50->dirty |= NV50_NEW_GEOMPROG_CB; } } @@ -696,6 +732,10 @@ nv50_init_state_functions(struct nv50_context *nv50) nv50->pipe.bind_fs_state = nv50_fp_state_bind; nv50->pipe.delete_fs_state = nv50_fp_state_delete; + nv50->pipe.create_gs_state = nv50_gp_state_create; + nv50->pipe.bind_gs_state = nv50_gp_state_bind; + nv50->pipe.delete_gs_state = nv50_gp_state_delete; + nv50->pipe.set_blend_color = nv50_set_blend_color; nv50->pipe.set_clip_state = nv50_set_clip_state; nv50->pipe.set_constant_buffer = nv50_set_constant_buffer; diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index f83232f43cf..fcd07b59cd6 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -199,6 +199,8 @@ nv50_state_emit(struct nv50_context *nv50) nv50->state.dirty |= NV50_NEW_VERTPROG; if (nv50->state.fragprog) nv50->state.dirty |= NV50_NEW_FRAGPROG; + if (nv50->state.geomprog) + nv50->state.dirty |= NV50_NEW_GEOMPROG; if (nv50->state.rast) nv50->state.dirty |= NV50_NEW_RASTERIZER; if (nv50->state.blend_colour) @@ -228,9 +230,14 @@ nv50_state_emit(struct nv50_context *nv50) so_emit(chan, nv50->state.vertprog); if (nv50->state.dirty & NV50_NEW_FRAGPROG) so_emit(chan, nv50->state.fragprog); + if (nv50->state.dirty & NV50_NEW_GEOMPROG && nv50->state.geomprog) + so_emit(chan, nv50->state.geomprog); if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG | - NV50_NEW_RASTERIZER)) - so_emit(chan, nv50->state.programs); + NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER)) + so_emit(chan, nv50->state.fp_linkage); + if ((nv50->state.dirty & (NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG)) + && nv50->state.gp_linkage) + so_emit(chan, nv50->state.gp_linkage); if (nv50->state.dirty & NV50_NEW_RASTERIZER) so_emit(chan, nv50->state.rast); if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR) @@ -291,9 +298,15 @@ nv50_state_validate(struct nv50_context *nv50) if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB)) nv50_fragprog_validate(nv50); + if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB)) + nv50_geomprog_validate(nv50); + if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG | - NV50_NEW_RASTERIZER)) - nv50_linkage_validate(nv50); + NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER)) + nv50_fp_linkage_validate(nv50); + + if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_VERTPROG)) + nv50_gp_linkage_validate(nv50); if (nv50->dirty & NV50_NEW_RASTERIZER) so_ref(nv50->rasterizer->so, &nv50->state.rast); @@ -400,8 +413,9 @@ viewport_uptodate: for (i = 0; i < PIPE_SHADER_TYPES; ++i) nr += nv50->sampler_nr[i]; - so = so_new(1+ 5 * PIPE_SHADER_TYPES, 1+ 19 * PIPE_SHADER_TYPES - + nr * 8, PIPE_SHADER_TYPES * 2); + so = so_new(1 + 5 * PIPE_SHADER_TYPES, + 1 + 19 * PIPE_SHADER_TYPES + nr * 8, + PIPE_SHADER_TYPES * 2); nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX); nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT); diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c index bef548b7286..871536dca9b 100644 --- a/src/gallium/drivers/nv50/nv50_tex.c +++ b/src/gallium/drivers/nv50/nv50_tex.c @@ -155,7 +155,7 @@ static boolean nv50_validate_textures(struct nv50_context *nv50, struct nouveau_stateobj *so, unsigned p) { - static const unsigned p_remap[PIPE_SHADER_TYPES] = { 0, 2 }; + static const unsigned p_remap[PIPE_SHADER_TYPES] = { 0, 2, 1 }; struct nouveau_grobj *eng2d = nv50->screen->eng2d; struct nouveau_grobj *tesla = nv50->screen->tesla; diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c index f2e510fba61..89a94d2fe81 100644 --- a/src/gallium/drivers/nv50/nv50_vbo.c +++ b/src/gallium/drivers/nv50/nv50_vbo.c @@ -55,6 +55,14 @@ nv50_prim(unsigned mode) case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS; case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP; case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON; + case PIPE_PRIM_LINES_ADJACENCY: + return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY; + case PIPE_PRIM_LINE_STRIP_ADJACENCY: + return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY; + case PIPE_PRIM_TRIANGLES_ADJACENCY: + return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY; + case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY: + return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY; default: break; } diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c index ffe066d5369..c14414fff6b 100644 --- a/src/gallium/drivers/r300/r300_blit.c +++ b/src/gallium/drivers/r300/r300_blit.c @@ -27,9 +27,9 @@ static void r300_blitter_save_states(struct r300_context* r300) { - util_blitter_save_blend(r300->blitter, r300->blend_state); - util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state); - util_blitter_save_rasterizer(r300->blitter, r300->rs_state); + 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); } diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c index 51fdb82ff34..92de297ef1d 