diff options
Diffstat (limited to 'src/mesa/drivers/dri/i965')
34 files changed, 1435 insertions, 583 deletions
diff --git a/src/mesa/drivers/dri/i965/Makefile b/src/mesa/drivers/dri/i965/Makefile index 2934414d99..9712c38725 100644 --- a/src/mesa/drivers/dri/i965/Makefile +++ b/src/mesa/drivers/dri/i965/Makefile @@ -14,6 +14,7 @@ DRIVER_SOURCES = \ intel_decode.c \ intel_extensions.c \ intel_fbo.c \ + intel_generatemipmap.c \ intel_mipmap_tree.c \ intel_regions.c \ intel_screen.c \ @@ -69,6 +70,7 @@ DRIVER_SOURCES = \ brw_vs_constval.c \ brw_vs_emit.c \ brw_vs_state.c \ + brw_vs_surface_state.c \ brw_vtbl.c \ brw_wm.c \ brw_wm_debug.c \ diff --git a/src/mesa/drivers/dri/i965/brw_context.c b/src/mesa/drivers/dri/i965/brw_context.c index d96ff29310..4dbe551d83 100644 --- a/src/mesa/drivers/dri/i965/brw_context.c +++ b/src/mesa/drivers/dri/i965/brw_context.c @@ -118,6 +118,8 @@ GLboolean brwCreateContext( const __GLcontextModes *mesaVis, ctx->Const.MaxCubeTextureLevels = 12; ctx->Const.MaxTextureRectSize = (1<<12); + ctx->Const.MaxTextureMaxAnisotropy = 16.0; + /* if conformance mode is set, swrast can handle any size AA point */ ctx->Const.MaxPointSizeAA = 255.0; diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h index 01e07c967f..873fc8ffff 100644 --- a/src/mesa/drivers/dri/i965/brw_context.h +++ b/src/mesa/drivers/dri/i965/brw_context.h @@ -131,6 +131,7 @@ struct brw_context; #define BRW_NEW_WM_INPUT_DIMENSIONS 0x100 #define BRW_NEW_INPUT_VARYING 0x200 #define BRW_NEW_PSP 0x800 +#define BRW_NEW_WM_SURFACES 0x1000 #define BRW_NEW_FENCE 0x2000 #define BRW_NEW_INDICES 0x4000 #define BRW_NEW_VERTICES 0x8000 @@ -141,7 +142,8 @@ struct brw_context; #define BRW_NEW_BATCH 0x10000 /** brw->depth_region updated */ #define BRW_NEW_DEPTH_BUFFER 0x20000 -#define BRW_NEW_NR_SURFACES 0x40000 +#define BRW_NEW_NR_WM_SURFACES 0x40000 +#define BRW_NEW_NR_VS_SURFACES 0x80000 struct brw_state_flags { /** State update flags signalled by mesa internals */ @@ -159,6 +161,8 @@ struct brw_state_flags { struct brw_vertex_program { struct gl_vertex_program program; GLuint id; + dri_bo *const_buffer; /** Program constant buffer/surface */ + GLboolean use_const_buffer; }; @@ -168,8 +172,8 @@ struct brw_fragment_program { GLuint id; /**< serial no. to identify frag progs, never re-used */ GLboolean isGLSL; /**< really, any IF/LOOP/CONT/BREAK instructions */ - /** Program constant buffer/surface */ - dri_bo *const_buffer; + dri_bo *const_buffer; /** Program constant buffer/surface */ + GLboolean use_const_buffer; }; @@ -186,7 +190,7 @@ struct brw_wm_prog_data { GLuint total_grf; GLuint total_scratch; - GLuint nr_params; + GLuint nr_params; /**< number of float params/constants */ GLboolean error; /* Pointer to tracked values (only valid once @@ -225,6 +229,7 @@ struct brw_vs_prog_data { GLuint urb_read_length; GLuint total_grf; GLuint outputs_written; + GLuint nr_params; /**< number of float params/constants */ GLuint inputs_read; @@ -241,16 +246,38 @@ struct brw_vs_ouput_sizes { }; +/** Number of general purpose registers (VS, WM, etc) */ +#define BRW_MAX_GRF 128 + /** Number of texture sampler units */ #define BRW_MAX_TEX_UNIT 16 /** - * Size of our surface binding table. + * Size of our surface binding table for the WM. * This contains pointers to the drawing surfaces and current texture - * objects and shader constant buffer (+1). + * objects and shader constant buffers (+2). */ #define BRW_WM_MAX_SURF (MAX_DRAW_BUFFERS + BRW_MAX_TEX_UNIT + 1) +/** + * Helpers to convert drawing buffers, textures and constant buffers + * to surface binding table indexes, for WM. + */ +#define SURF_INDEX_DRAW(d) (d) +#define SURF_INDEX_FRAG_CONST_BUFFER (MAX_DRAW_BUFFERS) +#define SURF_INDEX_TEXTURE(t) (MAX_DRAW_BUFFERS + 1 + (t)) + +/** + * Size of surface binding table for the VS. + * Only one constant buffer for now. + */ +#define BRW_VS_MAX_SURF 1 + +/** + * Only a VS constant buffer + */ +#define SURF_INDEX_VERT_CONST_BUFFER 0 + enum brw_cache_id { BRW_CC_VP, @@ -427,8 +454,6 @@ struct brw_context struct { struct brw_state_flags dirty; - struct brw_tracked_state **atoms; - GLuint nr_atoms; GLuint nr_color_regions; struct intel_region *color_regions[MAX_DRAW_BUFFERS]; @@ -448,7 +473,8 @@ struct brw_context int validated_bo_count; } state; - struct brw_cache cache; + struct brw_cache cache; /** non-surface items */ + struct brw_cache surface_cache; /* surface items */ struct brw_cached_batch_item *cached_batch_items; struct { @@ -532,11 +558,6 @@ struct brw_context GLuint vs_size; GLuint total_size; - /* Dynamic tracker which changes to reflect the state referenced - * by active fp and vp program parameters: - */ - struct brw_tracked_state tracked_state; - dri_bo *curbe_bo; /** Offset within curbe_bo of space for current curbe entry */ GLuint curbe_offset; @@ -557,6 +578,11 @@ struct brw_context dri_bo *prog_bo; dri_bo *state_bo; + + /** Binding table of pointers to surf_bo entries */ + dri_bo *bind_bo; + dri_bo *surf_bo[BRW_VS_MAX_SURF]; + GLuint nr_surfaces; } vs; struct { diff --git a/src/mesa/drivers/dri/i965/brw_curbe.c b/src/mesa/drivers/dri/i965/brw_curbe.c index a6bfb7507e..a1a6c53d0e 100644 --- a/src/mesa/drivers/dri/i965/brw_curbe.c +++ b/src/mesa/drivers/dri/i965/brw_curbe.c @@ -36,6 +36,7 @@ #include "main/macros.h" #include "main/enums.h" #include "shader/prog_parameter.h" +#include "shader/prog_print.h" #include "shader/prog_statevars.h" #include "intel_batchbuffer.h" #include "intel_regions.h" @@ -45,17 +46,21 @@ #include "brw_util.h" -/* Partition the CURBE between the various users of constant values: +/** + * Partition the CURBE between the various users of constant values: + * Note that vertex and fragment shaders can now fetch constants out + * of constant buffers. We no longer allocatea block of the GRF for + * constants. That greatly reduces the demand for space in the CURBE. + * Some of the comments within are dated... */ static void calculate_curbe_offsets( struct brw_context *brw ) { GLcontext *ctx = &brw->intel.ctx; /* CACHE_NEW_WM_PROG */ - GLuint nr_fp_regs = (brw->wm.prog_data->nr_params + 15) / 16; + const GLuint nr_fp_regs = (brw->wm.prog_data->nr_params + 15) / 16; /* BRW_NEW_VERTEX_PROGRAM */ - const struct brw_vertex_program *vp = brw_vertex_program_const(brw->vertex_program); - GLuint nr_vp_regs = (vp->program.Base.Parameters->NumParameters * 4 + 15) / 16; + const GLuint nr_vp_regs = (brw->vs.prog_data->nr_params + 15) / 16; GLuint nr_clip_regs = 0; GLuint total_regs; @@ -184,13 +189,6 @@ static void prepare_constant_buffer(struct brw_context *brw) GLfloat *buf; GLuint i; - /* Update our own dependency flags. This works because this - * function will also be called whenever fp or vp changes. - */ - brw->curbe.tracked_state.dirty.mesa = (_NEW_TRANSFORM|_NEW_PROJECTION); - brw->curbe.tracked_state.dirty.mesa |= vp->program.Base.Parameters->StateFlags; - brw->curbe.tracked_state.dirty.mesa |= fp->program.Base.Parameters->StateFlags; - if (sz == 0) { if (brw->curbe.last_buf) { free(brw->curbe.last_buf); @@ -248,7 +246,7 @@ static void prepare_constant_buffer(struct brw_context *brw) /* vertex shader constants */ if (brw->curbe.vs_size) { GLuint offset = brw->curbe.vs_start * 16; - GLuint nr = vp->program.Base.Parameters->NumParameters; + GLuint nr = brw->vs.prog_data->nr_params / 4; _mesa_load_state_parameters(ctx, vp->program.Base.Parameters); @@ -331,40 +329,11 @@ static void prepare_constant_buffer(struct brw_context *brw) */ } - -/** - * Vertex/fragment shader constants are stored in a pseudo 1D texture. - * This function updates the constants in that buffer. - */ -static void -update_texture_constant_buffer(struct brw_context *brw) -{ - struct brw_fragment_program *fp = - (struct brw_fragment_program *) brw->fragment_program; - const struct gl_program_parameter_list *params = fp->program.Base.Parameters; - const int size = params->NumParameters * 4 * sizeof(GLfloat); - - assert(fp->const_buffer); - assert(fp->const_buffer->size >= size); - - /* copy constants into the buffer */ - if (size > 0) { - GLubyte *map; - dri_bo_map(fp->const_buffer, GL_TRUE); - map = fp->const_buffer->virtual; - memcpy(map, params->ParameterValues, size); - dri_bo_unmap(fp->const_buffer); - } -} - - static void emit_constant_buffer(struct brw_context *brw) { struct intel_context *intel = &brw->intel; GLuint sz = brw->curbe.total_size; - update_texture_constant_buffer(brw); - BEGIN_BATCH(2, IGNORE_CLIPRECTS); if (sz == 0) { OUT_BATCH((CMD_CONST_BUFFER << 16) | (2 - 2)); @@ -386,7 +355,7 @@ static void emit_constant_buffer(struct brw_context *brw) */ const struct brw_tracked_state brw_constant_buffer = { .dirty = { - .mesa = (_NEW_TRANSFORM|_NEW_PROJECTION), /* plus fp and vp flags */ + .mesa = _NEW_PROGRAM_CONSTANTS, .brw = (BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_VERTEX_PROGRAM | BRW_NEW_URB_FENCE | /* Implicit - hardware requires this, not used above */ diff --git a/src/mesa/drivers/dri/i965/brw_draw_upload.c b/src/mesa/drivers/dri/i965/brw_draw_upload.c index b91b20bec6..1b8bcc14ec 100644 --- a/src/mesa/drivers/dri/i965/brw_draw_upload.c +++ b/src/mesa/drivers/dri/i965/brw_draw_upload.c @@ -343,7 +343,7 @@ static void brw_prepare_vertices(struct brw_context *brw) { GLcontext *ctx = &brw->intel.ctx; struct intel_context *intel = intel_context(ctx); - GLuint tmp = brw->vs.prog_data->inputs_read; + GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; GLuint i; const unsigned char *ptr = NULL; GLuint interleave = 0; @@ -362,11 +362,11 @@ static void brw_prepare_vertices(struct brw_context *brw) _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index); /* Accumulate the list of enabled arrays. */ - while (tmp) { - GLuint i = _mesa_ffsll(tmp)-1; + while (vs_inputs) { + GLuint i = _mesa_ffsll(vs_inputs) - 1; struct brw_vertex_element *input = &brw->vb.inputs[i]; - tmp &= ~(1<<i); + vs_inputs &= ~(1 << i); enabled[nr_enabled++] = input; } @@ -477,17 +477,17 @@ static void brw_emit_vertices(struct brw_context *brw) { GLcontext *ctx = &brw->intel.ctx; struct intel_context *intel = intel_context(ctx); - GLuint tmp = brw->vs.prog_data->inputs_read; + GLbitfield vs_inputs = brw->vs.prog_data->inputs_read; struct brw_vertex_element *enabled[VERT_ATTRIB_MAX]; GLuint i; GLuint nr_enabled = 0; /* Accumulate the list of enabled arrays. */ - while (tmp) { - i = _mesa_ffsll(tmp)-1; + while (vs_inputs) { + i = _mesa_ffsll(vs_inputs) - 1; struct brw_vertex_element *input = &brw->vb.inputs[i]; - tmp &= ~(1<<i); + vs_inputs &= ~(1 << i); enabled[nr_enabled++] = input; } diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h index d05f2e6c41..bc7756ceab 100644 --- a/src/mesa/drivers/dri/i965/brw_eu.h +++ b/src/mesa/drivers/dri/i965/brw_eu.h @@ -97,7 +97,7 @@ struct brw_glsl_call; #define BRW_EU_MAX_INSN_STACK 5 -#define BRW_EU_MAX_INSN 4000 +#define BRW_EU_MAX_INSN 10000 struct brw_compile { struct brw_instruction store[BRW_EU_MAX_INSN]; @@ -862,9 +862,17 @@ void brw_dp_READ_4( struct brw_compile *p, struct brw_reg dest, GLuint msg_reg_nr, GLboolean relAddr, - GLuint scratch_offset, + GLuint location, GLuint bind_table_index ); +void brw_dp_READ_4_vs( struct brw_compile *p, + struct brw_reg dest, + GLuint oword, + GLboolean relAddr, + struct brw_reg addrReg, + GLuint location, + GLuint bind_table_index ); + void brw_dp_WRITE_16( struct brw_compile *p, struct brw_reg src, GLuint msg_reg_nr, diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c index 21ce8369db..60ea44f7a9 100644 --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c @@ -952,7 +952,7 @@ void brw_dp_READ_16( struct brw_compile *p, /** * Read a float[4] vector from the data port Data Cache (const buffer). - * Scratch offset should be a multiple of 16. + * Location (in buffer) should be a multiple of 16. * Used for fetching shader constants. * If relAddr is true, we'll do an indirect fetch using the address register. */ @@ -960,7 +960,7 @@ void brw_dp_READ_4( struct brw_compile *p, struct brw_reg dest, GLuint msg_reg_nr, GLboolean relAddr, - GLuint scratch_offset, + GLuint location, GLuint bind_table_index ) { { @@ -969,18 +969,20 @@ void brw_dp_READ_4( struct brw_compile *p, brw_set_mask_control(p, BRW_MASK_DISABLE); /* set message header global offset field (reg 0, element 2) */ + /* Note that grf[0] will be copied to mrf[1] implicitly by the SEND instr */ brw_MOV(p, retype(brw_vec1_grf(0, 2), BRW_REGISTER_TYPE_UD), - brw_imm_d(scratch_offset)); + brw_imm_d(location)); brw_pop_insn_state(p); } { struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); - insn->header.predicate_control = 0; /* XXX */ + insn->header.predicate_control = BRW_PREDICATE_NONE; insn->header.compression_control = BRW_COMPRESSION_NONE; insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.mask_control = BRW_MASK_DISABLE; /* cast dest to a uword[8] vector */ dest = retype(vec8(dest), BRW_REGISTER_TYPE_UW); @@ -989,7 +991,7 @@ void brw_dp_READ_4( struct brw_compile *p, brw_set_src0(insn, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW)); brw_set_dp_read_message(insn, - bind_table_index, /* binding table index (255=stateless) */ + bind_table_index, 0, /* msg_control (0 means 1 Oword) */ BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */ 0, /* source cache = data cache */ @@ -1000,6 +1002,78 @@ void brw_dp_READ_4( struct brw_compile *p, } +/** + * Read float[4] constant(s) from VS constant buffer. + * For relative addressing, two float[4] constants will be read into 'dest'. + * Otherwise, one float[4] constant will be read into the lower half of 'dest'. + */ +void brw_dp_READ_4_vs(struct brw_compile *p, + struct brw_reg dest, + GLuint oword, + GLboolean relAddr, + struct brw_reg addrReg, + GLuint location, + GLuint bind_table_index) +{ + GLuint msg_reg_nr = 1; + + assert(oword < 2); + /* + printf("vs const read msg, location %u, msg_reg_nr %d\n", + location, msg_reg_nr); + */ + + /* Setup