X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fr300%2Fr300_fs.c;h=ffb041e2488c735dfde0572c7780d298cb38d294;hb=7691de0dcefd7a518ee1ecc4d2cd3803e42cc803;hp=4d61f6385301d53b8793e548509a72f31c912a97;hpb=64644ec3b21884d4a974fa29087fa98c4ed9e112;p=mesa.git diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c index 4d61f638530..ffb041e2488 100644 --- a/src/gallium/drivers/r300/r300_fs.c +++ b/src/gallium/drivers/r300/r300_fs.c @@ -22,20 +22,23 @@ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE * USE OR OTHER DEALINGS IN THE SOFTWARE. */ +#include "util/format/u_format.h" #include "util/u_math.h" #include "util/u_memory.h" #include "tgsi/tgsi_dump.h" #include "tgsi/tgsi_ureg.h" +#include "r300_cb.h" #include "r300_context.h" +#include "r300_emit.h" #include "r300_screen.h" #include "r300_fs.h" #include "r300_reg.h" +#include "r300_texture.h" #include "r300_tgsi_to_rc.h" -#include "radeon_code.h" -#include "radeon_compiler.h" +#include "compiler/radeon_compiler.h" /* Convert info about FS input semantics to r300_shader_semantics. */ void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, @@ -70,6 +73,11 @@ void r300_shader_read_fs_inputs(struct tgsi_shader_info* info, fs_inputs->wpos = i; break; + case TGSI_SEMANTIC_FACE: + assert(index == 0); + fs_inputs->face = i; + break; + default: fprintf(stderr, "r300: FP: Unknown input semantic: %i\n", info->input_semantic_name[i]); @@ -118,6 +126,9 @@ static void allocate_hardware_inputs( allocate(mydata, inputs->color[i], reg++); } } + if (inputs->face != ATTR_UNUSED) { + allocate(mydata, inputs->face, reg++); + } for (i = 0; i < ATTR_GENERIC_COUNT; i++) { if (inputs->generic[i] != ATTR_UNUSED) { allocate(mydata, inputs->generic[i], reg++); @@ -138,17 +149,21 @@ static void get_external_state( struct r300_textures_state *texstate = r300->textures_state.state; unsigned i; + state->alpha_to_one = r300->alpha_to_one && r300->msaa_enable; + for (i = 0; i < texstate->sampler_state_count; i++) { - struct r300_sampler_state* s = texstate->sampler_states[i]; + struct r300_sampler_state *s = texstate->sampler_states[i]; + struct r300_sampler_view *v = texstate->sampler_views[i]; + struct r300_resource *t; - if (!s) { + if (!s || !v) { continue; } + t = r300_resource(v->base.texture); + if (s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) { - /* XXX Gallium doesn't provide us with any information regarding - * this mode, so we are screwed. I'm setting 0 = LUMINANCE. */ - state->unit[i].depth_texture_mode = 0; + state->unit[i].compare_mode_enabled = 1; /* Fortunately, no need to translate this. */ state->unit[i].texture_compare_func = s->state.compare_func; @@ -156,35 +171,36 @@ static void get_external_state( state->unit[i].non_normalized_coords = !s->state.normalized_coords; - if (texstate->sampler_views[i]) { - struct r300_texture *t; - t = (struct r300_texture*)texstate->sampler_views[i]->base.texture; - - /* XXX this should probably take into account STR, not just S. */ - if (t->uses_pitch) { - switch (s->state.wrap_s) { - case PIPE_TEX_WRAP_REPEAT: - state->unit[i].wrap_mode = RC_WRAP_REPEAT; - state->unit[i].fake_npot = TRUE; - break; - - case PIPE_TEX_WRAP_MIRROR_REPEAT: - state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT; - state->unit[i].fake_npot = TRUE; - break; - - case PIPE_TEX_WRAP_MIRROR_CLAMP: - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: - case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: - state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP; - state->unit[i].fake_npot = TRUE; - break; - - default: - state->unit[i].wrap_mode = RC_WRAP_NONE; - break; - } + /* Pass texture swizzling to the compiler, some lowering passes need it. */ + if (state->unit[i].compare_mode_enabled) { + state->unit[i].texture_swizzle = + RC_MAKE_SWIZZLE(v->swizzle[0], v->swizzle[1], + v->swizzle[2], v->swizzle[3]); + } + + /* XXX this should probably take into account STR, not just S. */ + if (t->tex.is_npot) { + switch (s->state.wrap_s) { + case PIPE_TEX_WRAP_REPEAT: + state->unit[i].wrap_mode = RC_WRAP_REPEAT; + break; + + case PIPE_TEX_WRAP_MIRROR_REPEAT: + state->unit[i].wrap_mode = RC_WRAP_MIRRORED_REPEAT; + break; + + case PIPE_TEX_WRAP_MIRROR_CLAMP: + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: + case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: + state->unit[i].wrap_mode = RC_WRAP_MIRRORED_CLAMP; + break; + + default: + state->unit[i].wrap_mode = RC_WRAP_NONE; } + + if (t->b.b.target == PIPE_TEXTURE_3D) + state->unit[i].clamp_and_scale_before_fetch = TRUE; } } } @@ -204,7 +220,7 @@ static void r300_dummy_fragment_shader( struct ureg_src imm; /* Make a simple fragment shader which outputs (0, 0, 0, 1) */ - ureg = ureg_create(TGSI_PROCESSOR_FRAGMENT); + ureg = ureg_create(PIPE_SHADER_FRAGMENT); out = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0); imm = ureg_imm4f(ureg, 0, 0, 0, 1); @@ -219,6 +235,182 @@ static void r300_dummy_fragment_shader( ureg_destroy(ureg); } +static void r300_emit_fs_code_to_buffer( + struct r300_context *r300, + struct r300_fragment_shader_code *shader) +{ + struct rX00_fragment_program_code *generic_code = &shader->code; + unsigned imm_count = shader->immediates_count; + unsigned imm_first = shader->externals_count; + unsigned imm_end = generic_code->constants.Count; + struct rc_constant *constants = generic_code->constants.Constants; + unsigned i; + CB_LOCALS; + + if (r300->screen->caps.is_r500) { + struct r500_fragment_program_code *code = &generic_code->code.r500; + + shader->cb_code_size = 19 + + ((code->inst_end + 1) * 6) + + imm_count * 7 + + code->int_constant_count * 2; + + NEW_CB(shader->cb_code, shader->cb_code_size); + OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); + OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx); + OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl); + for(i = 0; i < code->int_constant_count; i++){ + OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4), + code->int_constants[i]); + } + OUT_CB_REG(R500_US_CODE_RANGE, + R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); + OUT_CB_REG(R500_US_CODE_OFFSET, 0); + OUT_CB_REG(R500_US_CODE_ADDR, + R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); + + OUT_CB_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); + OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); + for (i = 0; i <= code->inst_end; i++) { + OUT_CB(code->inst[i].inst0); + OUT_CB(code->inst[i].inst1); + OUT_CB(code->inst[i].inst2); + OUT_CB(code->inst[i].inst3); + OUT_CB(code->inst[i].inst4); + OUT_CB(code->inst[i].inst5); + } + + /* Emit immediates. */ + if (imm_count) { + for(i = imm_first; i < imm_end; ++i) { + if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { + const float *data = constants[i].u.Immediate; + + OUT_CB_REG(R500_GA_US_VECTOR_INDEX, + R500_GA_US_VECTOR_INDEX_TYPE_CONST | + (i & R500_GA_US_VECTOR_INDEX_MASK)); + OUT_CB_ONE_REG(R500_GA_US_VECTOR_DATA, 4); + OUT_CB_TABLE(data, 4); + } + } + } + } else { /* r300 */ + struct r300_fragment_program_code *code = &generic_code->code.r300; + unsigned int alu_length = code->alu.length; + unsigned int alu_iterations = ((alu_length - 1) / 64) + 1; + unsigned int tex_length = code->tex.length; + unsigned int tex_iterations = + tex_length > 0 ? ((tex_length - 1) / 32) + 1 : 0; + unsigned int iterations = + alu_iterations > tex_iterations ? alu_iterations : tex_iterations; + unsigned int bank = 0; + + shader->cb_code_size = 15 + + /* R400_US_CODE_BANK */ + (r300->screen->caps.is_r400 ? 