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 0633a8b8a72..28084864929 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_context.c b/src/gallium/drivers/r300/r300_context.c index d5c2d63d393..5e4f6552c36 100644 --- a/src/gallium/drivers/r300/r300_context.c +++ b/src/gallium/drivers/r300/r300_context.c @@ -30,6 +30,7 @@ #include "r300_blit.h" #include "r300_context.h" +#include "r300_emit.h" #include "r300_flush.h" #include "r300_query.h" #include "r300_render.h" @@ -69,11 +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->scissor_state.state); FREE(r300->vertex_info); - FREE(r300->viewport_state); + FREE(r300->viewport_state.state); + FREE(r300->ztop_state.state); FREE(r300); } @@ -107,6 +110,35 @@ 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 radeon_winsys* radeon_winsys) { @@ -155,11 +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->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state); r300->vertex_info = CALLOC_STRUCT(r300_vertex_info); - r300->viewport_state = CALLOC_STRUCT(r300_viewport_state); + 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, diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h index 232530b7dc4..682b9179c83 100644 --- a/src/gallium/drivers/r300/r300_context.h +++ b/src/gallium/drivers/r300/r300_context.h @@ -30,9 +30,28 @@ #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 */ @@ -62,11 +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 */ @@ -102,19 +116,6 @@ struct r300_sampler_state { unsigned min_lod, max_lod; }; -struct r300_scissor_regs { - uint32_t top_left; /* R300_SC_SCISSORS_TL: 0x43e0 */ - uint32_t bottom_right; /* R300_SC_SCISSORS_BR: 0x43e4 */ - - /* Whether everything is culled by scissoring. */ - boolean empty_area; -}; - -struct r300_scissor_state { - struct r300_scissor_regs framebuffer; - struct r300_scissor_regs scissor; -}; - struct r300_texture_state { uint32_t format0; /* R300_TX_FORMAT0: 0x4480 */ uint32_t format1; /* R300_TX_FORMAT1: 0x44c0 */ @@ -135,24 +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 @@ -194,6 +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; @@ -230,6 +230,9 @@ 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 { @@ -273,38 +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]; @@ -317,6 +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 d142fee0502..151f72b0fe4 100644 --- a/src/gallium/drivers/r300/r300_cs.h +++ b/src/gallium/drivers/r300/r300_cs.h @@ -52,7 +52,7 @@ #define CS_LOCALS(context) \ struct r300_context* const cs_context_copy = (context); \ struct radeon_winsys* cs_winsys = cs_context_copy->winsys; \ - int cs_count = 0; + int cs_count = 0; (void) cs_count; #define CHECK_CS(size) \ assert(cs_winsys->check_cs(cs_winsys, (size))) diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index f8bfa714fef..9f93327e598 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -25,6 +25,7 @@ #include "util/u_format.h" #include "util/u_math.h" +#include "util/u_simple_list.h" #include "r300_context.h" #include "r300_cs.h" @@ -36,11 +37,13 @@ #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(R300_RB3D_ROPCNTL, blend->rop); OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3); if (r300->framebuffer_state.nr_cbufs) { OUT_CS(blend->blend_control); @@ -52,14 +55,13 @@ void r300_emit_blend_state(struct r300_context* r300, OUT_CS(0); /* XXX also disable fastfill here once it's supported */ } - OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop); 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); @@ -76,9 +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); @@ -106,13 +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 */ @@ -131,7 +133,6 @@ void r300_emit_dsa_state(struct r300_context* r300, } 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) { @@ -145,6 +146,8 @@ static const float * get_shader_constant( struct rc_constant * constant, struct r300_constant_buffer * externals) { + 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; @@ -167,30 +170,28 @@ static const float * get_shader_constant( /* Texture compare-fail value. */ /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient, - * this is always (0,0,0,0). */ + * 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->rs_state->enable_vte) { - vec[0] = r300->viewport_state->xscale; - vec[1] = r300->viewport_state->yscale; - vec[2] = r300->viewport_state->zscale; - } else { + 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->rs_state->enable_vte) { - vec[0] = r300->viewport_state->xoffset; - vec[1] = r300->viewport_state->yoffset; - vec[2] = r300->viewport_state->zoffset; - } else { - /* Zeros. */ + if (!r300->tcl_bypass) { + vec[0] = viewport->xoffset; + vec[1] = viewport->yoffset; + vec[2] = viewport->zoffset; } break; @@ -419,8 +420,10 @@ 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)); @@ -443,8 +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; @@ -576,8 +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); @@ -640,26 +646,47 @@ void r300_emit_rs_block_state(struct r300_context* r300, END_CS; } -static void r300_emit_scissor_regs(struct r300_context* r300, - struct r300_scissor_regs* 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); - BEGIN_CS(3); - OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2); - OUT_CS(scissor->top_left); - OUT_CS(scissor->bottom_right); - END_CS; -} + minx = miny = 0; + maxx = r300->framebuffer_state.width; + maxy = r300->framebuffer_state.height; -void r300_emit_scissor_state(struct r300_context* r300, - struct r300_scissor_state* scissor) -{ - if (r300->rs_state->rs.scissor) { - r300_emit_scissor_regs(r300, &scissor->scissor); + 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 { - r300_emit_scissor_regs(r300, &scissor->framebuffer); + /* 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(top_left); + OUT_CS(bottom_right); + END_CS; } void r300_emit_texture(struct r300_context* r300, @@ -695,8 +722,10 @@ void r300_emit_texture(struct r300_context* r300, 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; } @@ -762,32 +791,6 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset) 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) { int i; @@ -912,26 +915,27 @@ 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; } - END_CS; } void r300_emit_texture_count(struct r300_context* r300) @@ -955,6 +959,16 @@ void r300_emit_texture_count(struct r300_context* r300) } +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; +} + void r300_flush_textures(struct r300_context* r300) { CS_LOCALS(r300); @@ -978,18 +992,24 @@ 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; + } } - /* Check size of CS. */ - /* Make sure we have at least 8*1024 spare dwords. */ + /* 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, 8*1024)) { + if (!r300->winsys->check_cs(r300->winsys, dwords)) { r300->context.flush(&r300->context, 0, NULL); } @@ -1029,10 +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) { @@ -1060,24 +1082,11 @@ 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) { @@ -1106,21 +1115,11 @@ 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)) { @@ -1142,11 +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 3797d3d332a..05a6bfeae86 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, @@ -89,11 +84,12 @@ 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); /* Emit all dirty state. */ diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c index 14a08241fc4..59819cb1061 100644 --- a/src/gallium/drivers/r300/r300_flush.c +++ b/src/gallium/drivers/r300/r300_flush.c @@ -37,8 +37,10 @@ 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; /* We probably need to flush Draw, but we may have been called from * within Draw. This feels kludgy, but it might be the best thing. * @@ -54,7 +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_reg.h b/src/gallium/drivers/r300/r300_reg.h index 034bfc15cf9..361813891fb 100644 --- a/src/gallium/drivers/r300/r300_reg.h +++ b/src/gallium/drivers/r300/r300_reg.h @@ -1619,18 +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 @@ -2283,9 +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) @@ -2544,9 +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) diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c index a4ac9ad9a7b..710d850163f 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" @@ -69,16 +71,11 @@ uint32_t r300_translate_primitive(unsigned prim) } } -static boolean r300_nothing_to_draw(struct r300_context *r300) -{ - return r300->rs_state->rs.scissor && - r300->scissor_state->scissor.empty_area; -} - static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, unsigned mode) { - uint32_t color_control = r300->rs_state->color_control; + 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. @@ -98,7 +95,7 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, * ~ C. */ - if (r300->rs_state->rs.flatshade_first) { + if (rs->rs.flatshade_first) { switch (mode) { case PIPE_PRIM_TRIANGLE_FAN: color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND; @@ -119,6 +116,44 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300, 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) @@ -212,17 +247,54 @@ 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. */ 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 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; @@ -235,23 +307,24 @@ void r300_draw_range_elements(struct pipe_context* pipe, return; } - if (r300_nothing_to_draw(r300)) { - return; - } - r300_update_derived_state(r300); if (!r300_setup_vertex_buffers(r300)) { return; } + if (indexSize == 1) { + indexBuffer = r300_translate_elts(r300, indexBuffer, + &indexSize, &mode, &count); + } + if (!r300->winsys->add_buffer(r300->winsys, indexBuffer, RADEON_GEM_DOMAIN_GTT, 0)) { - return; + goto cleanup; } if (!r300->winsys->validate(r300->winsys)) { - return; + goto