MRF[1] with location/offset into const buffer */ + { + struct brw_reg b; + + brw_push_insn_state(p); + brw_set_compression_control(p, BRW_COMPRESSION_NONE); + brw_set_mask_control(p, BRW_MASK_DISABLE); + brw_set_predicate_control(p, BRW_PREDICATE_NONE); + /*brw_set_access_mode(p, BRW_ALIGN_16);*/ + + /* XXX I think we're setting all the dwords of MRF[1] to 'location'. + * when the docs say only dword[2] should be set. Hmmm. But it works. + */ + b = brw_message_reg(msg_reg_nr); + b = retype(b, BRW_REGISTER_TYPE_UD); + /*b = get_element_ud(b, 2);*/ + if (relAddr) { + brw_ADD(p, b, addrReg, brw_imm_ud(location)); + } + else { + brw_MOV(p, b, brw_imm_ud(location)); + } + + brw_pop_insn_state(p); + } + + { + struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND); + + insn->header.predicate_control = BRW_PREDICATE_NONE; + insn->header.compression_control = BRW_COMPRESSION_NONE; + insn->header.destreg__conditonalmod = msg_reg_nr; + insn->header.mask_control = BRW_MASK_DISABLE; + /*insn->header.access_mode = BRW_ALIGN_16;*/ + + brw_set_dest(insn, dest); + brw_set_src0(insn, brw_null_reg()); + + brw_set_dp_read_message(insn, + bind_table_index, + oword, /* 0 = lower Oword, 1 = upper Oword */ + BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */ + 0, /* source cache = data cache */ + 1, /* msg_length */ + 1, /* response_length (1 Oword) */ + 0); /* eot */ + } +} + + + void brw_fb_WRITE(struct brw_compile *p, struct brw_reg dest, GLuint msg_reg_nr, diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c index 5c94a49f60..4784254bc7 100644 --- a/src/mesa/drivers/dri/i965/brw_misc_state.c +++ b/src/mesa/drivers/dri/i965/brw_misc_state.c @@ -101,6 +101,7 @@ const struct brw_tracked_state brw_drawing_rect = { static void prepare_binding_table_pointers(struct brw_context *brw) { + brw_add_validated_bo(brw, brw->vs.bind_bo); brw_add_validated_bo(brw, brw->wm.bind_bo); } @@ -117,13 +118,14 @@ static void upload_binding_table_pointers(struct brw_context *brw) BEGIN_BATCH(6, IGNORE_CLIPRECTS); OUT_BATCH(CMD_BINDING_TABLE_PTRS << 16 | (6 - 2)); - OUT_BATCH(0); /* vs */ + if (brw->vs.bind_bo != NULL) + OUT_RELOC(brw->vs.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* vs */ + else + OUT_BATCH(0); OUT_BATCH(0); /* gs */ OUT_BATCH(0); /* clip */ OUT_BATCH(0); /* sf */ - OUT_RELOC(brw->wm.bind_bo, - I915_GEM_DOMAIN_SAMPLER, 0, - 0); + OUT_RELOC(brw->wm.bind_bo, I915_GEM_DOMAIN_SAMPLER, 0, 0); /* wm/ps */ ADVANCE_BATCH(); } diff --git a/src/mesa/drivers/dri/i965/brw_program.c b/src/mesa/drivers/dri/i965/brw_program.c index 457bc2fc7f..bac69187c1 100644 --- a/src/mesa/drivers/dri/i965/brw_program.c +++ b/src/mesa/drivers/dri/i965/brw_program.c @@ -95,6 +95,12 @@ static struct gl_program *brwNewProgram( GLcontext *ctx, static void brwDeleteProgram( GLcontext *ctx, struct gl_program *prog ) { + if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) { + struct gl_fragment_program *fprog = (struct gl_fragment_program *) prog; + struct brw_fragment_program *brw_fprog = brw_fragment_program(fprog); + dri_bo_unreference(brw_fprog->const_buffer); + } + _mesa_delete_program( ctx, prog ); } @@ -111,7 +117,6 @@ static void brwProgramStringNotify( GLcontext *ctx, struct gl_program *prog ) { struct brw_context *brw = brw_context(ctx); - struct intel_context *intel = &brw->intel; if (target == GL_FRAGMENT_PROGRAM_ARB) { struct gl_fragment_program *fprog = (struct gl_fragment_program *) prog; @@ -128,24 +133,6 @@ static void brwProgramStringNotify( GLcontext *ctx, brw->state.dirty.brw |= BRW_NEW_FRAGMENT_PROGRAM; newFP->id = brw->program_id++; newFP->isGLSL = brw_wm_is_glsl(fprog); - - /* alloc constant buffer/surface */ - { - const struct gl_program_parameter_list *params = prog->Parameters; - const int size = params->NumParameters * 4 * sizeof(GLfloat); - - /* free old const buffer if too small */ - if (newFP->const_buffer && newFP->const_buffer->size < size) { - dri_bo_unreference(newFP->const_buffer); - newFP->const_buffer = NULL; - } - - if (!newFP->const_buffer) { - newFP->const_buffer = drm_intel_bo_alloc(intel->bufmgr, - "fp_const_buffer", - size, 64); - } - } } else if (target == GL_VERTEX_PROGRAM_ARB) { struct gl_vertex_program *vprog = (struct gl_vertex_program *) prog; diff --git a/src/mesa/drivers/dri/i965/brw_sf_state.c b/src/mesa/drivers/dri/i965/brw_sf_state.c index fc4eddda0a..c99918724b 100644 --- a/src/mesa/drivers/dri/i965/brw_sf_state.c +++ b/src/mesa/drivers/dri/i965/brw_sf_state.c @@ -66,7 +66,9 @@ static void upload_sf_vp(struct brw_context *brw) sfv.viewport.m31 = v[MAT_TY] * y_scale + y_bias; sfv.viewport.m32 = v[MAT_TZ] * depth_scale; - /* _NEW_SCISSOR */ + /* _NEW_SCISSOR | _NEW_BUFFERS | _NEW_VIEWPORT + * for DrawBuffer->_[XY]{min,max} + */ /* The scissor only needs to handle the intersection of drawable and * scissor rect. Clipping to the boundaries of static shared buffers @@ -97,7 +99,8 @@ static void upload_sf_vp(struct brw_context *brw) const struct brw_tracked_state brw_sf_vp = { .dirty = { .mesa = (_NEW_VIEWPORT | - _NEW_SCISSOR), + _NEW_SCISSOR | + _NEW_BUFFERS), .brw = 0, .cache = 0 }, @@ -147,7 +150,7 @@ sf_unit_populate_key(struct brw_context *brw, struct brw_sf_unit_key *key) key->line_smooth = ctx->Line.SmoothFlag; key->point_sprite = ctx->Point.PointSprite; - key->point_size = ctx->Point.Size; + key->point_size = CLAMP(ctx->Point.Size, ctx->Point.MinSize, ctx->Point.MaxSize); key->point_attenuated = ctx->Point._Attenuated; key->render_to_fbo = brw->intel.ctx.DrawBuffer->Name != 0; diff --git a/src/mesa/drivers/dri/i965/brw_state.h b/src/mesa/drivers/dri/i965/brw_state.h index 81b0a45998..bf9f6cae55 100644 --- a/src/mesa/drivers/dri/i965/brw_state.h +++ b/src/mesa/drivers/dri/i965/brw_state.h @@ -72,11 +72,13 @@ const struct brw_tracked_state brw_sf_vp; const struct brw_tracked_state brw_state_base_address; const struct brw_tracked_state brw_urb_fence; const struct brw_tracked_state brw_vertex_state; +const struct brw_tracked_state brw_vs_surfaces; const struct brw_tracked_state brw_vs_prog; const struct brw_tracked_state brw_vs_unit; const struct brw_tracked_state brw_wm_input_sizes; const struct brw_tracked_state brw_wm_prog; const struct brw_tracked_state brw_wm_samplers; +const struct brw_tracked_state brw_wm_constant_surface; const struct brw_tracked_state brw_wm_surfaces; const struct brw_tracked_state brw_wm_unit; @@ -91,6 +93,20 @@ const struct brw_tracked_state brw_drawing_rect; const struct brw_tracked_state brw_indices; const struct brw_tracked_state brw_vertices; +/** + * Use same key for WM and VS surfaces. + */ +struct brw_surface_key { + GLenum target, depthmode; + dri_bo *bo; + GLint format, internal_format; + GLint first_level, last_level; + GLint width, height, depth; + GLint pitch, cpp; + uint32_t tiling; + GLuint offset; +}; + /*********************************************************************** * brw_state.c */ @@ -135,8 +151,8 @@ dri_bo *brw_search_cache( struct brw_cache *cache, void *aux_return); void brw_state_cache_check_size( struct brw_context *brw ); -void brw_init_cache( struct brw_context *brw ); -void brw_destroy_cache( struct brw_context *brw ); +void brw_init_caches( struct brw_context *brw ); +void brw_destroy_caches( struct brw_context *brw ); /*********************************************************************** * brw_state_batch.c @@ -150,4 +166,9 @@ GLboolean brw_cached_batch_struct( struct brw_context *brw, void brw_destroy_batch_cache( struct brw_context *brw ); void brw_clear_batch_cache_flush( struct brw_context *brw ); +/* brw_wm_surface_state.c */ +dri_bo * +brw_create_constant_surface( struct brw_context *brw, + struct brw_surface_key *key ); + #endif diff --git a/src/mesa/drivers/dri/i965/brw_state_cache.c b/src/mesa/drivers/dri/i965/brw_state_cache.c index d5b5166406..e40d7a0416 100644 --- a/src/mesa/drivers/dri/i965/brw_state_cache.c +++ b/src/mesa/drivers/dri/i965/brw_state_cache.c @@ -56,9 +56,9 @@ * incorrect program is run for the other instance. */ +#include "main/imports.h" #include "brw_state.h" #include "intel_batchbuffer.h" -#include "main/imports.h" /* XXX: Fixme - have to include these to get the sizes of the prog_key * structs: @@ -69,8 +69,10 @@ #include "brw_sf.h" #include "brw_gs.h" -static GLuint hash_key( const void *key, GLuint key_size, - dri_bo **reloc_bufs, GLuint nr_reloc_bufs) + +static GLuint +hash_key(const void *key, GLuint key_size, + dri_bo **reloc_bufs, GLuint nr_reloc_bufs) { GLuint *ikey = (GLuint *)key; GLuint hash = 0, i; @@ -95,6 +97,7 @@ static GLuint hash_key( const void *key, GLuint key_size, return hash; } + /** * Marks a new buffer as being chosen for the given cache id. */ @@ -111,6 +114,7 @@ update_cache_last(struct brw_cache *cache, enum brw_cache_id cache_id, cache->brw->state.dirty.cache |= 1 << cache_id; } + static struct brw_cache_item * search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, GLuint hash, const void *key, GLuint key_size, @@ -143,7 +147,8 @@ search_cache(struct brw_cache *cache, enum brw_cache_id cache_id, } -static void rehash( struct brw_cache *cache ) +static void +rehash(struct brw_cache *cache) { struct brw_cache_item **items; struct brw_cache_item *c, *next; @@ -164,15 +169,17 @@ static void rehash( struct brw_cache *cache ) cache->size = size; } + /** * Returns the buffer object matching cache_id and key, or NULL. */ -dri_bo *brw_search_cache( struct brw_cache *cache, - enum brw_cache_id cache_id, - const void *key, - GLuint key_size, - dri_bo **reloc_bufs, GLuint nr_reloc_bufs, - void *aux_return ) +dri_bo * +brw_search_cache(struct brw_cache *cache, + enum brw_cache_id cache_id, + const void *key, + GLuint key_size, + dri_bo **reloc_bufs, GLuint nr_reloc_bufs, + void *aux_return) { struct brw_cache_item *item; GLuint hash = hash_key(key, key_size, reloc_bufs, nr_reloc_bufs); @@ -192,6 +199,7 @@ dri_bo *brw_search_cache( struct brw_cache *cache, return item->bo; } + dri_bo * brw_upload_cache( struct brw_cache *cache, enum brw_cache_id cache_id, @@ -265,7 +273,9 @@ brw_upload_cache( struct brw_cache *cache, return bo; } -/* This doesn't really work with aux data. Use search/upload instead + +/** + * This doesn't really work with aux data. Use search/upload instead */ dri_bo * brw_cache_data_sz(struct brw_cache *cache, @@ -296,6 +306,7 @@ brw_cache_data_sz(struct brw_cache *cache, return bo; } + /** * Wrapper around brw_cache_data_sz using the cache_id's canonical key size. * @@ -319,21 +330,22 @@ enum pool_type { DW_GENERAL_STATE }; + static void -brw_init_cache_id( struct brw_context *brw, - const char *name, - enum brw_cache_id id, - GLuint key_size, - GLuint aux_size) +brw_init_cache_id(struct brw_cache *cache, + const char *name, + enum brw_cache_id id, + GLuint key_size, + GLuint aux_size) { - struct brw_cache *cache = &brw->cache; - cache->name[id] = strdup(name); cache->key_size[id] = key_size; cache->aux_size[id] = aux_size; } -void brw_init_cache( struct brw_context *brw ) + +static void +brw_init_non_surface_cache(struct brw_context *brw) { struct brw_cache *cache = &brw->cache; @@ -342,114 +354,136 @@ void brw_init_cache( struct brw_context *brw ) cache->size = 7; cache->n_items = 0; cache->items = (struct brw_cache_item **) - _mesa_calloc(cache->size * - sizeof(struct brw_cache_item)); + _mesa_calloc(cache->size * sizeof(struct brw_cache_item)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CC_VP", BRW_CC_VP, sizeof(struct brw_cc_viewport), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CC_UNIT", BRW_CC_UNIT, sizeof(struct brw_cc_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "WM_PROG", BRW_WM_PROG, sizeof(struct brw_wm_prog_key), sizeof(struct brw_wm_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SAMPLER_DEFAULT_COLOR", BRW_SAMPLER_DEFAULT_COLOR, sizeof(struct brw_sampler_default_color), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SAMPLER", BRW_SAMPLER, 0, /* variable key/data size */ 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "WM_UNIT", BRW_WM_UNIT, sizeof(struct brw_wm_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SF_PROG", BRW_SF_PROG, sizeof(struct brw_sf_prog_key), sizeof(struct brw_sf_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SF_VP", BRW_SF_VP, sizeof(struct brw_sf_viewport), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SF_UNIT", BRW_SF_UNIT, sizeof(struct brw_sf_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "VS_UNIT", BRW_VS_UNIT, sizeof(struct brw_vs_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "VS_PROG", BRW_VS_PROG, sizeof(struct brw_vs_prog_key), sizeof(struct brw_vs_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CLIP_UNIT", BRW_CLIP_UNIT, sizeof(struct brw_clip_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "CLIP_PROG", BRW_CLIP_PROG, sizeof(struct brw_clip_prog_key), sizeof(struct brw_clip_prog_data)); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "GS_UNIT", BRW_GS_UNIT, sizeof(struct brw_gs_unit_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "GS_PROG", BRW_GS_PROG, sizeof(struct brw_gs_prog_key), sizeof(struct brw_gs_prog_data)); +} + + +static void +brw_init_surface_cache(struct brw_context *brw) +{ + struct brw_cache *cache = &brw->surface_cache; + + cache->brw = brw; - brw_init_cache_id(brw, + cache->size = 7; + cache->n_items = 0; + cache->items = (struct brw_cache_item **) + _mesa_calloc(cache->size * sizeof(struct brw_cache_item)); + + brw_init_cache_id(cache, "SS_SURFACE", BRW_SS_SURFACE, sizeof(struct brw_surface_state), 0); - brw_init_cache_id(brw, + brw_init_cache_id(cache, "SS_SURF_BIND", BRW_SS_SURF_BIND, 0, 0); } + +void +brw_init_caches(struct brw_context *brw) +{ + brw_init_non_surface_cache(brw); + brw_init_surface_cache(brw); +} + + static void -brw_clear_cache( struct brw_context *brw ) +brw_clear_cache(struct brw_context *brw, struct brw_cache *cache) { struct brw_cache_item *c, *next; GLuint i; @@ -457,8 +491,8 @@ brw_clear_cache( struct brw_context *brw ) if (INTEL_DEBUG & DEBUG_STATE) _mesa_printf("%s\n", __FUNCTION__); - for (i = 0; i < brw->cache.size; i++) { - for (c = brw->cache.items[i]; c; c = next) { + for (i = 0; i < cache->size; i++) { + for (c = cache->items[i]; c; c = next) { int j; next = c->next; @@ -468,10 +502,10 @@ brw_clear_cache( struct brw_context *brw ) free((void *)c->key); free(c); } - brw->cache.items[i] = NULL; + cache->items[i] = NULL; } - brw->cache.n_items = 0; + cache->n_items = 0; if (brw->curbe.last_buf) { _mesa_free(brw->curbe.last_buf); @@ -483,25 +517,46 @@ brw_clear_cache( struct brw_context *brw ) brw->state.dirty.cache |= ~0; } -void brw_state_cache_check_size( struct brw_context *brw ) + +void +brw_state_cache_check_size(struct brw_context *brw) { + if (INTEL_DEBUG & DEBUG_STATE) + _mesa_printf("%s (n_items=%d)\n", __FUNCTION__, brw->cache.n_items); + /* un-tuned guess. We've got around 20 state objects for a total of around * 32k, so 1000 of them is around 1.5MB. */ if (brw->cache.n_items > 1000) - brw_clear_cache(brw); + brw_clear_cache(brw, &brw->cache); + + if (brw->surface_cache.n_items > 1000) + brw_clear_cache(brw, &brw->surface_cache); } -void brw_destroy_cache( struct brw_context *brw ) + +static void +brw_destroy_cache(struct brw_context *brw, struct brw_cache *cache) { GLuint i; - brw_clear_cache(brw); + if (INTEL_DEBUG & DEBUG_STATE) + _mesa_printf("%s\n", __FUNCTION__); + + brw_clear_cache(brw, cache); for (i = 0; i < BRW_MAX_CACHE; i++) { - dri_bo_unreference(brw->cache.last_bo[i]); - free(brw->cache.name[i]); + dri_bo_unreference(cache->last_bo[i]); + free(cache->name[i]); } - free(brw->cache.items); - brw->cache.items = NULL; - brw->cache.size = 0; + free(cache->items); + cache->items = NULL; + cache->size = 0; +} + + +void +brw_destroy_caches(struct brw_context *brw) +{ + brw_destroy_cache(brw, &brw->cache); + brw_destroy_cache(brw, &brw->surface_cache); } diff --git a/src/mesa/drivers/dri/i965/brw_state_dump.c b/src/mesa/drivers/dri/i965/brw_state_dump.c index 5d332d010c..a713262269 100644 --- a/src/mesa/drivers/dri/i965/brw_state_dump.c +++ b/src/mesa/drivers/dri/i965/brw_state_dump.c @@ -84,6 +84,19 @@ get_965_surfacetype(unsigned int surfacetype) } } +static const char * +get_965_surface_format(unsigned int surface_format) +{ + switch (surface_format) { + case 0x000: return "r32g32b32a32_float"; + case 0x0c1: return "b8g8r8a8_unorm"; + case 0x100: return "b5g6r5_unorm"; + case 0x102: return "b5g5r5a1_unorm"; + case 0x104: return "b4g4r4a4_unorm"; + default: return "unknown"; + } +} + static void dump_wm_surface_state(struct brw_context *brw) { int i; @@ -95,7 +108,7 @@ static void dump_wm_surface_state(struct brw_context *brw) char name[20]; if (surf_bo == NULL) { - fprintf(stderr, "WM SS%d: NULL\n", i); + fprintf(stderr, " WM SS%d: NULL\n", i); continue; } dri_bo_map(surf_bo, GL_FALSE); @@ -103,8 +116,9 @@ static void dump_wm_surface_state(struct brw_context *brw) surf = (struct brw_surface_state *)(surf_bo->virtual); sprintf(name, "WM SS%d", i); - state_out(name, surf, surfoff, 0, "%s\n", - get_965_surfacetype(surf->ss0.surface_type)); + state_out(name, surf, surfoff, 0, "%s %s\n", + get_965_surfacetype(surf->ss0.surface_type), + get_965_surface_format(surf->ss0.surface_format)); state_out(name, surf, surfoff, 1, "offset\n"); state_out(name, surf, surfoff, 2, "%dx%d size, %d mips\n", surf->ss2.width + 1, surf->ss2.height + 1, surf->ss2.mip_count); diff --git a/src/mesa/drivers/dri/i965/brw_state_upload.c b/src/mesa/drivers/dri/i965/brw_state_upload.c index 5de1450e61..c6dfea4743 100644 --- a/src/mesa/drivers/dri/i965/brw_state_upload.c +++ b/src/mesa/drivers/dri/i965/brw_state_upload.c @@ -59,11 +59,12 @@ const struct brw_tracked_state *atoms[] = &brw_curbe_offsets, &brw_recalculate_urb_fence, - &brw_cc_vp, &brw_cc_unit, - &brw_wm_surfaces, /* must do before samplers */ + &brw_vs_surfaces, /* must do before unit */ + &brw_wm_constant_surface, /* must do before wm surfaces/bind bo */ + &brw_wm_surfaces, /* must do before samplers and unit */ &brw_wm_samplers, &brw_wm_unit, @@ -88,54 +89,26 @@ const struct brw_tracked_state *atoms[] = &brw_line_stipple, &brw_aa_line_parameters, - /* Ordering of the commands below is documented as fixed. - */ -#if 0 - &brw_pipelined_state_pointers, - &brw_urb_fence, - &brw_constant_buffer_state, -#else + &brw_psp_urb_cbs, -#endif &brw_drawing_rect, &brw_indices, &brw_vertices, - NULL, /* brw_constant_buffer */ + &brw_constant_buffer }; void brw_init_state( struct brw_context *brw ) { - GLuint i; - - brw_init_cache(brw); - - brw->state.atoms = _mesa_malloc(sizeof(atoms)); - brw->state.nr_atoms = sizeof(atoms)/sizeof(*atoms); - _mesa_memcpy(brw->state.atoms, atoms, sizeof(atoms)); - - /* Patch in a pointer to the dynamic state atom: - */ - for (i = 0; i < brw->state.nr_atoms; i++) - if (brw->state.atoms[i] == NULL) - brw->state.atoms[i] = &brw->curbe.tracked_state; - - _mesa_memcpy(&brw->curbe.tracked_state, - &brw_constant_buffer, - sizeof(brw_constant_buffer)); + brw_init_caches(brw); } void brw_destroy_state( struct brw_context *brw ) { - if (brw->state.atoms) { - _mesa_free(brw->state.atoms); - brw->state.atoms = NULL; - } - - brw_destroy_cache(brw); + brw_destroy_caches(brw); brw_destroy_batch_cache(brw); } @@ -218,6 +191,7 @@ static struct dirty_bit_map mesa_bits[] = { DEFINE_BIT(_NEW_MULTISAMPLE), DEFINE_BIT(_NEW_TRACK_MATRIX), DEFINE_BIT(_NEW_PROGRAM), + DEFINE_BIT(_NEW_PROGRAM_CONSTANTS), {0, 0, 0} }; @@ -336,7 +310,7 @@ void brw_validate_state( struct brw_context *brw ) /* do prepare stage for all atoms */ for (i = 0; i < Elements(atoms); i++) { - const struct brw_tracked_state *atom = brw->state.atoms[i]; + const struct brw_tracked_state *atom = atoms[i]; if (brw->intel.Fallback) break; @@ -367,8 +341,8 @@ void brw_upload_state(struct brw_context *brw) _mesa_memset(&examined, 0, sizeof(examined)); prev = *state; - for (i = 0; i < brw->state.nr_atoms; i++) { - const struct brw_tracked_state *atom = brw->state.atoms[i]; + for (i = 0; i < Elements(atoms); i++) { + const struct brw_tracked_state *atom = atoms[i]; struct brw_state_flags generated; assert(atom->dirty.mesa || @@ -397,7 +371,7 @@ void brw_upload_state(struct brw_context *brw) } else { for (i = 0; i < Elements(atoms); i++) { - const struct brw_tracked_state *atom = brw->state.atoms[i]; + const struct brw_tracked_state *atom = atoms[i]; if (brw->intel.Fallback) break; diff --git a/src/mesa/drivers/dri/i965/brw_tex_layout.c b/src/mesa/drivers/dri/i965/brw_tex_layout.c index 51a617fcb4..5c5455813a 100644 --- a/src/mesa/drivers/dri/i965/brw_tex_layout.c +++ b/src/mesa/drivers/dri/i965/brw_tex_layout.c @@ -28,7 +28,6 @@ * Authors: * Keith Whitwell <keith@tungstengraphics.com> */ - /* Code to layout images in a mipmap tree for i965. */ @@ -40,14 +39,15 @@ #define FILE_DEBUG_FLAG DEBUG_MIPTREE -GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_tree *mt ) +GLboolean brw_miptree_layout(struct intel_context *intel, + struct intel_mipmap_tree *mt, + uint32_t tiling) { - /* XXX: these vary depending on image format: - */ -/* GLint align_w = 4; */ + /* XXX: these vary depending on image format: */ + /* GLint align_w = 4; */ switch (mt->target) { - case GL_TEXTURE_CUBE_MAP: + case GL_TEXTURE_CUBE_MAP: case GL_TEXTURE_3D: { GLuint width = mt->width0; GLuint height = mt->height0; @@ -59,25 +59,25 @@ GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_t GLuint align_w = 4; mt->total_height = 0; - + if (mt->compressed) { align_w = intel_compressed_alignment(mt->internal_format); mt->pitch = ALIGN(width, align_w); pack_y_pitch = (height + 3) / 4; } else { - mt->pitch = intel_miptree_pitch_align (intel, mt, mt->width0); - pack_y_pitch = ALIGN(mt->height0, align_h); + mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->width0); + pack_y_pitch = ALIGN(mt->height0, align_h); } pack_x_pitch = mt->pitch; pack_x_nr = 1; - for ( level = mt->first_level ; level <= mt->last_level ; level++ ) { + for (level = mt->first_level ; level <= mt->last_level ; level++) { GLuint nr_images = mt->target == GL_TEXTURE_3D ? depth : 6; GLint x = 0; GLint y = 0; GLint q, j; - + intel_miptree_set_level_info(mt, level, nr_images, 0, mt->total_height, width, height, depth); @@ -89,7 +89,7 @@ GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_t } x = 0; - y += pack_y_pitch; + y += pack_y_pitch; } @@ -98,40 +98,40 @@ GLboolean brw_miptree_layout( struct intel_context *intel, struct intel_mipmap_t height = minify(height); depth = minify(depth); - if (mt->compressed) { - pack_y_pitch = (height + 3) / 4; - - if (pack_x_pitch > ALIGN(width, align_w)) { - pack_x_pitch = ALIGN(width, align_w); - pack_x_nr <<= 1; - } - } else { - if (pack_x_pitch > 4) { - pack_x_pitch >>= 1; - pack_x_nr <<= 1; - assert(pack_x_pitch * pack_x_nr <= mt->pitch); - } - - if (pack_y_pitch > 2) { - pack_y_pitch >>= 1; - pack_y_pitch = ALIGN(pack_y_pitch, align_h); - } - } + if (mt->compressed) { + pack_y_pitch = (height + 3) / 4; + + if (pack_x_pitch > ALIGN(width, align_w)) { + pack_x_pitch = ALIGN(width, align_w); + pack_x_nr <<= 1; + } + } else { + if (pack_x_pitch > 4) { + pack_x_pitch >>= 1; + pack_x_nr <<= 1; + assert(pack_x_pitch * pack_x_nr <= mt->pitch); + } + + if (pack_y_pitch > 2) { + pack_y_pitch >>= 1; + pack_y_pitch = ALIGN(pack_y_pitch, align_h); + } + } } break; } default: - i945_miptree_layout_2d(intel, mt); + i945_miptree_layout_2d(intel, mt, tiling); break; } - DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, - mt->pitch, + DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__, + mt->pitch, mt->total_height, mt->cpp, mt->pitch * mt->total_height * mt->cpp ); - + return GL_TRUE; } diff --git a/src/mesa/drivers/dri/i965/brw_vs.h b/src/mesa/drivers/dri/i965/brw_vs.h index 99d0e93722..1e4f66091e 100644 --- a/src/mesa/drivers/dri/i965/brw_vs.h +++ b/src/mesa/drivers/dri/i965/brw_vs.h @@ -75,6 +75,11 @@ struct brw_vs_compile { struct brw_reg userplane[6]; + /** we may need up to 3 constants per instruction (if use_const_buffer) */ + struct { + GLint index; + struct brw_reg reg; + } current_const[3]; }; void brw_vs_emit( struct brw_vs_compile *c ); diff --git a/src/mesa/drivers/dri/i965/brw_vs_constval.c b/src/mesa/drivers/dri/i965/brw_vs_constval.c index d29eb17f8c..2637344b48 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_constval.c +++ b/src/mesa/drivers/dri/i965/brw_vs_constval.c @@ -96,7 +96,7 @@ static GLubyte get_active( struct tracker *t, struct prog_src_register src ) { GLuint i; - GLubyte active = src.NegateBase; /* NOTE! */ + GLubyte active = src.Negate; /* NOTE! */ if (src.RelAddr) return 0xf; diff --git a/src/mesa/drivers/dri/i965/brw_vs_emit.c b/src/mesa/drivers/dri/i965/brw_vs_emit.c index 0d6c6ab9a8..d7f75e3685 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_emit.c +++ b/src/mesa/drivers/dri/i965/brw_vs_emit.c @@ -38,14 +38,49 @@ #include "brw_vs.h" +static struct brw_reg get_tmp( struct brw_vs_compile *c ) +{ + struct brw_reg tmp = brw_vec8_grf(c->last_tmp, 0); + + if (++c->last_tmp > c->prog_data.total_grf) + c->prog_data.total_grf = c->last_tmp; + + return tmp; +} -/* Do things as simply as possible. Allocate and populate all regs +static void release_tmp( struct brw_vs_compile *c, struct brw_reg tmp ) +{ + if (tmp.nr == c->last_tmp-1) + c->last_tmp--; +} + +static void release_tmps( struct brw_vs_compile *c ) +{ + c->last_tmp = c->first_tmp; +} + + +/** + * Preallocate GRF register before code emit. + * Do things as simply as possible. Allocate and populate all regs * ahead of time. */ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) { GLuint i, reg = 0, mrf; - GLuint nr_params; + + /* Determine whether to use a real constant buffer or use a block + * of GRF registers for constants. The later is faster but only + * works if everything fits in the GRF. + * XXX this heuristic/check may need some fine tuning... + */ + if (c->vp->program.Base.Parameters->NumParameters + + c->vp->program.Base.NumTemporaries + 20 > BRW_MAX_GRF) + c->vp->use_const_buffer = GL_TRUE; + else + c->vp->use_const_buffer = GL_FALSE; + + /*printf("use_const_buffer = %d\n", c->vp->use_const_buffer);*/ /* r0 -- reserved as usual */ @@ -66,13 +101,22 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) /* Vertex program parameters from curbe: */ - nr_params = c->vp->program.Base.Parameters->NumParameters; - for (i = 0; i < nr_params; i++) { - c->regs[PROGRAM_STATE_VAR][i] = stride( brw_vec4_grf(reg+i/2, (i%2) * 4), 0, 4, 1); - } - reg += (nr_params + 1) / 2; + if (c->vp->use_const_buffer) { + /* get constants from a real constant buffer */ + c->prog_data.curb_read_length = 0; + c->prog_data.nr_params = 4; /* XXX 0 causes a bug elsewhere... */ + } + else { + /* use a section of the GRF for constants */ + GLuint nr_params = c->vp->program.Base.Parameters->NumParameters; + for (i = 0; i < nr_params; i++) { + c->regs[PROGRAM_STATE_VAR][i] = stride( brw_vec4_grf(reg+i/2, (i%2) * 4), 0, 4, 1); + } + reg += (nr_params + 1) / 2; + c->prog_data.curb_read_length = reg - 1; - c->prog_data.curb_read_length = reg - 1; + c->prog_data.nr_params = nr_params * 4; + } /* Allocate input regs: */ @@ -133,6 +177,14 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) reg++; } + if (c->vp->use_const_buffer) { + for (i = 0; i < 3; i++) { + c->current_const[i].index = -1; + c->current_const[i].reg = brw_vec8_grf(reg, 0); + reg++; + } + } + for (i = 0; i < 128; i++) { if (c->output_regs[i].used_in_src) { c->output_regs[i].reg = brw_vec8_grf(reg, 0); @@ -165,28 +217,6 @@ static void brw_vs_alloc_regs( struct brw_vs_compile *c ) } -static struct brw_reg get_tmp( struct brw_vs_compile *c ) -{ - struct brw_reg tmp = brw_vec8_grf(c->last_tmp, 0); - - if (++c->last_tmp > c->prog_data.total_grf) - c->prog_data.total_grf = c->last_tmp; - - return tmp; -} - -static void release_tmp( struct brw_vs_compile *c, struct brw_reg tmp ) -{ - if (tmp.nr == c->last_tmp-1) - c->last_tmp--; -} - -static void release_tmps( struct brw_vs_compile *c ) -{ - c->last_tmp = c->first_tmp; -} - - /** * If an instruction uses a temp reg both as a src and the dest, we * sometimes need to allocate an intermediate temporary. @@ -633,6 +663,8 @@ static void emit_lit_noalias( struct brw_vs_compile *c, } brw_ENDIF(p, if_insn); + + release_tmp(c, tmp); } static void emit_lrp_noalias(struct brw_vs_compile *c, @@ -673,13 +705,84 @@ static void emit_nrm( struct brw_vs_compile *c, } +static struct brw_reg +get_constant(struct brw_vs_compile *c, + const struct prog_instruction *inst, + GLuint argIndex) +{ + const struct prog_src_register *src = &inst->SrcReg[argIndex]; + struct brw_compile *p = &c->func; + struct brw_reg const_reg; + struct brw_reg const2_reg; + const GLboolean relAddr = src->RelAddr; + + assert(argIndex < 3); + + if (c->current_const[argIndex].index != src->Index || relAddr) { + struct brw_reg addrReg = c->regs[PROGRAM_ADDRESS][0]; + + c->current_const[argIndex].index = src->Index; + +#if 0 + printf(" fetch const[%d] for arg %d into reg %d\n", + src->Index, argIndex, c->current_const[argIndex].reg.nr); +#endif + /* need to fetch the constant now */ + brw_dp_READ_4_vs(p, + c->current_const[argIndex].reg,/* writeback dest */ + 0, /* oword */ + relAddr, /* relative indexing? */ + addrReg, /* address register */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_VERT_CONST_BUFFER /* binding table index */ + ); + + if (relAddr) { + /* second read */ + const2_reg = get_tmp(c); + + /* use upper half of address reg for second read */ + addrReg = stride(addrReg, 0, 4, 0); + addrReg.subnr = 16; + + brw_dp_READ_4_vs(p, + const2_reg, /* writeback dest */ + 1, /* oword */ + relAddr, /* relative indexing? */ + addrReg, /* address register */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_VERT_CONST_BUFFER + ); + } + } + + const_reg = c->current_const[argIndex].reg; + + if (relAddr) { + /* merge the two Owords into the constant register */ + /* const_reg[7..4] = const2_reg[7..4] */ + brw_MOV(p, + suboffset(stride(const_reg, 0, 4, 1), 4), + suboffset(stride(const2_reg, 0, 4, 1), 4)); + release_tmp(c, const2_reg); + } + else { + /* replicate lower four floats into upper half (to get XYZWXYZW) */ + const_reg = stride(const_reg, 0, 4, 0); + const_reg.subnr = 0; + } + + return const_reg; +} + + + /* TODO: relative addressing! */ static struct brw_reg get_reg( struct brw_vs_compile *c, gl_register_file file, GLuint index ) { - switch (file) { case PROGRAM_TEMPORARY: case PROGRAM_INPUT: @@ -708,13 +811,17 @@ static struct brw_reg get_reg( struct brw_vs_compile *c, } +/** + * Indirect addressing: get reg[[arg] + offset]. + */ static struct brw_reg deref( struct brw_vs_compile *c, struct brw_reg arg, GLint offset) { struct brw_compile *p = &c->func; struct brw_reg tmp = vec4(get_tmp(c)); - struct brw_reg vp_address = retype(vec1(get_reg(c, PROGRAM_ADDRESS, 0)), BRW_REGISTER_TYPE_UW); + struct brw_reg addr_reg = c->regs[PROGRAM_ADDRESS][0]; + struct brw_reg vp_address = retype(vec1(addr_reg), BRW_REGISTER_TYPE_UW); GLuint byte_offset = arg.nr * 32 + arg.subnr + offset * 16; struct brw_reg indirect = brw_vec4_indirect(0,0); @@ -735,10 +842,67 @@ static struct brw_reg deref( struct brw_vs_compile *c, brw_pop_insn_state(p); } + /* NOTE: tmp not released */ return vec8(tmp); } +/** + * Get brw reg corresponding to the instruction's [argIndex] src reg. + * TODO: relative addressing! + */ +static struct brw_reg +get_src_reg( struct brw_vs_compile *c, + const struct prog_instruction *inst, + GLuint argIndex ) +{ + const GLuint file = inst->SrcReg[argIndex].File; + const GLint index = inst->SrcReg[argIndex].Index; + const GLboolean relAddr = inst->SrcReg[argIndex].RelAddr; + + switch (file) { + case PROGRAM_TEMPORARY: + case PROGRAM_INPUT: + case PROGRAM_OUTPUT: + if (relAddr) { + return deref(c, c->regs[file][0], index); + } + else { + assert(c->regs[file][index].nr != 0); + return c->regs[file][index]; + } + + case PROGRAM_STATE_VAR: + case PROGRAM_CONSTANT: + case PROGRAM_UNIFORM: + if (c->vp->use_const_buffer) { + return get_constant(c, inst, argIndex); + } + else if (relAddr) { + return deref(c, c->regs[PROGRAM_STATE_VAR][0], index); + } + else { + assert(c->regs[PROGRAM_STATE_VAR][index].nr != 0); + return c->regs[PROGRAM_STATE_VAR][index]; + } + case PROGRAM_ADDRESS: + assert(index == 0); + return c->regs[file][index]; + + case PROGRAM_UNDEFINED: + /* this is a normal case since we loop over all three src args */ + return brw_null_reg(); + + case PROGRAM_LOCAL_PARAM: + case PROGRAM_ENV_PARAM: + case PROGRAM_WRITE_ONLY: + default: + assert(0); + return brw_null_reg(); + } +} + + static void emit_arl( struct brw_vs_compile *c, struct brw_reg dst, struct brw_reg arg0 ) @@ -750,30 +914,31 @@ static void emit_arl( struct brw_vs_compile *c, if (need_tmp) tmp = get_tmp(c); - brw_RNDD(p, tmp, arg0); - brw_MUL(p, dst, tmp, brw_imm_d(16)); + brw_RNDD(p, tmp, arg0); /* tmp = round(arg0) */ + brw_MUL(p, dst, tmp, brw_imm_d(16)); /* dst = tmp * 16 */ if (need_tmp) release_tmp(c, tmp); } -/* Will return mangled results for SWZ op. The emit_swz() function +/** + * Return the brw reg for the given instruction's src argument. + * Will return mangled results for SWZ op. The emit_swz() function * ignores this result and recalculates taking extended swizzles into * account. */ static struct brw_reg get_arg( struct brw_vs_compile *c, - struct prog_src_register *src ) + const struct prog_instruction *inst, + GLuint argIndex ) { + const struct prog_src_register *src = &inst->SrcReg[argIndex]; struct brw_reg reg; if (src->File == PROGRAM_UNDEFINED) return brw_null_reg(); - if (src->RelAddr) - reg = deref(c, c->regs[PROGRAM_STATE_VAR][0], src->Index); - else - reg = get_reg(c, src->File, src->Index); + reg = get_src_reg(c, inst, argIndex); /* Convert 3-bit swizzle to 2-bit. */ @@ -784,16 +949,38 @@ static struct brw_reg get_arg( struct brw_vs_compile *c, /* Note this is ok for non-swizzle instructions: */ - reg.negate = src->NegateBase ? 1 : 0; + reg.negate = src->Negate ? 1 : 0; return reg; } +/** + * Get brw register for the given program dest register. + */ static struct brw_reg get_dst( struct brw_vs_compile *c, struct prog_dst_register dst ) { - struct brw_reg reg = get_reg(c, dst.File, dst.Index); + struct brw_reg reg; + + switch (dst.File) { + case PROGRAM_TEMPORARY: + case PROGRAM_OUTPUT: + assert(c->regs[dst.File][dst.Index].nr != 0); + reg = c->regs[dst.File][dst.Index]; + break; + case PROGRAM_ADDRESS: + assert(dst.Index == 0); + reg = c->regs[dst.File][dst.Index]; + break; + case PROGRAM_UNDEFINED: + /* we may hit this for OPCODE_END, OPCODE_KIL, etc */ + reg = brw_null_reg(); + break; + default: + assert(0); + reg = brw_null_reg(); + } reg.dw1.bits.writemask = dst.WriteMask; @@ -803,14 +990,16 @@ static struct brw_reg get_dst( struct brw_vs_compile *c, static void emit_swz( struct brw_vs_compile *c, struct brw_reg dst, - struct prog_src_register src ) + const struct prog_instruction *inst) { + const GLuint argIndex = 0; + const struct prog_src_register src = inst->SrcReg[argIndex]; struct brw_compile *p = &c->func; GLuint zeros_mask = 0; GLuint ones_mask = 0; GLuint src_mask = 0; GLubyte src_swz[4]; - GLboolean need_tmp = (src.NegateBase && + GLboolean need_tmp = (src.Negate && dst.file != BRW_GENERAL_REGISTER_FILE); struct brw_reg tmp = dst; GLuint i; @@ -844,10 +1033,7 @@ static void emit_swz( struct brw_vs_compile *c, if (src_mask) { struct brw_reg arg0; - if (src.RelAddr) - arg0 = deref(c, c->regs[PROGRAM_STATE_VAR][0], src.Index); - else - arg0 = get_reg(c, src.File, src.Index); + arg0 = get_src_reg(c, inst, argIndex); arg0 = brw_swizzle(arg0, src_swz[0], src_swz[1], @@ -862,8 +1048,8 @@ static void emit_swz( struct brw_vs_compile *c, if (ones_mask) brw_MOV(p, brw_writemask(tmp, ones_mask), brw_imm_f(1)); - if (src.NegateBase) - brw_MOV(p, brw_writemask(tmp, src.NegateBase), negate(tmp)); + if (src.Negate) + brw_MOV(p, brw_writemask(tmp, src.Negate), negate(tmp)); if (need_tmp) { brw_MOV(p, dst, tmp); @@ -1039,21 +1225,26 @@ void brw_vs_emit(struct brw_vs_compile *c ) for (insn = 0; insn < nr_insns; insn++) { - struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn]; + const struct prog_instruction *inst = &c->vp->program.Base.Instructions[insn]; struct brw_reg args[3], dst; GLuint i; +#if 0 + printf("%d: ", insn); + _mesa_print_instruction(inst); +#endif + /* Get argument regs. SWZ is special and does this itself. */ if (inst->Opcode != OPCODE_SWZ) for (i = 0; i < 3; i++) { - struct prog_src_register *src = &inst->SrcReg[i]; + const struct prog_src_register *src = &inst->SrcReg[i]; index = src->Index; file = src->File; if (file == PROGRAM_OUTPUT && c->output_regs[index].used_in_src) args[i] = c->output_regs[index].reg; else - args[i] = get_arg(c, src); + args[i] = get_arg(c, inst, i); } /* Get dest regs. Note that it is possible for a reg to be both @@ -1181,7 +1372,7 @@ void brw_vs_emit(struct brw_vs_compile *c ) /* The args[0] value can't be used here as it won't have * correctly encoded the full swizzle: */ - emit_swz(c, dst, inst->SrcReg[0] ); + emit_swz(c, dst, inst); break; case OPCODE_TRUNC: /* round toward zero */ diff --git a/src/mesa/drivers/dri/i965/brw_vs_state.c b/src/mesa/drivers/dri/i965/brw_vs_state.c index 1a63766ea1..3d29538843 100644 --- a/src/mesa/drivers/dri/i965/brw_vs_state.c +++ b/src/mesa/drivers/dri/i965/brw_vs_state.c @@ -44,6 +44,8 @@ struct brw_vs_unit_key { unsigned int curbe_offset; unsigned int nr_urb_entries, urb_size; + + unsigned int nr_surfaces; }; static void @@ -62,6 +64,9 @@ vs_unit_populate_key(struct brw_context *brw, struct brw_vs_unit_key *key) key->nr_urb_entries = brw->urb.nr_vs_entries; key->urb_size = brw->urb.vsize; + /* BRW_NEW_NR_VS_SURFACES */ + key->nr_surfaces = brw->vs.nr_surfaces; + /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM */ if (ctx->Transform.ClipPlanesEnabled) { /* Note that we read in the userclip planes as well, hence @@ -92,6 +97,8 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key) * brw_urb_WRITE() results. */ vs.thread1.single_program_flow = 0; + vs.thread1.binding_table_entry_count = key->nr_surfaces; + vs.thread3.urb_entry_read_length = key->urb_entry_read_length; vs.thread3.const_urb_entry_read_length = key->curb_entry_read_length; vs.thread3.dispatch_grf_start_reg = 1; @@ -158,6 +165,7 @@ const struct brw_tracked_state brw_vs_unit = { .dirty = { .mesa = _NEW_TRANSFORM, .brw = (BRW_NEW_CURBE_OFFSETS | + BRW_NEW_NR_VS_SURFACES | BRW_NEW_URB_FENCE), .cache = CACHE_NEW_VS_PROG }, diff --git a/src/mesa/drivers/dri/i965/brw_vs_surface_state.c b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c new file mode 100644 index 0000000000..89f47522a1 --- /dev/null +++ b/src/mesa/drivers/dri/i965/brw_vs_surface_state.c @@ -0,0 +1,226 @@ +/* + Copyright (C) Intel Corp. 2006. All Rights Reserved. + Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to + develop this 3D driver. + + Permission is hereby granted, free of charge, to any person obtaining + a copy of this software and associated documentation files (the + "Software"), to deal in the Software without restriction, including + without limitation the rights to use, copy, modify, merge, publish, + distribute, sublicense, and/or sell copies of the Software, and to + permit persons to whom the Software is furnished to do so, subject to + the following conditions: + + The above copyright notice and this permission notice (including the + next paragraph) shall be included in all copies or substantial + portions of the Software. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE + LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION + OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION + WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + + **********************************************************************/ + /* + * Authors: + * Keith Whitwell <keith@tungstengraphics.com> + */ + +#include "main/mtypes.h" +#include "main/texformat.h" +#include "main/texstore.h" +#include "shader/prog_parameter.h" + +#include "brw_context.h" +#include "brw_state.h" +#include "brw_defines.h" + +/* Creates a new VS constant buffer reflecting the current VS program's + * constants, if needed by the VS program. + * + * Otherwise, constants go through the CURBEs using the brw_constant_buffer + * state atom. + */ +static drm_intel_bo * +brw_vs_update_constant_buffer(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + struct brw_vertex_program *vp = + (struct brw_vertex_program *) brw->vertex_program; + const struct gl_program_parameter_list *params = vp->program.Base.Parameters; + const int size = params->NumParameters * 4 * sizeof(GLfloat); + drm_intel_bo *const_buffer; + + /* BRW_NEW_VERTEX_PROGRAM */ + if (!vp->use_const_buffer) + return NULL; + + const_buffer = drm_intel_bo_alloc(intel->bufmgr, "vp_const_buffer", + size, 64); + + /* _NEW_PROGRAM_CONSTANTS */ + dri_bo_subdata(const_buffer, 0, size, params->ParameterValues); + + return const_buffer; +} + +/** + * Update the surface state for a VS constant buffer. + * + * Sets brw->vs.surf_bo[surf] and brw->vp->const_buffer. + */ +static void +brw_update_vs_constant_surface( GLcontext *ctx, + GLuint surf) +{ + struct brw_context *brw = brw_context(ctx); + struct brw_surface_key key; + struct brw_vertex_program *vp = + (struct brw_vertex_program *) brw->vertex_program; + const struct gl_program_parameter_list *params = vp->program.Base.Parameters; + + assert(surf == 0); + + /* If we're in this state update atom, we need to update VS constants, so + * free the old buffer and create a new one for the new contents. + */ + dri_bo_unreference(vp->const_buffer); + vp->const_buffer = brw_vs_update_constant_buffer(brw); + + /* If there's no constant buffer, then no surface BO is needed to point at + * it. + */ + if (vp->const_buffer == 0) { + drm_intel_bo_unreference(brw->vs.surf_bo[surf]); + brw->vs.surf_bo[surf] = NULL; + return; + } + + memset(&key, 0, sizeof(key)); + + key.format = MESA_FORMAT_RGBA_FLOAT32; + key.internal_format = GL_RGBA; + key.bo = vp->const_buffer; + key.depthmode = GL_NONE; + key.pitch = params->NumParameters; + key.width = params->NumParameters; + key.height = 1; + key.depth = 1; + key.cpp = 16; + + /* + printf("%s:\n", __FUNCTION__); + printf(" width %d height %d depth %d cpp %d pitch %d\n", + key.width, key.height, key.depth, key.cpp, key.pitch); + */ + + drm_intel_bo_unreference(brw->vs.surf_bo[surf]); + brw->vs.surf_bo[surf] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, key.bo ? 