diff options
author | Corbin Simpson <MostAwesomeDude@gmail.com> | 2009-08-02 16:56:52 -0700 |
---|---|---|
committer | Corbin Simpson <MostAwesomeDude@gmail.com> | 2009-08-02 16:56:52 -0700 |
commit | 30bca7a4e6ad4e5a9fc74aba2eb854bb1341cca7 (patch) | |
tree | 225de18a7604621744e35875261a247abd114e42 /src/gallium/drivers/r300/r300_emit.c | |
parent | d07bf2b35c1419bdef824092e509c70ffb11fd67 (diff) | |
parent | d0c398a8e2985b855f923aec3470cef8734a622a (diff) |
Merge commit 'nha/r300-compiler-gallium'
Diffstat (limited to 'src/gallium/drivers/r300/r300_emit.c')
-rw-r--r-- | src/gallium/drivers/r300/r300_emit.c | 251 |
1 files changed, 170 insertions, 81 deletions
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c index ac510ffc2e..e0c38a06e4 100644 --- a/src/gallium/drivers/r300/r300_emit.c +++ b/src/gallium/drivers/r300/r300_emit.c @@ -24,6 +24,9 @@ #include "r300_emit.h" +#include "r300_fs.h" +#include "r300_vs.h" + void r300_emit_blend_state(struct r300_context* r300, struct r300_blend_state* blend) { @@ -109,73 +112,158 @@ void r300_emit_dsa_state(struct r300_context* r300, END_CS; } -void r300_emit_fragment_shader(struct r300_context* r300, - struct r3xx_fragment_shader* fs) +static const float * get_shader_constant( + struct r300_context * r300, + struct rc_constant * constant, + struct r300_constant_buffer * externals) +{ + static const float zero[4] = { 0.0, 0.0, 0.0, 0.0 }; + switch(constant->Type) { + case RC_CONSTANT_EXTERNAL: + return externals->constants[constant->u.External]; + + case RC_CONSTANT_IMMEDIATE: + return constant->u.Immediate; + + default: + debug_printf("r300: Implementation error: Unhandled constant type %i\n", + constant->Type); + return zero; + } +} + +/* Convert a normal single-precision float into the 7.16 format + * used by the R300 fragment shader. + */ +static uint32_t pack_float24(float f) { + union { + float fl; + uint32_t u; + } u; + float mantissa; + int exponent; + uint32_t float24 = 0; + + if (f == 0.0) + return 0; + + u.fl = f; + + mantissa = frexpf(f, &exponent); + + /* Handle -ve */ + if (mantissa < 0) { + float24 |= (1 << 23); + mantissa = mantissa * -1.0; + } + /* Handle exponent, bias of 63 */ + exponent += 62; + float24 |= (exponent << 16); + /* Kill 7 LSB of mantissa */ + float24 |= (u.u & 0x7FFFFF) >> 7; + + return float24; +} + +void r300_emit_fragment_program_code(struct r300_context* r300, + struct rX00_fragment_program_code* generic_code, + struct r300_constant_buffer* externals) +{ + struct r300_fragment_program_code * code = &generic_code->code.r300; + struct rc_constant_list * constants = &generic_code->constants; int i; CS_LOCALS(r300); - BEGIN_CS(22); - - OUT_CS_REG(R300_US_CONFIG, fs->indirections); - OUT_CS_REG(R300_US_PIXSIZE, fs->shader.stack_size); - /* XXX figure out exactly how big the sizes are on this reg */ - OUT_CS_REG(R300_US_CODE_OFFSET, 0x40); - /* XXX figure these ones out a bit better kthnx */ - OUT_CS_REG(R300_US_CODE_ADDR_0, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_1, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_2, 0x0); - OUT_CS_REG(R300_US_CODE_ADDR_3, 0x40 | R300_RGBA_OUT); - - for (i = 0; i < fs->alu_instruction_count; i++) { - OUT_CS_REG(R300_US_ALU_RGB_INST_0 + (4 * i), - fs->instructions[i].alu_rgb_inst); - OUT_CS_REG(R300_US_ALU_RGB_ADDR_0 + (4 * i), - fs->instructions[i].alu_rgb_addr); - OUT_CS_REG(R300_US_ALU_ALPHA_INST_0 + (4 * i), - fs->instructions[i].alu_alpha_inst); - OUT_CS_REG(R300_US_ALU_ALPHA_ADDR_0 + (4 * i), - fs->instructions[i].alu_alpha_addr); + BEGIN_CS(15 + + code->alu.length * 4 + + (code->tex.length ? (1 + code->tex.length) : 0) + + (constants->Count ? (1 + constants->Count * 4) : 0)); + + OUT_CS_REG(R300_US_CONFIG, code->config); + OUT_CS_REG(R300_US_PIXSIZE, code->pixsize); + OUT_CS_REG(R300_US_CODE_OFFSET, code->code_offset); + + OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0, 4); + for(i = 0; i < 