cleanup; } r300_emit_dirty_state(r300); @@ -260,6 +333,11 @@ void r300_draw_range_elements(struct pipe_context* pipe, r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex, mode, start, count); + +cleanup: + if (indexBuffer != orgIndexBuffer) { + pipe->screen->buffer_destroy(indexBuffer); + } } /* Simple helpers for context setup. Should probably be moved to util. */ @@ -273,7 +351,7 @@ void r300_draw_elements(struct pipe_context* pipe, } void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, - unsigned start, unsigned count) + unsigned start, unsigned count) { struct r300_context* r300 = r300_context(pipe); @@ -288,10 +366,6 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, return; } - if (r300_nothing_to_draw(r300)) { - return; - } - r300_update_derived_state(r300); if (!r300_setup_vertex_buffers(r300)) { @@ -300,9 +374,12 @@ void r300_draw_arrays(struct pipe_context* pipe, unsigned mode, r300_emit_dirty_state(r300); - r300_emit_aos(r300, start); - - r300_emit_draw_arrays(r300, mode, count); + 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); + } } /**************************************************************************** @@ -323,10 +400,6 @@ void r300_swtcl_draw_arrays(struct pipe_context* pipe, return; } - if (r300_nothing_to_draw(r300)) { - return; - } - for (i = 0; i < r300->vertex_buffer_count; i++) { void* buf = pipe_buffer_map(pipe->screen, r300->vertex_buffer[i].buffer, @@ -368,10 +441,6 @@ void r300_swtcl_draw_range_elements(struct pipe_context* pipe, return; } - if (r300_nothing_to_draw(r300)) { - return; - } - for (i = 0; i < r300->vertex_buffer_count; i++) { void* buf = pipe_buffer_map(pipe->screen, r300->vertex_buffer[i].buffer, diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c index a145a7f18a5..60ad763cf48 100644 --- a/src/gallium/drivers/r300/r300_state.c +++ b/src/gallium/drivers/r300/r300_state.c @@ -317,8 +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. */ @@ -340,20 +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, &uc); - r300->blend_color_state->blend_color = uc.ui; + 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, @@ -362,12 +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. @@ -460,9 +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. */ @@ -472,37 +481,11 @@ static void r300_delete_dsa_state(struct pipe_context* pipe, FREE(state); } -static void r300_set_scissor_regs(const struct pipe_scissor_state* state, - struct r300_scissor_regs *scissor, - boolean is_r500) -{ - if (is_r500) { - scissor->top_left = - (state->minx << R300_SCISSORS_X_SHIFT) | - (state->miny << R300_SCISSORS_Y_SHIFT); - 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. */ - scissor->top_left = - ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) | - ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT); - scissor->bottom_right = - (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) | - (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT); - } - - scissor->empty_area = state->minx >= state->maxx || - state->miny >= state->maxy; -} - static void r300_set_framebuffer_state(struct pipe_context* pipe, const struct pipe_framebuffer_state* state) { struct r300_context* r300 = r300_context(pipe); - struct pipe_scissor_state scissor; if (r300->draw) { draw_flush(r300->draw); @@ -510,19 +493,12 @@ static void r300->framebuffer_state = *state; - scissor.minx = scissor.miny = 0; - scissor.maxx = state->width; - scissor.maxy = state->height; - r300_set_scissor_regs(&scissor, &r300->scissor_state->framebuffer, - r300_screen(r300->context.screen)->caps->is_r500); - /* Don't rely on the order of states being set for the first time. */ - if (!r300->rs_state || !r300->rs_state->rs.scissor) { - r300->dirty_state |= R300_NEW_SCISSOR; - } r300->dirty_state |= R300_NEW_FRAMEBUFFERS; - r300->dirty_state |= R300_NEW_BLEND; - r300->dirty_state |= R300_NEW_DSA; + + r300->blend_state.dirty = TRUE; + r300->dsa_state.dirty = TRUE; + r300->scissor_state.dirty = TRUE; } /* Create fragment shader state. */ @@ -601,8 +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 @@ -718,12 +692,20 @@ 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; } @@ -862,49 +844,48 @@ static void r300_set_scissor_state(struct pipe_context* pipe, { struct r300_context* r300 = r300_context(pipe); - r300_set_scissor_regs(state, &r300->scissor_state->scissor, - r300_screen(r300->context.screen)->caps->is_r500); + memcpy(r300->scissor_state.state, state, + sizeof(struct pipe_scissor_state)); - /* Don't rely on the order of states being set for the first time. */ - if (!r300->rs_state || r300->rs_state->rs.scissor) { - 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; } @@ -1011,22 +992,22 @@ static void r300_delete_vs_state(struct pipe_context* pipe, void* shader) static void r300_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct r300_context* r300 = r300_context(pipe); void *mapped; - if (buf == NULL || buf->buffer->size == 0 || - (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) + if (buf == NULL || buf->size == 0 || + (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL) { r300->shader_constants[shader].count = 0; return; } - assert((buf->buffer->size % 4 * sizeof(float)) == 0); - memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size); - r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float)); - pipe_buffer_unmap(pipe->screen, buf->buffer); + assert((buf->size % 4 * sizeof(float)) == 0); + memcpy(r300->shader_constants[shader].constants, mapped, buf->size); + r300->shader_constants[shader].count = buf->size / (4 * sizeof(float)); + pipe_buffer_unmap(pipe->screen, buf); if (shader == PIPE_SHADER_VERTEX) r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS; diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c index 22660a52d9a..192846411ba 100644 --- a/src/gallium/drivers/r300/r300_state_derived.c +++ b/src/gallium/drivers/r300/r300_state_derived.c @@ -139,10 +139,10 @@ static void r300_vertex_psc(struct r300_context* r300) /* If TCL is bypassed, map vertex streams to equivalent VS output * locations. */ - if (r300->rs_state->enable_vte) { - stream_tab = identity; - } else { + if (r300->tcl_bypass) { stream_tab = r300->vs->stream_loc_notcl; + } else { + stream_tab = identity; } /* Vertex shaders have no semantics on their inputs, @@ -508,7 +508,8 @@ static boolean r300_dsa_alpha_test_enabled(struct r300_dsa_state* dsa) 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. * @@ -530,31 +531,37 @@ static void r300_update_ztop(struct r300_context* r300) * 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. */ /* ZS writes */ - if (r300_dsa_writes_depth_stencil(r300->dsa_state) && - (r300_dsa_alpha_test_enabled(r300->dsa_state) || /* (1) */ - r300->fs->info.uses_kill)) { /* (2) */ - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; - } else if (r300_fragment_shader_writes_depth(r300->fs)) { /* (5) */ - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; - } else if (r300->query_current) { /* (6) */ - r300->ztop_state.z_buffer_top = R300_ZTOP_DISABLE; + 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 (r300->dirty_state & (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER | - R300_NEW_VERTEX_FORMAT)) { + 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_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c index f25f3ca217d..b0f309695c9 100644 --- a/src/gallium/drivers/r300/r300_state_invariant.c +++ b/src/gallium/drivers/r300/r300_state_invariant.c @@ -115,10 +115,12 @@ void r300_emit_invariant_state(struct r300_context* r300) OUT_CS_REG(R300_SC_HYPERZ, 0x0000001C); OUT_CS_REG(R300_SC_EDGERULE, 0x2DA49525); OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, 0x00000000); - if (caps->is_r500) { + + 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_BW_CNTL, 0x00000000); OUT_CS_REG(R300_ZB_DEPTHCLEARVALUE, 0x00000000); OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0x00000000); diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c index 9a96206a4dc..a9bbdd56d84 100644 --- a/src/gallium/drivers/r300/r300_texture.c +++ b/src/gallium/drivers/r300/r300_texture.c @@ -30,6 +30,18 @@ #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; @@ -92,33 +104,67 @@ 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(util_format_get_stride(tex->tex.format, u_minify(tex->tex.width0, 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++) { - unsigned nblocksy = util_format_get_nblocksy(base->format, u_minify(base->height0, 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); + nblocksy = r300_texture_get_nblocksy(tex, i); layer_size = stride * nblocksy; if (base->target == PIPE_TEXTURE_CUBE) @@ -132,9 +178,9 @@ static void r300_setup_miptree(struct r300_texture* tex) 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", + "(%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); + u_minify(base->depth0, i), stride, tex->size); } } @@ -163,7 +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); diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c index f3ac6760db5..8e017939402 100644 --- a/src/gallium/drivers/softpipe/sp_context.c +++ b/src/gallium/drivers/softpipe/sp_context.c @@ -113,8 +113,8 @@ softpipe_destroy( struct pipe_context *pipe ) } for (i = 0; i < Elements(softpipe->constants); i++) { - if (softpipe->constants[i].buffer) { - pipe_buffer_reference(&softpipe->constants[i].buffer, NULL); + if (softpipe->constants[i]) { + pipe_buffer_reference(&softpipe->constants[i], NULL); } } @@ -256,6 +256,8 @@ softpipe_create( struct pipe_screen *screen ) softpipe->pipe.draw_arrays = softpipe_draw_arrays; softpipe->pipe.draw_elements = softpipe_draw_elements; softpipe->pipe.draw_range_elements = softpipe_draw_range_elements; + softpipe->pipe.draw_arrays_instanced = softpipe_draw_arrays_instanced; + softpipe->pipe.draw_elements_instanced = softpipe_draw_elements_instanced; softpipe->pipe.clear = softpipe_clear; softpipe->pipe.flush = softpipe_flush; diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h index 73fa744f9d4..da673c57ada 100644 --- a/src/gallium/drivers/softpipe/sp_context.h +++ b/src/gallium/drivers/softpipe/sp_context.h @@ -63,7 +63,7 @@ struct softpipe_context { /** Other rendering state */ struct pipe_blend_color blend_color; struct pipe_clip_state clip; - struct pipe_constant_buffer constants[PIPE_SHADER_TYPES]; + struct pipe_buffer *constants[PIPE_SHADER_TYPES]; struct pipe_framebuffer_state framebuffer; struct pipe_poly_stipple poly_stipple; struct pipe_scissor_state scissor; diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c index 03d35fb3cb5..03b58d2fb72 100644 --- a/src/gallium/drivers/softpipe/sp_draw_arrays.c +++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c @@ -52,18 +52,18 @@ softpipe_map_constant_buffers(struct softpipe_context *sp) uint i, vssize, gssize; for (i = 0; i < PIPE_SHADER_TYPES; i++) { - if (sp->constants[i].buffer && sp->constants[i].buffer->size) - sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i].buffer, + if (sp->constants[i] && sp->constants[i]->size) + sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i], PIPE_BUFFER_USAGE_CPU_READ); } - if (sp->constants[PIPE_SHADER_VERTEX].buffer) - vssize = sp->constants[PIPE_SHADER_VERTEX].buffer->size; + if (sp->constants[PIPE_SHADER_VERTEX]) + vssize = sp->constants[PIPE_SHADER_VERTEX]->size; else vssize = 0; - if (sp->constants[PIPE_SHADER_GEOMETRY].buffer) - gssize = sp->constants[PIPE_SHADER_GEOMETRY].buffer->size; + if (sp->constants[PIPE_SHADER_GEOMETRY]) + gssize = sp->constants[PIPE_SHADER_GEOMETRY]->size; else gssize = 0; @@ -91,26 +91,48 @@ softpipe_unmap_constant_buffers(struct softpipe_context *sp) draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_GEOMETRY, NULL, 0); for (i = 0; i < PIPE_SHADER_TYPES; i++) { - if (sp->constants[i].buffer && sp->constants[i].buffer->size) - ws->buffer_unmap(ws, sp->constants[i].buffer); + if (sp->constants[i] && sp->constants[i]->size) + ws->buffer_unmap(ws, sp->constants[i]); sp->mapped_constants[i] = NULL; } } +/** + * Draw vertex arrays, with optional indexing. + * Basically, map the vertex buffers (and drawing surfaces), then hand off + * the drawing to the 'draw' module. + */ +static void +softpipe_draw_range_elements_instanced(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count, + unsigned startInstance, + unsigned instanceCount); + + void softpipe_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count) { - softpipe_draw_elements(pipe, NULL, 0, mode, start, count); + softpipe_draw_range_elements_instanced(pipe, + NULL, + 0, + 0, + 0xffffffff, + mode, + start, + count, + 0, + 1); } -/** - * Draw vertex arrays, with optional indexing. - * Basically, map the vertex buffers (and drawing surfaces), then hand off - * the drawing to the 'draw' module. - */ void softpipe_draw_range_elements(struct pipe_context *pipe, struct pipe_buffer *indexBuffer, @@ -119,6 +141,91 @@ softpipe_draw_range_elements(struct pipe_context *pipe, unsigned max_index, unsigned mode, unsigned start, unsigned count) { + softpipe_draw_range_elements_instanced(pipe, + indexBuffer, + indexSize, + min_index, + max_index, + mode, + start, + count, + 0, + 1); +} + + +void +softpipe_draw_elements(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, unsigned start, unsigned count) +{ + softpipe_draw_range_elements_instanced(pipe, + indexBuffer, + indexSize, + 0, + 0xffffffff, + mode, + start, + count, + 0, + 1); +} + +void +softpipe_draw_arrays_instanced(struct pipe_context *pipe, + unsigned mode, + unsigned start, + unsigned count, + unsigned startInstance, + unsigned instanceCount) +{ + softpipe_draw_range_elements_instanced(pipe, + NULL, + 0, + 0, + 0xffffffff, + mode, + start, + count, + startInstance, + instanceCount); +} + +void +softpipe_draw_elements_instanced(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, + unsigned start, + unsigned count, + unsigned startInstance, + unsigned instanceCount) +{ + softpipe_draw_range_elements_instanced(pipe, + indexBuffer, + indexSize, + 0, + 0xffffffff, + mode, + start, + count, + startInstance, + instanceCount); +} + +static void +softpipe_draw_range_elements_instanced(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned minIndex, + unsigned maxIndex, + unsigned mode, + unsigned start, + unsigned count, + unsigned startInstance, + unsigned instanceCount) +{ struct softpipe_context *sp = softpipe_context(pipe); struct draw_context *draw = sp->draw; unsigned i; @@ -128,45 +235,48 @@ softpipe_draw_range_elements(struct pipe_context *pipe, sp->reduced_api_prim = u_reduced_prim(mode); - if (sp->dirty) - softpipe_update_derived( sp ); + if (sp->dirty) { + softpipe_update_derived(sp); + } softpipe_map_transfers(sp); softpipe_map_constant_buffers(sp); - /* - * Map