1 : 0, + NULL); + if (brw->vs.surf_bo[surf] == NULL) { + brw->vs.surf_bo[surf] = brw_create_constant_surface(brw, &key); + } +} + + +/** + * Constructs the binding table for the VS surface state. + */ +static dri_bo * +brw_vs_get_binding_table(struct brw_context *brw) +{ + dri_bo *bind_bo; + + bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->vs.surf_bo, BRW_VS_MAX_SURF, + NULL); + + if (bind_bo == NULL) { + GLuint data_size = BRW_VS_MAX_SURF * sizeof(GLuint); + uint32_t *data = malloc(data_size); + int i; + + for (i = 0; i < BRW_VS_MAX_SURF; i++) + if (brw->vs.surf_bo[i]) + data[i] = brw->vs.surf_bo[i]->offset; + else + data[i] = 0; + + bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, + NULL, 0, + brw->vs.surf_bo, BRW_VS_MAX_SURF, + data, data_size, + NULL, NULL); + + /* Emit binding table relocations to surface state */ + for (i = 0; i < BRW_VS_MAX_SURF; i++) { + if (brw->vs.surf_bo[i] != NULL) { + /* The presumed offsets were set in the data values for + * brw_upload_cache. + */ + drm_intel_bo_emit_reloc(bind_bo, i * 4, + brw->vs.surf_bo[i], 0, + I915_GEM_DOMAIN_INSTRUCTION, 0); + } + } + + free(data); + } + + return bind_bo; +} + +/** + * Vertex shader surfaces (constant buffer). + * + * This consumes the state updates for the constant buffer needing + * to be updated, and produces BRW_NEW_NR_VS_SURFACES for the VS unit and + * CACHE_NEW_SURF_BIND for the binding table upload. + */ +static void prepare_vs_surfaces(struct brw_context *brw ) +{ + GLcontext *ctx = &brw->intel.ctx; + int i; + int nr_surfaces = 0; + + brw_update_vs_constant_surface(ctx, SURF_INDEX_VERT_CONST_BUFFER); + + for (i = 0; i < BRW_VS_MAX_SURF; i++) { + if (brw->vs.surf_bo[i] != NULL) { + nr_surfaces = i + 1; + } + } + + if (brw->vs.nr_surfaces != nr_surfaces) { + brw->state.dirty.brw |= BRW_NEW_NR_VS_SURFACES; + brw->vs.nr_surfaces = nr_surfaces; + } + + /* Note that we don't end up updating the bind_bo if we don't have a + * surface to be pointing at. This should be relatively harmless, as it + * just slightly increases our working set size. + */ + if (brw->vs.nr_surfaces != 0) { + dri_bo_unreference(brw->vs.bind_bo); + brw->vs.bind_bo = brw_vs_get_binding_table(brw); + } +} + +const struct brw_tracked_state brw_vs_surfaces = { + .dirty = { + .mesa = (_NEW_PROGRAM_CONSTANTS), + .brw = (BRW_NEW_VERTEX_PROGRAM), + .cache = 0 + }, + .prepare = prepare_vs_surfaces, +}; + + + diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c index 960bbb311e..ba03afd6c1 100644 --- a/src/mesa/drivers/dri/i965/brw_vtbl.c +++ b/src/mesa/drivers/dri/i965/brw_vtbl.c @@ -79,6 +79,7 @@ static void brw_destroy_context( struct intel_context *intel ) dri_bo_release(&brw->curbe.curbe_bo); dri_bo_release(&brw->vs.prog_bo); dri_bo_release(&brw->vs.state_bo); + dri_bo_release(&brw->vs.bind_bo); dri_bo_release(&brw->gs.prog_bo); dri_bo_release(&brw->gs.state_bo); dri_bo_release(&brw->clip.prog_bo); diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c index 90d74c2885..3e476fd3be 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.c +++ b/src/mesa/drivers/dri/i965/brw_wm.c @@ -146,6 +146,13 @@ static void do_wm_prog( struct brw_context *brw, if (c == NULL) { brw->wm.compile_data = calloc(1, sizeof(*brw->wm.compile_data)); c = brw->wm.compile_data; + if (c == NULL) { + /* Ouch - big out of memory problem. Can't continue + * without triggering a segfault, no way to signal, + * so just return. + */ + return; + } } else { memset(c, 0, sizeof(*brw->wm.compile_data)); } @@ -312,6 +319,9 @@ static void brw_wm_populate_key( struct brw_context *brw, key->drawable_height = brw->intel.driDrawable->h; } + /* CACHE_NEW_VS_PROG */ + key->vp_outputs_written = brw->vs.prog_data->outputs_written & DO_SETUP_BITS; + /* The unique fragment program ID */ key->program_string_id = fp->id; } @@ -350,7 +360,7 @@ const struct brw_tracked_state brw_wm_prog = { .brw = (BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_WM_INPUT_DIMENSIONS | BRW_NEW_REDUCED_PRIMITIVE), - .cache = 0 + .cache = CACHE_NEW_VS_PROG, }, .prepare = brw_prepare_wm_prog }; diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index d0ab3bdc65..fb15c03e83 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -75,6 +75,7 @@ struct brw_wm_prog_key { GLuint program_string_id:32; GLuint origin_x, origin_y; GLuint drawable_height; + GLuint vp_outputs_written; }; @@ -240,22 +241,25 @@ struct brw_wm_compile { GLuint max_wm_grf; GLuint last_scratch; + GLuint cur_inst; /**< index of current instruction */ + + GLboolean out_of_regs; /**< ran out of GRF registers? */ + /** Mapping from Mesa registers to hardware registers */ struct { GLboolean inited; struct brw_reg reg; } wm_regs[PROGRAM_PAYLOAD+1][256][4]; + GLboolean used_grf[BRW_WM_MAX_GRF]; + GLuint first_free_grf; struct brw_reg stack; struct brw_reg emit_mask_reg; - GLuint reg_index; /**< Index of next free GRF register */ GLuint tmp_regs[BRW_WM_MAX_GRF]; GLuint tmp_index; GLuint tmp_max; GLuint subroutines[BRW_WM_MAX_SUBROUTINE]; - /** using a real constant buffer? */ - GLboolean use_const_buffer; /** we may need up to 3 constants per instruction (if use_const_buffer) */ struct { GLint index; diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index d65b1332c6..14ab9042de 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -353,6 +353,19 @@ static void emit_mad( struct brw_compile *p, } } +static void emit_trunc( struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) +{ + GLuint i; + + for (i = 0; i < 4; i++) { + if (mask & (1<<i)) { + brw_RNDZ(p, dst[i], arg0[i]); + } + } +} static void emit_lrp( struct brw_compile *p, const struct brw_reg *dst, @@ -742,7 +755,7 @@ static void emit_tex( struct brw_wm_compile *c, retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW), 1, retype(c->payload.depth[0].hw_reg, BRW_REGISTER_TYPE_UW), - inst->tex_unit + MAX_DRAW_BUFFERS, /* surface */ + SURF_INDEX_TEXTURE(inst->tex_unit), inst->tex_unit, /* sampler */ inst->writemask, (inst->tex_shadow ? @@ -791,7 +804,7 @@ static void emit_txb( struct brw_wm_compile *c, retype(vec16(dst[0]), BRW_REGISTER_TYPE_UW), 1, retype(c->payload.depth[0].hw_reg, BRW_REGISTER_TYPE_UW), - inst->tex_unit + MAX_DRAW_BUFFERS, /* surface */ + SURF_INDEX_TEXTURE(inst->tex_unit), inst->tex_unit, /* sampler */ inst->writemask, BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS, @@ -1224,6 +1237,10 @@ void brw_wm_emit( struct brw_wm_compile *c ) emit_dph(p, dst, dst_flags, args[0], args[1]); break; + case OPCODE_TRUNC: + emit_trunc(p, dst, dst_flags, args[0]); + break; + case OPCODE_LRP: emit_lrp(p, dst, dst_flags, args[0], args[1], args[2]); break; diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c index a7f5f1b9a2..1798d842c7 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_fp.c +++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c @@ -80,9 +80,8 @@ static struct prog_src_register src_reg(GLuint file, GLuint idx) reg.Index = idx; reg.Swizzle = SWIZZLE_NOOP; reg.RelAddr = 0; - reg.NegateBase = 0; + reg.Negate = NEGATE_NONE; reg.Abs = 0; - reg.NegateAbs = 0; return reg; } @@ -569,7 +568,7 @@ static void precalc_dst( struct brw_wm_compile *c, src_undef(), src_undef()); /* Avoid letting negation flag of src0 affect our 1 constant. */ - swz->SrcReg[0].NegateBase &= ~NEGATE_X; + swz->SrcReg[0].Negate &= ~NEGATE_X; } if (dst.WriteMask & WRITEMASK_W) { /* dst.w = mov src1.w @@ -604,7 +603,7 @@ static void precalc_lit( struct brw_wm_compile *c, src_undef(), src_undef()); /* Avoid letting the negation flag of src0 affect our 1 constant. */ - swz->SrcReg[0].NegateBase = 0; + swz->SrcReg[0].Negate = NEGATE_NONE; } if (dst.WriteMask & WRITEMASK_YZ) { @@ -651,7 +650,7 @@ static void precalc_tex( struct brw_wm_compile *c, src0, src_undef(), src_undef()); - out->SrcReg[0].NegateBase = 0; + out->SrcReg[0].Negate = NEGATE_NONE; out->SrcReg[0].Abs = 1; /* tmp0 = MAX(coord.X, coord.Y) */ @@ -1050,14 +1049,14 @@ void brw_wm_pass_fp( struct brw_wm_compile *c ) case OPCODE_ABS: out = emit_insn(c, inst); out->Opcode = OPCODE_MOV; - out->SrcReg[0].NegateBase = 0; + out->SrcReg[0].Negate = NEGATE_NONE; out->SrcReg[0].Abs = 1; break; case OPCODE_SUB: out = emit_insn(c, inst); out->Opcode = OPCODE_ADD; - out->SrcReg[1].NegateBase ^= 0xf; + out->SrcReg[1].Negate ^= NEGATE_XYZW; break; case OPCODE_SCS: diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 575cd45d57..0e6a2f8ef0 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -1,5 +1,7 @@ #include "main/macros.h" #include "shader/prog_parameter.h" +#include "shader/prog_print.h" +#include "shader/prog_optimize.h" #include "brw_context.h" #include "brw_eu.h" #include "brw_wm.h" @@ -21,7 +23,6 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) const struct prog_instruction *inst = &fp->Base.Instructions[i]; switch (inst->Opcode) { case OPCODE_IF: - case OPCODE_TRUNC: case OPCODE_ENDIF: case OPCODE_CAL: case OPCODE_BRK: @@ -42,6 +43,83 @@ GLboolean brw_wm_is_glsl(const struct gl_fragment_program *fp) } + +static void +reclaim_temps(struct brw_wm_compile *c); + + +/** Mark GRF register as used. */ +static void +prealloc_grf(struct brw_wm_compile *c, int r) +{ + c->used_grf[r] = GL_TRUE; +} + + +/** Mark given GRF register as not in use. */ +static void +release_grf(struct brw_wm_compile *c, int r) +{ + /*assert(c->used_grf[r]);*/ + c->used_grf[r] = GL_FALSE; + c->first_free_grf = MIN2(c->first_free_grf, r); +} + + +/** Return index of a free GRF, mark it as used. */ +static int +alloc_grf(struct brw_wm_compile *c) +{ + GLuint r; + for (r = c->first_free_grf; r < BRW_WM_MAX_GRF; r++) { + if (!c->used_grf[r]) { + c->used_grf[r] = GL_TRUE; + c->first_free_grf = r + 1; /* a guess */ + return r; + } + } + + /* no free temps, try to reclaim some */ + reclaim_temps(c); + c->first_free_grf = 0; + + /* try alloc again */ + for (r = c->first_free_grf; r < BRW_WM_MAX_GRF; r++) { + if (!c->used_grf[r]) { + c->used_grf[r] = GL_TRUE; + c->first_free_grf = r + 1; /* a guess */ + return r; + } + } + + for (r = 0; r < BRW_WM_MAX_GRF; r++) { + assert(c->used_grf[r]); + } + + /* really, no free GRF regs found */ + if (!c->out_of_regs) { + /* print warning once per compilation */ + _mesa_warning(NULL, "i965: ran out of registers for fragment program"); + c->out_of_regs = GL_TRUE; + } + + return -1; +} + + +/** Return number of GRF registers used */ +static int +num_grf_used(const struct brw_wm_compile *c) +{ + int r; + for (r = BRW_WM_MAX_GRF - 1; r >= 0; r--) + if (c->used_grf[r]) + return r + 1; + return 0; +} + + + /** * Record the mapping of a Mesa register to a hardware register. */ @@ -56,7 +134,7 @@ static void set_reg(struct brw_wm_compile *c, int file, int index, * Examine instruction's write mask to find index of first component * enabled for writing. */ -static int get_scalar_dst_index(struct prog_instruction *inst) +static int get_scalar_dst_index(const struct prog_instruction *inst) { int i; for (i = 0; i < 4; i++) @@ -68,11 +146,23 @@ static int get_scalar_dst_index(struct prog_instruction *inst) static struct brw_reg alloc_tmp(struct brw_wm_compile *c) { struct brw_reg reg; - if(c->tmp_index == c->tmp_max) - c->tmp_regs[ c->tmp_max++ ] = c->reg_index++; - + + /* if we need to allocate another temp, grow the tmp_regs[] array */ + if (c->tmp_index == c->tmp_max) { + int r = alloc_grf(c); + if (r < 0) { + /*printf("Out of temps in %s\n", __FUNCTION__);*/ + r = 50; /* XXX random register! */ + } + c->tmp_regs[ c->tmp_max++ ] = r; + } + + /* form the GRF register */ reg = brw_vec8_grf(c->tmp_regs[ c->tmp_index++ ], 0); + /*printf("alloc_temp %d\n", reg.nr);*/ + assert(reg.nr < BRW_WM_MAX_GRF); return reg; + } /** @@ -130,35 +220,29 @@ get_reg(struct brw_wm_compile *c, int file, int index, int component, return brw_null_reg(); } + assert(index < 256); + assert(component < 4); + /* see if we've already allocated a HW register for this Mesa register */ if (c->wm_regs[file][index][component].inited) { - /* yes, re-use */ - reg = c->wm_regs[file][index][component].reg; + /* yes, re-use */ + reg = c->wm_regs[file][index][component].reg; } else { /* no, allocate new register */ - reg = brw_vec8_grf(c->reg_index, 0); - } + int grf = alloc_grf(c); + /*printf("alloc grf %d for reg %d:%d.%d\n", grf, file, index, component);*/ + if (grf < 0) { + /* totally out of temps */ + grf = 51; /* XXX random register! */ + } - /* if this is a new register allocation, record it in the table */ - if (!c->wm_regs[file][index][component].inited) { - set_reg(c, file, index, component, reg); - c->reg_index++; - } + reg = brw_vec8_grf(grf, 0); + /*printf("Alloc new grf %d for %d.