2 * (iterations + 1): 0) + + /* R400_US_CODE_EXT */ + (r300->screen->caps.is_r400 ? 2 : 0) + + /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0, R400_US_ALU_EXT_ADDR_0 */ + (code->r390_mode ? (5 * alu_iterations) : 4) + + /* R400_US_ALU_EXT_ADDR_[0-63] */ + (code->r390_mode ? (code->alu.length) : 0) + + /* R300_US_ALU_{RGB,ALPHA}_{INST,ADDR}_0 */ + code->alu.length * 4 + + /* R300_US_TEX_INST_0, R300_US_TEX_INST_[0-31] */ + (code->tex.length > 0 ? code->tex.length + tex_iterations : 0) + + imm_count * 5; + + NEW_CB(shader->cb_code, shader->cb_code_size); + + OUT_CB_REG(R300_US_CONFIG, code->config); + OUT_CB_REG(R300_US_PIXSIZE, code->pixsize); + OUT_CB_REG(R300_US_CODE_OFFSET, code->code_offset); + + if (code->r390_mode) { + OUT_CB_REG(R400_US_CODE_EXT, code->r400_code_offset_ext); + } else if (r300->screen->caps.is_r400) { + /* This register appears to affect shaders even if r390_mode is + * disabled, so it needs to be set to 0 for shaders that + * don't use r390_mode. */ + OUT_CB_REG(R400_US_CODE_EXT, 0); + } + + OUT_CB_REG_SEQ(R300_US_CODE_ADDR_0, 4); + OUT_CB_TABLE(code->code_addr, 4); + + do { + unsigned int bank_alu_length = (alu_length < 64 ? alu_length : 64); + unsigned int bank_alu_offset = bank * 64; + unsigned int bank_tex_length = (tex_length < 32 ? tex_length : 32); + unsigned int bank_tex_offset = bank * 32; + + if (r300->screen->caps.is_r400) { + OUT_CB_REG(R400_US_CODE_BANK, code->r390_mode ? + (bank << R400_BANK_SHIFT) | R400_R390_MODE_ENABLE : 0);//2 + } + + if (bank_alu_length > 0) { + OUT_CB_REG_SEQ(R300_US_ALU_RGB_INST_0, bank_alu_length); + for (i = 0; i < bank_alu_length; i++) + OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_inst); + + OUT_CB_REG_SEQ(R300_US_ALU_RGB_ADDR_0, bank_alu_length); + for (i = 0; i < bank_alu_length; i++) + OUT_CB(code->alu.inst[i + bank_alu_offset].rgb_addr); + + OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_INST_0, bank_alu_length); + for (i = 0; i < bank_alu_length; i++) + OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_inst); + + OUT_CB_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, bank_alu_length); + for (i = 0; i < bank_alu_length; i++) + OUT_CB(code->alu.inst[i + bank_alu_offset].alpha_addr); + + if (code->r390_mode) { + OUT_CB_REG_SEQ(R400_US_ALU_EXT_ADDR_0, bank_alu_length); + for (i = 0; i < bank_alu_length; i++) + OUT_CB(code->alu.inst[i + bank_alu_offset].r400_ext_addr); + } + } + + if (bank_tex_length > 0) { + OUT_CB_REG_SEQ(R300_US_TEX_INST_0, bank_tex_length); + OUT_CB_TABLE(code->tex.inst + bank_tex_offset, bank_tex_length); + } + + alu_length -= bank_alu_length; + tex_length -= bank_tex_length; + bank++; + } while(code->r390_mode && (alu_length > 0 || tex_length > 0)); + + /* R400_US_CODE_BANK needs to be reset to 0, otherwise some shaders + * will be rendered incorrectly. */ + if (r300->screen->caps.is_r400) { + OUT_CB_REG(R400_US_CODE_BANK, + code->r390_mode ? R400_R390_MODE_ENABLE : 0); + } + + /* Emit immediates. */ + if (imm_count) { + for(i = imm_first; i < imm_end; ++i) { + if (constants[i].Type == RC_CONSTANT_IMMEDIATE) { + const float *data = constants[i].u.Immediate; + + OUT_CB_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4); + OUT_CB(pack_float24(data[0])); + OUT_CB(pack_float24(data[1])); + OUT_CB(pack_float24(data[2])); + OUT_CB(pack_float24(data[3])); + } + } + } + } + + OUT_CB_REG(R300_FG_DEPTH_SRC, shader->fg_depth_src); + OUT_CB_REG(R300_US_W_FMT, shader->us_out_w); + END_CB; +} + static void r300_translate_fragment_shader( struct r300_context* r300, struct r300_fragment_shader_code* shader, @@ -226,30 +418,46 @@ static void r300_translate_fragment_shader( { struct r300_fragment_program_compiler compiler; struct