vertex buffers - */ + /* Map vertex buffers */ for (i = 0; i < sp->num_vertex_buffers; i++) { - void *buf - = pipe_buffer_map(pipe->screen, - sp->vertex_buffer[i].buffer, - PIPE_BUFFER_USAGE_CPU_READ); + void *buf; + + buf = pipe_buffer_map(pipe->screen, + sp->vertex_buffer[i].buffer, + PIPE_BUFFER_USAGE_CPU_READ); draw_set_mapped_vertex_buffer(draw, i, buf); } /* Map index buffer, if present */ if (indexBuffer) { - void *mapped_indexes - = pipe_buffer_map(pipe->screen, indexBuffer, - PIPE_BUFFER_USAGE_CPU_READ); - draw_set_mapped_element_buffer_range(draw, indexSize, - min_index, - max_index, + void *mapped_indexes; + + mapped_indexes = pipe_buffer_map(pipe->screen, + indexBuffer, + PIPE_BUFFER_USAGE_CPU_READ); + draw_set_mapped_element_buffer_range(draw, + indexSize, + minIndex, + maxIndex, mapped_indexes); - } - else { + } else { /* no index/element buffer */ - draw_set_mapped_element_buffer_range(draw, 0, start, - start + count - 1, NULL); + draw_set_mapped_element_buffer_range(draw, + 0, + start, + start + count - 1, + NULL); } /* draw! */ - draw_arrays(draw, mode, start, count); + draw_arrays_instanced(draw, mode, start, count, startInstance, instanceCount); - /* - * unmap vertex/index buffers - will cause draw module to flush - */ + /* unmap vertex/index buffers - will cause draw module to flush */ for (i = 0; i < sp->num_vertex_buffers; i++) { draw_set_mapped_vertex_buffer(draw, i, NULL); pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer); @@ -176,22 +286,8 @@ softpipe_draw_range_elements(struct pipe_context *pipe, pipe_buffer_unmap(pipe->screen, indexBuffer); } - /* Note: leave drawing surfaces mapped */ softpipe_unmap_constant_buffers(sp); sp->dirty_render_cache = TRUE; } - - -void -softpipe_draw_elements(struct pipe_context *pipe, - struct pipe_buffer *indexBuffer, - unsigned indexSize, - unsigned mode, unsigned start, unsigned count) -{ - softpipe_draw_range_elements( pipe, indexBuffer, - indexSize, - 0, 0xffffffff, - mode, start, count ); -} diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h index 9b18dac67bd..7f244c4fd49 100644 --- a/src/gallium/drivers/softpipe/sp_state.h +++ b/src/gallium/drivers/softpipe/sp_state.h @@ -139,7 +139,7 @@ void softpipe_set_clip_state( struct pipe_context *, void softpipe_set_constant_buffer(struct pipe_context *, uint shader, uint index, - const struct pipe_constant_buffer *buf); + struct pipe_buffer *buf); void *softpipe_create_fs_state(struct pipe_context *, const struct pipe_shader_state *); @@ -200,6 +200,24 @@ softpipe_draw_range_elements(struct pipe_context *pipe, unsigned mode, unsigned start, unsigned count); void +softpipe_draw_arrays_instanced(struct pipe_context *pipe, + unsigned mode, + unsigned start, + unsigned count, + unsigned startInstance, + unsigned instanceCount); + +void +softpipe_draw_elements_instanced(struct pipe_context *pipe, + struct pipe_buffer *indexBuffer, + unsigned indexSize, + unsigned mode, + unsigned start, + unsigned count, + unsigned startInstance, + unsigned instanceCount); + +void softpipe_map_transfers(struct softpipe_context *sp); void diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c index aa12bb215a8..b7ed4441b43 100644 --- a/src/gallium/drivers/softpipe/sp_state_fs.c +++ b/src/gallium/drivers/softpipe/sp_state_fs.c @@ -159,7 +159,7 @@ softpipe_delete_vs_state(struct pipe_context *pipe, void *vs) void softpipe_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct softpipe_context *softpipe = softpipe_context(pipe); @@ -169,8 +169,7 @@ softpipe_set_constant_buffer(struct pipe_context *pipe, draw_flush(softpipe->draw); /* note: reference counting */ - pipe_buffer_reference(&softpipe->constants[shader].buffer, - buf ? buf->buffer : NULL); + pipe_buffer_reference(&softpipe->constants[shader], buf); softpipe->dirty |= SP_NEW_CONSTANTS; } diff --git a/src/gallium/drivers/svga/svga_pipe_constants.c b/src/gallium/drivers/svga/svga_pipe_constants.c index 10e7a121892..ca2c7c49d72 100644 --- a/src/gallium/drivers/svga/svga_pipe_constants.c +++ b/src/gallium/drivers/svga/svga_pipe_constants.c @@ -49,7 +49,7 @@ struct svga_constbuf static void svga_set_constant_buffer(struct pipe_context *pipe, uint shader, uint index, - const struct pipe_constant_buffer *buf) + struct pipe_buffer *buf) { struct svga_context *svga = svga_context(pipe); @@ -57,7 +57,7 @@ static void svga_set_constant_buffer(struct pipe_context *pipe, assert(index == 0); pipe_buffer_reference( &svga->curr.cb[shader], - buf->buffer ); + buf ); if (shader == PIPE_SHADER_FRAGMENT) svga->dirty |= SVGA_NEW_FS_CONST_BUFFER; diff --git a/src/gallium/drivers/svga/svga_state_fs.c b/src/gallium/drivers/svga/svga_state_fs.c index bba80a93a56..ec2886348b9 100644 --- a/src/gallium/drivers/svga/svga_state_fs.c +++ b/src/gallium/drivers/svga/svga_state_fs.c @@ -72,7 +72,7 @@ static enum pipe_error