%d\n", reg.nr, index, component);*/ - if (c->reg_index >= BRW_WM_MAX_GRF - 12) { - /* ran out of temporary registers! */ -#if 1 - /* This is a big hack for now. - * Return bad register index, just don't hang the GPU. - */ - _mesa_fprintf(stderr, "out of regs %d\n", c->reg_index); - c->reg_index = BRW_WM_MAX_GRF - 13; -#else - return brw_null_reg(); -#endif + set_reg(c, file, index, component, reg); } - + if (neg & (1 << component)) { reg = negate(reg); } @@ -168,6 +252,46 @@ get_reg(struct brw_wm_compile *c, int file, int index, int component, } + +/** + * This is called if we run out of GRF registers. Examine the live intervals + * of temp regs in the program and free those which won't be used again. + */ +static void +reclaim_temps(struct brw_wm_compile *c) +{ + GLint intBegin[MAX_PROGRAM_TEMPS]; + GLint intEnd[MAX_PROGRAM_TEMPS]; + int index; + + /*printf("Reclaim temps:\n");*/ + + _mesa_find_temp_intervals(c->prog_instructions, c->nr_fp_insns, + intBegin, intEnd); + + for (index = 0; index < MAX_PROGRAM_TEMPS; index++) { + if (intEnd[index] != -1 && intEnd[index] < c->cur_inst) { + /* program temp[i] can be freed */ + int component; + /*printf(" temp[%d] is dead\n", index);*/ + for (component = 0; component < 4; component++) { + if (c->wm_regs[PROGRAM_TEMPORARY][index][component].inited) { + int r = c->wm_regs[PROGRAM_TEMPORARY][index][component].reg.nr; + release_grf(c, r); + /* + printf(" Reclaim temp %d, reg %d at inst %d\n", + index, r, c->cur_inst); + */ + c->wm_regs[PROGRAM_TEMPORARY][index][component].inited = GL_FALSE; + } + } + } + } +} + + + + /** * Preallocate registers. This sets up the Mesa to hardware register * mapping for certain registers, such as constants (uniforms/state vars) @@ -177,8 +301,12 @@ static void prealloc_reg(struct brw_wm_compile *c) { int i, j; struct brw_reg reg; - int nr_interp_regs = 0; + int urb_read_length = 0; GLuint inputs = FRAG_BIT_WPOS | c->fp_interp_emitted | c->fp_deriv_emitted; + GLuint reg_index = 0; + + memset(c->used_grf, GL_FALSE, sizeof(c->used_grf)); + c->first_free_grf = 0; for (i = 0; i < 4; i++) { if (i < c->key.nr_depth_regs) @@ -187,16 +315,22 @@ static void prealloc_reg(struct brw_wm_compile *c) reg = brw_vec8_grf(0, 0); set_reg(c, PROGRAM_PAYLOAD, PAYLOAD_DEPTH, i, reg); } - c->reg_index += 2 * c->key.nr_depth_regs; + reg_index += 2 * c->key.nr_depth_regs; /* constants */ { - const int nr_params = c->fp->program.Base.Parameters->NumParameters; + const GLuint nr_params = c->fp->program.Base.Parameters->NumParameters; + const GLuint nr_temps = c->fp->program.Base.NumTemporaries; /* use a real constant buffer, or just use a section of the GRF? */ - c->use_const_buffer = GL_FALSE; /* (nr_params > 8);*/ + /* XXX this heuristic may need adjustment... */ + if ((nr_params + nr_temps) * 4 + reg_index > 80) + c->fp->use_const_buffer = GL_TRUE; + else + c->fp->use_const_buffer = GL_FALSE; + /*printf("WM use_const_buffer = %d\n", c->fp->use_const_buffer);*/ - if (c->use_const_buffer) { + if (c->fp->use_const_buffer) { /* We'll use a real constant buffer and fetch constants from * it with a dataport read message. */ @@ -216,7 +350,7 @@ static void prealloc_reg(struct brw_wm_compile *c) for (i = 0; i < nr_params; i++) { /* loop over XYZW channels */ for (j = 0; j < 4; j++, index++) { - reg = brw_vec1_grf(c->reg_index + index / 8, index % 8); + reg = brw_vec1_grf(reg_index + index / 8, index % 8); /* Save pointer to parameter/constant value. * Constants will be copied in prepare_constant_buffer() */ @@ -226,42 +360,62 @@ static void prealloc_reg(struct brw_wm_compile *c) } /* number of constant regs used (each reg is float[8]) */ c->nr_creg = 2 * ((4 * nr_params + 15) / 16); - c->reg_index += c->nr_creg; + reg_index += c->nr_creg; } } /* fragment shader inputs */ - for (i = 0; i < FRAG_ATTRIB_MAX; i++) { - if (inputs & (1<<i)) { - nr_interp_regs++; - reg = brw_vec8_grf(c->reg_index, 0); - for (j = 0; j < 4; j++) - set_reg(c, PROGRAM_PAYLOAD, i, j, reg); - c->reg_index += 2; - } + for (i = 0; i < VERT_RESULT_MAX; i++) { + int fp_input; + + if (i >= VERT_RESULT_VAR0) + fp_input = i - VERT_RESULT_VAR0 + FRAG_ATTRIB_VAR0; + else if (i <= VERT_RESULT_TEX7) + fp_input = i; + else + fp_input = -1; + + if (fp_input >= 0 && inputs & (1 << fp_input)) { + urb_read_length = reg_index; + reg = brw_vec8_grf(reg_index, 0); + for (j = 0; j < 4; j++) + set_reg(c, PROGRAM_PAYLOAD, fp_input, j, reg); + } + if (c->key.vp_outputs_written & (1 << i)) { + reg_index += 2; + } } c->prog_data.first_curbe_grf = c->key.nr_depth_regs * 2; - c->prog_data.urb_read_length = nr_interp_regs * 2; + c->prog_data.urb_read_length = urb_read_length; c->prog_data.curb_read_length = c->nr_creg; - c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0); - c->reg_index++; - c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, c->reg_index, 0); - c->reg_index += 2; + c->emit_mask_reg = brw_uw1_reg(BRW_GENERAL_REGISTER_FILE, reg_index, 0); + reg_index++; + c->stack = brw_uw16_reg(BRW_GENERAL_REGISTER_FILE, reg_index, 0); + reg_index += 2; + + /* mark GRF regs [0..reg_index-1] as in-use */ + for (i = 0; i < reg_index; i++) + prealloc_grf(c, i); + + /* Don't use GRF 126, 127. Using them seems to lead to GPU lock-ups */ + prealloc_grf(c, 126); + prealloc_grf(c, 127); /* An instruction may reference up to three constants. * They'll be found in these registers. * XXX alloc these on demand! */ - if (c->use_const_buffer) { - c->current_const[0].reg = alloc_tmp(c); - c->current_const[1].reg = alloc_tmp(c); - c->current_const[2].reg = alloc_tmp(c); - } - /* - printf("USE CONST BUFFER? %d\n", c->use_const_buffer); - printf("AFTER PRE_ALLOC, reg_index = %d\n", c->reg_index); - */ + if (c->fp->use_const_buffer) { + for (i = 0; i < 3; i++) { + c->current_const[i].index = -1; + c->current_const[i].reg = brw_vec8_grf(alloc_grf(c), 0); + } + } +#if 0 + printf("USE CONST BUFFER? %d\n", c->fp->use_const_buffer); + printf("AFTER PRE_ALLOC, reg_index = %d\n", reg_index); +#endif } @@ -282,45 +436,21 @@ static void fetch_constants(struct brw_wm_compile *c, if (src->File == PROGRAM_STATE_VAR || src->File == PROGRAM_CONSTANT || src->File == PROGRAM_UNIFORM) { - if (c->current_const[i].index != src->Index) { - - c->current_const[i].index = src->Index; - /*c->current_const[i].reg = alloc_tmp(c);*/ - - /* - printf(" fetch const[%d] for arg %d into reg %d\n", - src->Index, i, c->current_const[i].reg.nr); - */ - - /* need to fetch the constant now */ - brw_dp_READ_4(p, - c->current_const[i].reg, /* writeback dest */ - 1, /* msg_reg */ - src->RelAddr, /* relative indexing? */ - 16 * src->Index, /* byte offset */ - BRW_WM_MAX_SURF - 1 /* binding table index */ - ); + c->current_const[i].index = src->Index; #if 0 - /* dependency stall */ - { - int response_length = 1; - int mark = mark_tmps( c ); - struct brw_reg src = c->current_const[i].reg; - struct brw_reg tmp = alloc_tmp(c); - - /* mov (8) r9.0<1>:f r9.0<8;8,1>:f { Align1 } - */ - brw_push_insn_state(p); - brw_set_compression_control(p, BRW_COMPRESSION_NONE); - brw_MOV(p, tmp, src); - brw_MOV(p, src, tmp); - brw_pop_insn_state(p); - - release_tmps( c, mark ); - } + printf(" fetch const[%d] for arg %d into reg %d\n", + src->Index, i, c->current_const[i].reg.nr); #endif - } + + /* need to fetch the constant now */ + brw_dp_READ_4(p, + c->current_const[i].reg, /* writeback dest */ + 1, /* msg_reg */ + src->RelAddr, /* relative indexing? */ + 16 * src->Index, /* byte offset */ + SURF_INDEX_FRAG_CONST_BUFFER/* binding table index */ + ); } } } @@ -361,18 +491,18 @@ get_src_reg_const(struct brw_wm_compile *c, const_reg = stride(const_reg, 0, 1, 0); const_reg.subnr = component * 4; - if (src->NegateBase) + if (src->Negate & (1 << component)) const_reg = negate(const_reg); if (src->Abs) const_reg = brw_abs(const_reg); - /* - printf(" form const[%d] for arg %d, comp %d, reg %d\n", +#if 0 + printf(" form const[%d].%d for arg %d, reg %d\n", c->current_const[srcRegIndex].index, - srcRegIndex, component, + srcRegIndex, const_reg.nr); - */ +#endif return const_reg; } @@ -389,7 +519,15 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c, const GLuint nr = 1; const GLuint component = GET_SWZ(src->Swizzle, channel); - if (c->use_const_buffer && + /* Extended swizzle terms */ + if (component == SWIZZLE_ZERO) { + return brw_imm_f(0.0F); + } + else if (component == SWIZZLE_ONE) { + return brw_imm_f(1.0F); + } + + if (c->fp->use_const_buffer && (src->File == PROGRAM_STATE_VAR || src->File == PROGRAM_CONSTANT || src->File == PROGRAM_UNIFORM)) { @@ -398,7 +536,7 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c, else { /* other type of source register */ return get_reg(c, src->File, src->Index, component, nr, - src->NegateBase, src->Abs); + src->Negate, src->Abs); } } @@ -423,11 +561,13 @@ static struct brw_reg get_src_reg_imm(struct brw_wm_compile *c, const GLfloat *param = c->fp->program.Base.Parameters->ParameterValues[src->Index]; GLfloat value = param[component]; - if (src->NegateBase) + if (src->Negate & (1 << channel)) value = -value; if (src->Abs) value = FABSF(value); - /*printf(" form imm reg %f\n", value);*/ +#if 0 + printf(" form immed value %f for chan %d\n", value, channel); +#endif return brw_imm_f(value); } else { @@ -501,7 +641,7 @@ static void invoke_subroutine( struct brw_wm_compile *c, } static void emit_abs( struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { int i; struct brw_compile *p = &c->func; @@ -518,7 +658,7 @@ static void emit_abs( struct brw_wm_compile *c, } static void emit_trunc( struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { int i; struct brw_compile *p = &c->func; @@ -536,7 +676,7 @@ static void emit_trunc( struct brw_wm_compile *c, } static void emit_mov( struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { int i; struct brw_compile *p = &c->func; @@ -546,7 +686,9 @@ static void emit_mov( struct brw_wm_compile *c, if (mask & (1<<i)) { struct brw_reg src, dst; dst = get_dst_reg(c, inst, i); - src = get_src_reg_imm(c, inst, 0, i); + /* XXX some moves from immediate value don't work reliably!!! */ + /*src = get_src_reg_imm(c, inst, 0, i);*/ + src = get_src_reg(c, inst, 0, i); brw_MOV(p, dst, src); } } @@ -554,7 +696,7 @@ static void emit_mov( struct brw_wm_compile *c, } static void emit_pixel_xy(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_reg r1 = brw_vec1_grf(1, 0); struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); @@ -584,7 +726,7 @@ static void emit_pixel_xy(struct brw_wm_compile *c, } static void emit_delta_xy(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_reg r1 = brw_vec1_grf(1, 0); struct brw_reg dst0, dst1, src0, src1; @@ -644,7 +786,7 @@ static void fire_fb_write( struct brw_wm_compile *c, } static void emit_fb_write(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; int nr = 2; @@ -682,27 +824,26 @@ static void emit_fb_write(struct brw_wm_compile *c, } if (c->key.dest_depth_reg) { - GLuint comp = c->key.dest_depth_reg / 2; - GLuint off = c->key.dest_depth_reg % 2; + const GLuint comp = c->key.dest_depth_reg / 2; + const GLuint off = c->key.dest_depth_reg % 2; - assert(comp == 1); - assert(off == 0); -#if 0 - /* XXX do we need this code? comp always 1, off always 0, it seems */ if (off != 0) { + /* XXX this code needs review/testing */ + struct brw_reg arg1_0 = get_src_reg(c, inst, 1, comp); + struct brw_reg arg1_1 = get_src_reg(c, inst, 1, comp+1); + brw_push_insn_state(p); brw_set_compression_control(p, BRW_COMPRESSION_NONE); - brw_MOV(p, brw_message_reg(nr), offset(arg1[comp],1)); + brw_MOV(p, brw_message_reg(nr), offset(arg1_0, 1)); /* 2nd half? */ - brw_MOV(p, brw_message_reg(nr+1), arg1[comp+1]); + brw_MOV(p, brw_message_reg(nr+1), arg1_1); brw_pop_insn_state(p); } else -#endif { - struct brw_reg src = get_src_reg(c, inst, 1, 1); - brw_MOV(p, brw_message_reg(nr), src); + struct brw_reg src = get_src_reg(c, inst, 1, 1); + brw_MOV(p, brw_message_reg(nr), src); } nr += 2; } @@ -713,7 +854,7 @@ static void emit_fb_write(struct brw_wm_compile *c, } static void emit_pixel_w( struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -743,7 +884,7 @@ static void emit_pixel_w( struct brw_wm_compile *c, } static void emit_linterp(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -772,7 +913,7 @@ static void emit_linterp(struct brw_wm_compile *c, } static void emit_cinterp(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -798,7 +939,7 @@ static void emit_cinterp(struct brw_wm_compile *c, } static void emit_pinterp(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -832,7 +973,7 @@ static void emit_pinterp(struct brw_wm_compile *c, /* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */ static void emit_frontfacing(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD); @@ -861,7 +1002,7 @@ static void emit_frontfacing(struct brw_wm_compile *c, } static void emit_xpd(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { int i; struct brw_compile *p = &c->func; @@ -886,7 +1027,7 @@ static void emit_xpd(struct brw_wm_compile *c, } static void emit_dp3(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_reg src0[3], src1[3], dst; int i; @@ -905,7 +1046,7 @@ static void emit_dp3(struct brw_wm_compile *c, } static void emit_dp4(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_reg src0[4], src1[4], dst; int i; @@ -924,7 +1065,7 @@ static void emit_dp4(struct brw_wm_compile *c, } static void emit_dph(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_reg src0[4], src1[4], dst; int i; @@ -948,7 +1089,7 @@ static void emit_dph(struct brw_wm_compile *c, * register's X, Y, Z and W channels (subject to writemasking of course). */ static void emit_math1(struct brw_wm_compile *c, - struct prog_instruction *inst, GLuint func) + const struct prog_instruction *inst, GLuint func) { struct brw_compile *p = &c->func; struct brw_reg src0, dst, tmp; @@ -985,43 +1126,43 @@ static void emit_math1(struct brw_wm_compile *c, } static void emit_rcp(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_math1(c, inst, BRW_MATH_FUNCTION_INV); } static void emit_rsq(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_math1(c, inst, BRW_MATH_FUNCTION_RSQ); } static void emit_sin(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_math1(c, inst, BRW_MATH_FUNCTION_SIN); } static void emit_cos(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_math1(c, inst, BRW_MATH_FUNCTION_COS); } static void emit_ex2(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_math1(c, inst, BRW_MATH_FUNCTION_EXP); } static void