tgsi_to_rc ttr; - int wpos; + int wpos, face; unsigned i; tgsi_scan_shader(tokens, &shader->info); r300_shader_read_fs_inputs(&shader->info, &shader->inputs); wpos = shader->inputs.wpos; + face = shader->inputs.face; /* Setup the compiler. */ memset(&compiler, 0, sizeof(compiler)); - rc_init(&compiler.Base); - compiler.Base.Debug = DBG_ON(r300, DBG_FP); + rc_init(&compiler.Base, &r300->fs_regalloc_state); + DBG_ON(r300, DBG_FP) ? compiler.Base.Debug |= RC_DBG_LOG : 0; + DBG_ON(r300, DBG_P_STAT) ? compiler.Base.Debug |= RC_DBG_STATS : 0; compiler.code = &shader->code; compiler.state = shader->compare_state; - compiler.is_r500 = r300->screen->caps.is_r500; - compiler.max_temp_regs = compiler.is_r500 ? 128 : 32; + compiler.Base.is_r500 = r300->screen->caps.is_r500; + compiler.Base.is_r400 = r300->screen->caps.is_r400; + compiler.Base.disable_optimizations = DBG_ON(r300, DBG_NO_OPT); + compiler.Base.has_half_swizzles = TRUE; + compiler.Base.has_presub = TRUE; + compiler.Base.has_omod = TRUE; + compiler.Base.max_temp_regs = + compiler.Base.is_r500 ? 128 : (compiler.Base.is_r400 ? 64 : 32); + compiler.Base.max_constants = compiler.Base.is_r500 ? 256 : 32; + compiler.Base.max_alu_insts = + (compiler.Base.is_r500 || compiler.Base.is_r400) ? 512 : 64; + compiler.Base.max_tex_insts = + (compiler.Base.is_r500 || compiler.Base.is_r400) ? 512 : 32; compiler.AllocateHwInputs = &allocate_hardware_inputs; compiler.UserData = &shader->inputs; find_output_registers(&compiler, shader); - if (compiler.Base.Debug) { - debug_printf("r300: Initial fragment program\n"); + shader->write_all = + shader->info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS]; + + if (compiler.Base.Debug & RC_DBG_LOG) { + DBG(r300, DBG_FP, "r300: Initial fragment program\n"); tgsi_dump(tokens, 0); } @@ -260,6 +468,18 @@ static void r300_translate_fragment_shader( r300_tgsi_to_rc(&ttr, tokens); + if (ttr.error) { + fprintf(stderr, "r300 FP: Cannot translate a shader. " + "Using a dummy shader instead.\n"); + r300_dummy_fragment_shader(r300, shader); + return; + } + + if (!r300->screen->caps.is_r500 || + compiler.Base.Program.Constants.Count > 200) { + compiler.Base.remove_unused_constants = TRUE; + } + /** * Transform the program to support WPOS. * @@ -272,6 +492,10 @@ static void r300_translate_fragment_shader( rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE); } + if (face != ATTR_UNUSED) { + rc_transform_fragment_face(&compiler.Base, face); + } + /* Invoke the compiler */ r3xx_compile_fragment_program(&compiler); @@ -290,8 +514,21 @@ static void r300_translate_fragment_shader( return; } + /* Shaders with zero instructions are invalid, + * use the dummy shader instead. */ + if (shader->code.code.r500.inst_end == -1) { + rc_destroy(&compiler.Base); + r300_dummy_fragment_shader(r300, shader); + return; + } + /* Initialize numbers of constants for each type. */ - shader->externals_count = ttr.immediate_offset; + shader->externals_count = 0; + for (i = 0; + i < shader->code.constants.Count && + shader->code.constants.Constants[i].Type == RC_CONSTANT_EXTERNAL; i++) { + shader->externals_count = i+1; + } shader->immediates_count = 0; shader->rc_state_count = 0; @@ -319,14 +556,18 @@ static void r300_translate_fragment_shader( /* And, finally... */ rc_destroy(&compiler.Base); + + /* Build the command buffer. */ + r300_emit_fs_code_to_buffer(r300, shader); } boolean r300_pick_fragment_shader(struct r300_context* r300) { struct r300_fragment_shader* fs = r300_fs(r300); - struct r300_fragment_program_external_state state = {{{ 0 }}}; + struct r300_fragment_program_external_state state; struct r300_fragment_shader_code* ptr; + memset(&state, 0, sizeof(state)); get_external_state(r300, &state); if (!fs->first) {