compile_fs( struct svga_context *svga, struct svga_shader_result **out_result ) { struct svga_shader_result *result; - enum pipe_error ret; + enum pipe_error ret = PIPE_ERROR; result = svga_translate_fragment_program( fs, key ); if (result == NULL) { @@ -81,8 +81,10 @@ static enum pipe_error compile_fs( struct svga_context *svga, } result->id = util_bitmask_add(svga->fs_bm); - if(result->id == UTIL_BITMASK_INVALID_INDEX) + if(result->id == UTIL_BITMASK_INVALID_INDEX) { + ret = PIPE_ERROR_OUT_OF_MEMORY; goto fail; + } ret = SVGA3D_DefineShader(svga->swc, result->id, diff --git a/src/gallium/drivers/svga/svga_state_vs.c b/src/gallium/drivers/svga/svga_state_vs.c index ae1e77e7d44..e7e6c084321 100644 --- a/src/gallium/drivers/svga/svga_state_vs.c +++ b/src/gallium/drivers/svga/svga_state_vs.c @@ -71,7 +71,7 @@ static enum pipe_error compile_vs( struct svga_context *svga, struct svga_shader_result **out_result ) { struct svga_shader_result *result; - enum pipe_error ret = PIPE_OK; + enum pipe_error ret = PIPE_ERROR; result = svga_translate_vertex_program( vs, key ); if (result == NULL) { @@ -80,8 +80,10 @@ static enum pipe_error compile_vs( struct svga_context *svga, } result->id = util_bitmask_add(svga->vs_bm); - if(result->id == UTIL_BITMASK_INVALID_INDEX) + if(result->id == UTIL_BITMASK_INVALID_INDEX) { + ret = PIPE_ERROR_OUT_OF_MEMORY; goto fail; + } ret = SVGA3D_DefineShader(svga->swc, result->id, @@ -200,10 +202,12 @@ static int update_zero_stride( struct svga_context *svga, key.output_stride = 4 * sizeof(float); key.nr_elements = 1; + key.element[0].type = TRANSLATE_ELEMENT_NORMAL; key.element[0].input_format = vel->src_format; key.element[0].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT; key.element[0].input_buffer = vel->vertex_buffer_index; key.element[0].input_offset = vel->src_offset; + key.element[0].instance_divisor = vel->instance_divisor; key.element[0].output_offset = const_idx * 4 * sizeof(float); translate_key_sanitize(&key); @@ -222,7 +226,7 @@ static int update_zero_stride( struct svga_context *svga, translate->set_buffer(translate, vel->vertex_buffer_index, mapped_buffer, vbuffer->stride); - translate->run(translate, 0, 1, + translate->run(translate, 0, 1, 0, svga->curr.zero_stride_constants); pipe_buffer_unmap(svga->pipe.screen, diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c index 075e4f9a0b2..5a9f0fc6901 100644 --- a/src/gallium/drivers/trace/tr_context.c +++ b/src/gallium/drivers/trace/tr_context.c @@ -812,13 +812,13 @@ trace_context_set_clip_state(struct pipe_context *_pipe, static INLINE void trace_context_set_constant_buffer(struct pipe_context *_pipe, uint shader, uint index, - const struct pipe_constant_buffer *buffer) + struct pipe_buffer *buffer) { struct trace_context *tr_ctx = trace_context(_pipe); struct pipe_context *pipe = tr_ctx->pipe; if (buffer) - trace_screen_user_buffer_update(_pipe->screen, buffer->buffer); + trace_screen_user_buffer_update(_pipe->screen, buffer); trace_dump_call_begin("pipe_context", "set_constant_buffer"); @@ -827,10 +827,11 @@ trace_context_set_constant_buffer(struct pipe_context *_pipe, trace_dump_arg(uint, index); trace_dump_arg(constant_buffer, buffer); + /* XXX hmm? */ if (buffer) { - struct pipe_constant_buffer _buffer; - _buffer.buffer = trace_buffer_unwrap(tr_ctx, buffer->buffer); - pipe->set_constant_buffer(pipe, shader, index, &_buffer); + struct pipe_buffer *_buffer; + _buffer = trace_buffer_unwrap(tr_ctx, buffer); + pipe->set_constant_buffer(pipe, shader, index, _buffer); } else { pipe->set_constant_buffer(pipe, shader, index, buffer); } diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c index 86237e03bcc..32f61f8c944 100644 --- a/src/gallium/drivers/trace/tr_dump_state.c +++ b/src/gallium/drivers/trace/tr_dump_state.c @@ -229,7 +229,7 @@ void trace_dump_clip_state(const struct pipe_clip_state *state) } -void trace_dump_constant_buffer(const struct pipe_constant_buffer *state) +void trace_dump_constant_buffer(const struct pipe_buffer *state) { if (!trace_dumping_enabled_locked()) return; @@ -241,7 +241,7 @@ void trace_dump_constant_buffer(const struct pipe_constant_buffer *state) trace_dump_struct_begin("pipe_constant_buffer"); - trace_dump_member(buffer_ptr, state, buffer); + trace_dump_reference(&state->reference); trace_dump_struct_end(); } diff --git a/src/gallium/drivers/trace/tr_dump_state.h b/src/gallium/drivers/trace/tr_dump_state.h index 07ad6fbb205..c7860fd6e18 100644 --- a/src/gallium/drivers/trace/tr_dump_state.h +++ b/src/gallium/drivers/trace/tr_dump_state.h @@ -47,7 +47,7 @@ void trace_dump_scissor_state(const struct pipe_scissor_state *state); void trace_dump_clip_state(const struct pipe_clip_state *state); -void trace_dump_constant_buffer(const struct pipe_constant_buffer *state); +void trace_dump_constant_buffer(const struct pipe_buffer *state); void trace_dump_token(const struct tgsi_token *token); |