emit_lg2(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_math1(c, inst, BRW_MATH_FUNCTION_LOG); } static void emit_add(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg src0, src1, dst; @@ -1040,7 +1181,7 @@ static void emit_add(struct brw_wm_compile *c, } static void emit_arl(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg src0, addr_reg; @@ -1053,7 +1194,7 @@ static void emit_arl(struct brw_wm_compile *c, } static void emit_sub(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg src0, src1, dst; @@ -1072,7 +1213,7 @@ static void emit_sub(struct brw_wm_compile *c, } static void emit_mul(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg src0, src1, dst; @@ -1091,7 +1232,7 @@ static void emit_mul(struct brw_wm_compile *c, } static void emit_frc(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg src0, dst; @@ -1110,7 +1251,7 @@ static void emit_frc(struct brw_wm_compile *c, } static void emit_flr(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg src0, dst; @@ -1177,7 +1318,7 @@ static void emit_min_max(struct brw_wm_compile *c, } static void emit_pow(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg dst, src0, src1; @@ -1199,7 +1340,7 @@ static void emit_pow(struct brw_wm_compile *c, } static void emit_lrp(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -1252,7 +1393,7 @@ static void emit_kil(struct brw_wm_compile *c) } static void emit_mad(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -1275,7 +1416,7 @@ static void emit_mad(struct brw_wm_compile *c, } static void emit_sop(struct brw_wm_compile *c, - struct prog_instruction *inst, GLuint cond) + const struct prog_instruction *inst, GLuint cond) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -1299,43 +1440,43 @@ static void emit_sop(struct brw_wm_compile *c, } static void emit_slt(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_sop(c, inst, BRW_CONDITIONAL_L); } static void emit_sle(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_sop(c, inst, BRW_CONDITIONAL_LE); } static void emit_sgt(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_sop(c, inst, BRW_CONDITIONAL_G); } static void emit_sge(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_sop(c, inst, BRW_CONDITIONAL_GE); } static void emit_seq(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_sop(c, inst, BRW_CONDITIONAL_EQ); } static void emit_sne(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { emit_sop(c, inst, BRW_CONDITIONAL_NEQ); } static void emit_ddx(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -1362,7 +1503,7 @@ static void emit_ddx(struct brw_wm_compile *c, } static void emit_ddy(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -1505,7 +1646,7 @@ static void noise1_sub( struct brw_wm_compile *c ) { } static void emit_noise1( struct brw_wm_compile *c, - struct prog_instruction *inst ) + const struct prog_instruction *inst ) { struct brw_compile *p = &c->func; struct brw_reg src, param, dst; @@ -1675,7 +1816,7 @@ static void noise2_sub( struct brw_wm_compile *c ) { } static void emit_noise2( struct brw_wm_compile *c, - struct prog_instruction *inst ) + const struct prog_instruction *inst ) { struct brw_compile *p = &c->func; struct brw_reg src0, src1, param0, param1, dst; @@ -1978,7 +2119,7 @@ static void noise3_sub( struct brw_wm_compile *c ) { } static void emit_noise3( struct brw_wm_compile *c, - struct prog_instruction *inst ) + const struct prog_instruction *inst ) { struct brw_compile *p = &c->func; struct brw_reg src0, src1, src2, param0, param1, param2, dst; @@ -2401,7 +2542,7 @@ static void noise4_sub( struct brw_wm_compile *c ) } static void emit_noise4( struct brw_wm_compile *c, - struct prog_instruction *inst ) + const struct prog_instruction *inst ) { struct brw_compile *p = &c->func; struct brw_reg src0, src1, src2, src3, param0, param1, param2, param3, dst; @@ -2443,7 +2584,7 @@ static void emit_noise4( struct brw_wm_compile *c, } static void emit_wpos_xy(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; GLuint mask = inst->DstReg.WriteMask; @@ -2479,7 +2620,7 @@ static void emit_wpos_xy(struct brw_wm_compile *c, BIAS on SIMD8 not working yet... */ static void emit_txb(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg dst[4], src[4], payload_reg; @@ -2517,7 +2658,7 @@ static void emit_txb(struct brw_wm_compile *c, retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), /* dest */ 1, /* msg_reg_nr */ retype(payload_reg, BRW_REGISTER_TYPE_UW), /* src0 */ - unit + MAX_DRAW_BUFFERS, /* surface */ + SURF_INDEX_TEXTURE(unit), unit, /* sampler */ inst->DstReg.WriteMask, /* writemask */ BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS, /* msg_type */ @@ -2528,7 +2669,7 @@ static void emit_txb(struct brw_wm_compile *c, static void emit_tex(struct brw_wm_compile *c, - struct prog_instruction *inst) + const struct prog_instruction *inst) { struct brw_compile *p = &c->func; struct brw_reg dst[4], src[4], payload_reg; @@ -2581,7 +2722,7 @@ static void emit_tex(struct brw_wm_compile *c, retype(vec8(dst[0]), BRW_REGISTER_TYPE_UW), /* dest */ 1, /* msg_reg_nr */ retype(payload_reg, BRW_REGISTER_TYPE_UW), /* src0 */ - unit + MAX_DRAW_BUFFERS, /* surface */ + SURF_INDEX_TEXTURE(unit), unit, /* sampler */ inst->DstReg.WriteMask, /* writemask */ BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE, /* msg_type */ @@ -2612,28 +2753,31 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) struct brw_compile *p = &c->func; struct brw_indirect stack_index = brw_indirect(0, 0); - c->reg_index = 0; + c->out_of_regs = GL_FALSE; + prealloc_reg(c); brw_set_compression_control(p, BRW_COMPRESSION_NONE); brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack)); for (i = 0; i < c->nr_fp_insns; i++) { - struct prog_instruction *inst = &c->prog_instructions[i]; + const struct prog_instruction *inst = &c->prog_instructions[i]; - if (inst->CondUpdate) - brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); - else - brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); + c->cur_inst = i; - /* +#if 0 _mesa_printf("Inst %d: ", i); _mesa_print_instruction(inst); - */ +#endif /* fetch any constants that this instruction needs */ - if (c->use_const_buffer) + if (c->fp->use_const_buffer) fetch_constants(c, inst); + if (inst->CondUpdate) + brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ); + else + brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE); + switch (inst->Opcode) { case WM_PIXELXY: emit_pixel_xy(c, inst); @@ -2819,6 +2963,7 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) break; case OPCODE_BGNLOOP: + /* XXX may need to invalidate the current_constant regs */ loop_inst[loop_insn++] = brw_DO(p, BRW_EXECUTE_8); break; case OPCODE_BRK: @@ -2849,17 +2994,13 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) _mesa_printf("unsupported IR in fragment shader %d\n", inst->Opcode); } + if (inst->CondUpdate) brw_set_predicate_control(p, BRW_PREDICATE_NORMAL); else brw_set_predicate_control(p, BRW_PREDICATE_NONE); } post_wm_emit(c); - - if (c->reg_index >= BRW_WM_MAX_GRF) { - _mesa_problem(NULL, "Ran out of registers in brw_wm_emit_glsl()"); - /* XXX we need to do some proper error recovery here */ - } } @@ -2883,6 +3024,6 @@ void brw_wm_glsl_emit(struct brw_context *brw, struct brw_wm_compile *c) brw_wm_print_program(c, "brw_wm_glsl_emit done"); } - c->prog_data.total_grf = c->reg_index; + c->prog_data.total_grf = num_grf_used(c); c->prog_data.total_scratch = 0; } diff --git a/src/mesa/drivers/dri/i965/brw_wm_iz.c b/src/mesa/drivers/dri/i965/brw_wm_iz.c index bd60ac9b31..8fd067abe7 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_iz.c +++ b/src/mesa/drivers/dri/i965/brw_wm_iz.c @@ -116,6 +116,10 @@ const struct { { C, 0, 1, 1, 1 } }; +/** + * \param line_aa AA_NEVER, AA_ALWAYS or AA_SOMETIMES + * \param lookup bitmask of IZ_* flags + */ void brw_wm_lookup_iz( GLuint line_aa, GLuint lookup, struct brw_wm_prog_key *key ) diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass0.c b/src/mesa/drivers/dri/i965/brw_wm_pass0.c index 2debd0678a..92142764f5 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass0.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass0.c @@ -322,7 +322,7 @@ static struct brw_wm_ref *get_new_ref( struct brw_wm_compile *c, newref->value->lastuse = newref; } - if (src.NegateBase & (1<<i)) + if (src.Negate & (1 << i)) newref->hw_reg.negate ^= 1; if (src.Abs) { diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass1.c b/src/mesa/drivers/dri/i965/brw_wm_pass1.c index ab9aa2f10d..3436a24717 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass1.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass1.c @@ -159,6 +159,7 @@ void brw_wm_pass1( struct brw_wm_compile *c ) case OPCODE_FRC: case OPCODE_MOV: case OPCODE_SWZ: + case OPCODE_TRUNC: read0 = writemask; break; diff --git a/src/mesa/drivers/dri/i965/brw_wm_pass2.c b/src/mesa/drivers/dri/i965/brw_wm_pass2.c index 780edbc42e..6faea018fb 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_pass2.c +++ b/src/mesa/drivers/dri/i965/brw_wm_pass2.c @@ -69,8 +69,6 @@ static void prealloc_reg(struct brw_wm_compile *c, */ static void init_registers( struct brw_wm_compile *c ) { - struct brw_context *brw = c->func.brw; - GLuint inputs = (brw->vs.prog_data->outputs_written & DO_SETUP_BITS); GLuint nr_interp_regs = 0; GLuint i = 0; GLuint j; @@ -85,16 +83,19 @@ static void init_registers( struct brw_wm_compile *c ) prealloc_reg(c, &c->creg[j], i++); for (j = 0; j < FRAG_ATTRIB_MAX; j++) { - if (inputs & (1<<j)) { - /* index for vs output and ps input are not the same - in shader varying */ - GLuint index; - if (j > FRAG_ATTRIB_VAR0) - index = j - (VERT_RESULT_VAR0 - FRAG_ATTRIB_VAR0); + if (c->key.vp_outputs_written & (1<<j)) { + int fp_index; + + if (j >= VERT_RESULT_VAR0) + fp_index = j - (VERT_RESULT_VAR0 - FRAG_ATTRIB_VAR0); + else if (j <= VERT_RESULT_TEX7) + fp_index = j; else - index = j; + fp_index = -1; + nr_interp_regs++; - prealloc_reg(c, &c->payload.input_interp[index], i++); + if (fp_index >= 0) + prealloc_reg(c, &c->payload.input_interp[fp_index], i++); } } diff --git a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c index 1fc9f01372..3fc18ff1f3 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_sampler_state.c @@ -152,7 +152,7 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key, sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC; if (key->max_aniso > 2.0) { - sampler->ss3.max_aniso = MAX2((key->max_aniso - 2) / 2, + sampler->ss3.max_aniso = MIN2((key->max_aniso - 2) / 2, BRW_ANISORATIO_16); } } @@ -178,6 +178,16 @@ static void brw_update_sampler_state(struct wm_sampler_entry *key, sampler->ss1.s_wrap_mode = BRW_TEXCOORDMODE_CUBE; sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_CUBE; } + else if (key->tex_target == GL_TEXTURE_1D) { + /* There's a bug in 1D texture sampling - it actually pays + * attention to the wrap_t value, though it should not. + * Override the wrap_t value here to GL_REPEAT to keep + * any nonexistent border pixels from floating in. + */ + sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); + sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); + sampler->ss1.t_wrap_mode = BRW_TEXCOORDMODE_WRAP; + } else { sampler->ss1.r_wrap_mode = translate_wrap_mode(key->wrap_r); sampler->ss1.s_wrap_mode = translate_wrap_mode(key->wrap_s); diff --git a/src/mesa/drivers/dri/i965/brw_wm_state.c b/src/mesa/drivers/dri/i965/brw_wm_state.c index 58fa6aaf8f..67b41173fb 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_state.c @@ -290,7 +290,7 @@ const struct brw_tracked_state brw_wm_unit = { .brw = (BRW_NEW_FRAGMENT_PROGRAM | BRW_NEW_CURBE_OFFSETS | - BRW_NEW_NR_SURFACES), + BRW_NEW_NR_WM_SURFACES), .cache = (CACHE_NEW_WM_PROG | CACHE_NEW_SAMPLER) diff --git a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c index e7d55d5dbd..c49a5f6b4e 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_surface_state.c +++ b/src/mesa/drivers/dri/i965/brw_wm_surface_state.c @@ -38,7 +38,7 @@ #include "intel_mipmap_tree.h" #include "intel_batchbuffer.h" #include "intel_tex.h" - +#include "intel_fbo.h" #include "brw_context.h" #include "brw_state.h" @@ -176,17 +176,6 @@ static GLuint translate_tex_format( GLuint mesa_format, GLenum internal_format, } } -struct brw_wm_surface_key { - GLenum target, depthmode; - dri_bo *bo; - GLint format, internal_format; - GLint first_level, last_level; - GLint width, height, depth; - GLint pitch, cpp; - uint32_t tiling; - GLuint offset; -}; - static void brw_set_surface_tiling(struct brw_surface_state *surf, uint32_t tiling) { @@ -208,7 +197,7 @@ brw_set_surface_tiling(struct brw_surface_state *surf, uint32_t tiling) static dri_bo * brw_create_texture_surface( struct brw_context *brw, - struct brw_wm_surface_key *key ) + struct brw_surface_key *key ) { struct brw_surface_state surf; dri_bo *bo; @@ -263,7 +252,7 @@ brw_create_texture_surface( struct brw_context *brw, surf.ss0.cube_neg_z = 1; } - bo = brw_upload_cache(&brw->cache, BRW_SS_SURFACE, + bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, key, sizeof(*key), &key->bo, key->bo ? 1 : 0, &surf, sizeof(surf), @@ -287,8 +276,8 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit ) struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current; struct intel_texture_object *intelObj = intel_texture_object(tObj); struct gl_texture_image *firstImage = tObj->Image[0][intelObj->firstLevel]; - struct brw_wm_surface_key key; - const GLuint j = MAX_DRAW_BUFFERS + unit; + struct brw_surface_key key; + const GLuint surf = SURF_INDEX_TEXTURE(unit); memset(&key, 0, sizeof(key)); @@ -315,25 +304,26 @@ brw_update_texture_surface( GLcontext *ctx, GLuint unit ) key.cpp = intelObj->mt->cpp; key.tiling = intelObj->mt->region->tiling; - dri_bo_unreference(brw->wm.surf_bo[j]); - brw->wm.surf_bo[j] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), - &key.bo, key.bo ? 1 : 0, - NULL); - if (brw->wm.surf_bo[j] == NULL) { - brw->wm.surf_bo[j] = brw_create_texture_surface(brw, &key); + dri_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, key.bo ? 1 : 0, + NULL); + if (brw->wm.surf_bo[surf] == NULL) { + brw->wm.surf_bo[surf] = brw_create_texture_surface(brw, &key); } } /** - * Create the constant buffer surface. Fragment shader constanst will be + * Create the constant buffer surface. Vertex/fragment shader constants will be * read from this buffer with Data Port Read instructions/messages. */ -static dri_bo * +dri_bo * brw_create_constant_surface( struct brw_context *brw, - struct brw_wm_surface_key *key ) + struct brw_surface_key *key ) { const GLint w = key->width - 1; struct brw_surface_state surf; @@ -345,8 +335,6 @@ brw_create_constant_surface( struct brw_context *brw, surf.ss0.surface_type = BRW_SURFACE_BUFFER; surf.ss0.surface_format = BRW_SURFACEFORMAT_R32G32B32A32_FLOAT; - /* This is ok for all textures with channel width 8bit or less: - */ assert(key->bo); if (key->bo) surf.ss1.base_addr = key->bo->offset; /* reloc */ @@ -356,10 +344,10 @@ brw_create_constant_surface( struct brw_context *brw, surf.ss2.width = w & 0x7f; /* bits 6:0 of size or width */ surf.ss2.height = (w >> 7) & 0x1fff; /* bits 19:7 of size or width */ surf.ss3.depth = (w >> 20) & 0x7f; /* bits 26:20 of size or width */ - surf.ss3.pitch = (key->pitch * key->cpp) - 1; - brw_set_surface_tiling(&surf, key->tiling); + surf.ss3.pitch = (key->pitch * key->cpp) - 1; /* ignored?? */ + brw_set_surface_tiling(&surf, key->tiling); /* tiling now allowed */ - bo = brw_upload_cache(&brw->cache, BRW_SS_SURFACE, + bo = brw_upload_cache(&brw->surface_cache, BRW_SS_SURFACE, key, sizeof(*key), &key->bo, key->bo ? 1 : 0, &surf, sizeof(surf), @@ -377,28 +365,73 @@ brw_create_constant_surface( struct brw_context *brw, return bo; } +/* Creates a new WM constant buffer reflecting the current fragment program's + * constants, if needed by the fragment program. + * + * Otherwise, constants go through the CURBEs using the brw_constant_buffer + * state atom. + */ +static drm_intel_bo * +brw_wm_update_constant_buffer(struct brw_context *brw) +{ + struct intel_context *intel = &brw->intel; + struct brw_fragment_program *fp = + (struct brw_fragment_program *) brw->fragment_program; + const struct gl_program_parameter_list *params = fp->program.Base.Parameters; + const int size = params->NumParameters * 4 * sizeof(GLfloat); + drm_intel_bo *const_buffer; + + /* BRW_NEW_FRAGMENT_PROGRAM */ + if (!fp->use_const_buffer) + return NULL; + + const_buffer = drm_intel_bo_alloc(intel->bufmgr, "fp_const_buffer", + size, 64); + + /* _NEW_PROGRAM_CONSTANTS */ + dri_bo_subdata(const_buffer, 0, size, params->ParameterValues); + + return const_buffer; +} /** - * Update the constant buffer surface. + * Update the surface state for a WM constant buffer. + * The constant buffer will be (re)allocated here if needed. */ static void -brw_update_constant_surface( GLcontext *ctx, - const struct brw_fragment_program *fp ) +brw_update_wm_constant_surface( GLcontext *ctx, + GLuint surf) { struct brw_context *brw = brw_context(ctx); - struct brw_wm_surface_key key; - const GLuint j = BRW_WM_MAX_SURF - 1; - const GLuint numParams = fp->program.Base.Parameters->NumParameters; + struct brw_surface_key key; + struct brw_fragment_program *fp = + (struct brw_fragment_program *) brw->fragment_program; + const struct gl_program_parameter_list *params = + fp->program.Base.Parameters; + + /* If we're in this state update atom, we need to update WM constants, so + * free the old buffer and create a new one for the new contents. + */ + dri_bo_unreference(fp->const_buffer); + fp->const_buffer = brw_wm_update_constant_buffer(brw); + + /* If there's no constant buffer, then no surface BO is needed to point at + * it. + */ + if (fp->const_buffer == 0) { + drm_intel_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = NULL; + return; + } memset(&key, 0, sizeof(key)); key.format = MESA_FORMAT_RGBA_FLOAT32; key.internal_format = GL_RGBA; key.bo = fp->const_buffer; - key.depthmode = GL_NONE; - key.pitch = numParams; - key.width = numParams; + key.pitch = params->NumParameters; + key.width = params->NumParameters; key.height = 1; key.depth = 1; key.cpp = 16; @@ -409,16 +442,60 @@ brw_update_constant_surface( GLcontext *ctx, key.width, key.height, key.depth, key.cpp, key.pitch); */ - dri_bo_unreference(brw->wm.surf_bo[j]); - brw->wm.surf_bo[j] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), - &key.bo, key.bo ? 1 : 0, - NULL); - if (brw->wm.surf_bo[j] == NULL) { - brw->wm.surf_bo[j] = brw_create_constant_surface(brw, &key); + dri_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + &key.bo, key.bo ? 1 : 0, + NULL); + if (brw->wm.surf_bo[surf] == NULL) { + brw->wm.surf_bo[surf] = brw_create_constant_surface(brw, &key); } + brw->state.dirty.brw |= BRW_NEW_WM_SURFACES; } +/** + * Updates surface / buffer for fragment shader constant buffer, if + * one is required. + * + * This consumes the state updates for the constant buffer, and produces + * BRW_NEW_WM_SURFACES to get picked up by brw_prepare_wm_surfaces for + * inclusion in the binding table. + */ +static void prepare_wm_constant_surface(struct brw_context *brw ) +{ + GLcontext *ctx = &brw->intel.ctx; + struct brw_fragment_program *fp = + (struct brw_fragment_program *) brw->fragment_program; + GLuint surf = SURF_INDEX_FRAG_CONST_BUFFER; + + drm_intel_bo_unreference(fp->const_buffer); + fp->const_buffer = brw_wm_update_constant_buffer(brw); + + /* If there's no constant buffer, then no surface BO is needed to point at + * it. + */ + if (fp->const_buffer == 0) { + if (brw->wm.surf_bo[surf] != NULL) { + drm_intel_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = NULL; + brw->state.dirty.brw |= BRW_NEW_WM_SURFACES; + } + return; + } + + brw_update_wm_constant_surface(ctx, surf); +} + +const struct brw_tracked_state brw_wm_constant_surface = { + .dirty = { + .mesa = (_NEW_PROGRAM_CONSTANTS), + .brw = (BRW_NEW_FRAGMENT_PROGRAM), + .cache = 0 + }, + .prepare = prepare_wm_constant_surface, +}; + /** * Sets up a surface state structure to point at the given region. @@ -426,18 +503,22 @@ brw_update_constant_surface( GLcontext *ctx, * usable for further buffers when doing ARB_draw_buffer support. */ static void -brw_update_region_surface(struct brw_context *brw, struct intel_region *region, - unsigned int unit, GLboolean cached) +brw_update_renderbuffer_surface(struct brw_context *brw, + struct gl_renderbuffer *rb, + unsigned int unit) { GLcontext *ctx = &brw->intel.ctx; dri_bo *region_bo = NULL; + struct intel_renderbuffer *irb = intel_renderbuffer(rb); + struct intel_region *region = irb ? irb->region : NULL; struct { unsigned int surface_type; unsigned int surface_format; - unsigned int width, height, cpp; + unsigned int width, height, pitch, cpp; GLubyte color_mask[4]; GLboolean color_blend; uint32_t tiling; + uint32_t draw_offset; } key; memset(&key, 0, sizeof(key)); @@ -446,14 +527,29 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region, region_bo = region->buffer; key.surface_type = BRW_SURFACE_2D; - if (region->cpp == 4) + switch (irb->texformat->MesaFormat) { + case MESA_FORMAT_ARGB8888: key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; - else + break; + case MESA_FORMAT_RGB565: key.surface_format = BRW_SURFACEFORMAT_B5G6R5_UNORM; + break; + case MESA_FORMAT_ARGB1555: + key.surface_format = BRW_SURFACEFORMAT_B5G5R5A1_UNORM; + break; + case MESA_FORMAT_ARGB4444: + key.surface_format = BRW_SURFACEFORMAT_B4G4R4A4_UNORM; + break; + default: + _mesa_problem(ctx, "Bad renderbuffer format: %d\n", + irb->texformat->MesaFormat); + } key.tiling = region->tiling; - key.width = region->pitch; /* XXX: not really! */ + key.width = region->width; key.height = region->height; + key.pitch = region->pitch; key.cpp = region->cpp; + key.draw_offset = region->draw_offset; /* cur 3d or cube face offset */ } else { key.surface_type = BRW_SURFACE_NULL; key.surface_format = BRW_SURFACEFORMAT_B8G8R8A8_UNORM; @@ -461,6 +557,7 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region, key.width = 1; key.height = 1; key.cpp = 4; + key.draw_offset = 0; } memcpy(key.color_mask, ctx->Color.ColorMask, sizeof(key.color_mask)); @@ -468,12 +565,11 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region, ctx->Color.BlendEnabled); dri_bo_unreference(brw->wm.surf_bo[unit]); - brw->wm.surf_bo[unit] = NULL; - if (cached) - brw->wm.surf_bo[unit] = brw_search_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), - ®ion_bo, 1, - NULL); + brw->wm.surf_bo[unit] = brw_search_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), + ®ion_bo, 1, + NULL); if (brw->wm.surf_bo[unit] == NULL) { struct brw_surface_state surf; @@ -482,13 +578,14 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region, surf.ss0.surface_format = key.surface_format; surf.ss0.surface_type = key.surface_type; + surf.ss1.base_addr = key.draw_offset; if (region_bo != NULL) - surf.ss1.base_addr = region_bo->offset; /* reloc */ + surf.ss1.base_addr += region_bo->offset; /* reloc */ surf.ss2.width = key.width - 1; surf.ss2.height = key.height - 1; brw_set_surface_tiling(&surf, key.tiling); - surf.ss3.pitch = (key.width * key.cpp) - 1; + surf.ss3.pitch = (key.pitch * key.cpp) - 1; /* _NEW_COLOR */ surf.ss0.color_blend = key.color_blend; @@ -498,8 +595,9 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region, surf.ss0.writedisable_alpha = !key.color_mask[3]; /* Key size will never match key size for textures, so we're safe. */ - brw->wm.surf_bo[unit] = brw_upload_cache(&brw->cache, BRW_SS_SURFACE, - &key, sizeof(key), + brw->wm.surf_bo[unit] = brw_upload_cache(&brw->surface_cache, + BRW_SS_SURFACE, + &key, sizeof(key), ®ion_bo, 1, &surf, sizeof(surf), NULL, NULL); @@ -508,12 +606,12 @@ brw_update_region_surface(struct brw_context *brw, struct intel_region *region, * them both. We might be able to figure out from other state * a more restrictive relocation to emit. */ - dri_bo_emit_reloc(brw->wm.surf_bo[unit], - I915_GEM_DOMAIN_RENDER, - I915_GEM_DOMAIN_RENDER, - 0, - offsetof(struct brw_surface_state, ss1), - region_bo); + drm_intel_bo_emit_reloc(brw->wm.surf_bo[unit], + offsetof(struct brw_surface_state, ss1), + region_bo, + key.draw_offset, + I915_GEM_DOMAIN_RENDER, + I915_GEM_DOMAIN_RENDER); } } } @@ -528,7 +626,9 @@ brw_wm_get_binding_table(struct brw_context *brw) { dri_bo *bind_bo; - bind_bo = brw_search_cache(&brw->cache, BRW_SS_SURF_BIND, + assert(brw->wm.nr_surfaces <= BRW_WM_MAX_SURF); + + bind_bo = brw_search_cache(&brw->surface_cache, BRW_SS_SURF_BIND, NULL, 0, brw->wm.surf_bo, brw->wm.nr_surfaces, NULL); @@ -544,7 +644,7 @@ brw_wm_get_binding_table(struct brw_context *brw) else data[i] = 0; - bind_bo = brw_upload_cache( &brw->cache, BRW_SS_SURF_BIND, + bind_bo = brw_upload_cache( &brw->surface_cache, BRW_SS_SURF_BIND, NULL, 0, brw->wm.surf_bo, brw->wm.nr_surfaces, data, data_size, @@ -574,66 +674,62 @@ static void prepare_wm_surfaces(struct brw_context *brw ) GLuint i; int old_nr_surfaces; + /* _NEW_BUFFERS */ /* Update surfaces for drawing buffers */ - if (brw->state.nr_color_regions > 1) { - for (i = 0; i < brw->state.nr_color_regions; i++) { - brw_update_region_surface(brw, brw->state.color_regions[i], i, - GL_FALSE); + if (ctx->DrawBuffer->_NumColorDrawBuffers >= 1) { + for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) { + brw_update_renderbuffer_surface(brw, + ctx->DrawBuffer->_ColorDrawBuffers[i], + i); } } else { - brw_update_region_surface(brw, brw->state.color_regions[0], 0, GL_TRUE); + brw_update_renderbuffer_surface(brw, NULL, 0); } old_nr_surfaces = brw->wm.nr_surfaces; brw->wm.nr_surfaces = MAX_DRAW_BUFFERS; + if (brw->wm.surf_bo[SURF_INDEX_FRAG_CONST_BUFFER] != NULL) + brw->wm.nr_surfaces = SURF_INDEX_FRAG_CONST_BUFFER + 1; + /* Update surfaces for textures */ for (i = 0; i < BRW_MAX_TEX_UNIT; i++) { const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[i]; - const GLuint j = MAX_DRAW_BUFFERS + i; + const GLuint surf = SURF_INDEX_TEXTURE(i); /* _NEW_TEXTURE, BRW_NEW_TEXDATA */ if (texUnit->_ReallyEnabled) { if (texUnit->_Current == intel->frame_buffer_texobj) { /* render to texture */ - dri_bo_unreference(brw->wm.surf_bo[j]); - brw->wm.surf_bo[j] = brw->wm.surf_bo[0]; - dri_bo_reference(brw->wm.surf_bo[j]); - brw->wm.nr_surfaces = j + 1; + dri_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = brw->wm.surf_bo[0]; + dri_bo_reference(brw->wm.surf_bo[surf]); + brw->wm.nr_surfaces = surf + 1; } else { /* regular texture */ brw_update_texture_surface(ctx, i); - brw->wm.nr_surfaces = j + 1; + brw->wm.nr_surfaces = surf + 1; } } else { - dri_bo_unreference(brw->wm.surf_bo[j]); - brw->wm.surf_bo[j] = NULL; + dri_bo_unreference(brw->wm.surf_bo[surf]); + brw->wm.surf_bo[surf] = NULL; } } - /* Update surface for fragment shader constant buffer */ - { - const GLuint j = BRW_WM_MAX_SURF - 1; - const struct brw_fragment_program *fp = - brw_fragment_program_const(brw->fragment_program); - - brw_update_constant_surface(ctx, fp); - brw->wm.nr_surfaces = j + 1; - } - - dri_bo_unreference(brw->wm.bind_bo); brw->wm.bind_bo = brw_wm_get_binding_table(brw); if (brw->wm.nr_surfaces != old_nr_surfaces) - brw->state.dirty.brw |= BRW_NEW_NR_SURFACES; + brw->state.dirty.brw |= BRW_NEW_NR_WM_SURFACES; } - const struct brw_tracked_state brw_wm_surfaces = { .dirty = { - .mesa = _NEW_COLOR | _NEW_TEXTURE | _NEW_BUFFERS, - .brw = BRW_NEW_CONTEXT, + .mesa = (_NEW_COLOR | + _NEW_TEXTURE | + _NEW_BUFFERS), + .brw = (BRW_NEW_CONTEXT | + BRW_NEW_WM_SURFACES), .cache = 0 }, .prepare = prepare_wm_surfaces, diff --git a/src/mesa/drivers/dri/i965/intel_generatemipmap.c b/src/mesa/drivers/dri/i965/intel_generatemipmap.c new file mode 120000 index 0000000000..4c6b37ada0 --- /dev/null +++ b/src/mesa/drivers/dri/i965/intel_generatemipmap.c @@ -0,0 +1 @@ +../intel/intel_generatemipmap.c
\ No newline at end of file |