4; ++i) + OUT_CS(code->code_addr[i]); + + OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].rgb_inst); + + OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].rgb_addr); + + OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].alpha_inst); + + OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length); + for (i = 0; i < code->alu.length; i++) + OUT_CS(code->alu.inst[i].alpha_addr); + + if (code->tex.length) { + OUT_CS_REG_SEQ(R300_US_TEX_INST_0, code->tex.length); + for(i = 0; i < code->tex.length; ++i) + OUT_CS(code->tex.inst[i]); + } + + if (constants->Count) { + OUT_CS_ONE_REG(R300_PFS_PARAM_0_X, constants->Count * 4); + for(i = 0; i < constants->Count; ++i) { + const float * data = get_shader_constant(r300, &constants->Constants[i], externals); + OUT_CS(pack_float24(data[0])); + OUT_CS(pack_float24(data[1])); + OUT_CS(pack_float24(data[2])); + OUT_CS(pack_float24(data[3])); + } } END_CS; } -void r500_emit_fragment_shader(struct r300_context* r300, - struct r5xx_fragment_shader* fs) +void r500_emit_fragment_program_code(struct r300_context* r300, + struct rX00_fragment_program_code* generic_code, + struct r300_constant_buffer* externals) { + struct r500_fragment_program_code * code = &generic_code->code.r500; + struct rc_constant_list * constants = &generic_code->constants; int i; - struct r300_constant_buffer* constants = - &r300->shader_constants[PIPE_SHADER_FRAGMENT]; CS_LOCALS(r300); - BEGIN_CS(9 + (fs->instruction_count * 6) + (constants->count ? 3 : 0) + - (constants->count * 4)); - OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO); - OUT_CS_REG(R500_US_PIXSIZE, fs->shader.stack_size); - OUT_CS_REG(R500_US_CODE_ADDR, R500_US_CODE_START_ADDR(0) | - R500_US_CODE_END_ADDR(fs->instruction_count)); + BEGIN_CS(13 + + ((code->inst_end + 1) * 6) + + (constants->Count ? (3 + (constants->Count * 4)) : 0)); + OUT_CS_REG(R500_US_CONFIG, 0); + OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx); + OUT_CS_REG(R500_US_CODE_RANGE, + R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end)); + OUT_CS_REG(R500_US_CODE_OFFSET, 0); + OUT_CS_REG(R500_US_CODE_ADDR, + R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end)); OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, fs->instruction_count * 6); - for (i = 0; i < fs->instruction_count; i++) { - OUT_CS(fs->instructions[i].inst0); - OUT_CS(fs->instructions[i].inst1); - OUT_CS(fs->instructions[i].inst2); - OUT_CS(fs->instructions[i].inst3); - OUT_CS(fs->instructions[i].inst4); - OUT_CS(fs->instructions[i].inst5); - } - - if (constants->count) { - OUT_CS_REG(R500_GA_US_VECTOR_INDEX, - R500_GA_US_VECTOR_INDEX_TYPE_CONST); - OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->count * 4); - for (i = 0; i < constants->count; i++) { - OUT_CS_32F(constants->constants[i][0]); - OUT_CS_32F(constants->constants[i][1]); - OUT_CS_32F(constants->constants[i][2]); - OUT_CS_32F(constants->constants[i][3]); + OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6); + for (i = 0; i <= code->inst_end; i++) { + OUT_CS(code->inst[i].inst0); + OUT_CS(code->inst[i].inst1); + OUT_CS(code->inst[i].inst2); + OUT_CS(code->inst[i].inst3); + OUT_CS(code->inst[i].inst4); + OUT_CS(code->inst[i].inst5); + } + + if (constants->Count) { + OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST); + OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4); + for (i = 0; i < constants->Count; i++) { + const float * data = get_shader_constant(r300, &constants->Constants[i], externals); + OUT_CS_32F(data[0]); + OUT_CS_32F(data[1]); + OUT_CS_32F(data[2]); + OUT_CS_32F(data[3]); } } @@ -380,13 +468,13 @@ void r300_emit_vertex_format_state(struct r300_context* r300) END_CS; } -void r300_emit_vertex_shader(struct r300_context* r300, - struct r300_vertex_shader* vs) +void r300_emit_vertex_program_code(struct r300_context* r300, + struct r300_vertex_program_code* code, + struct r300_constant_buffer* constants) { int i; struct r300_screen* r300screen = r300_screen(r300->context.screen); - struct r300_constant_buffer* constants = - &r300->shader_constants[PIPE_SHADER_VERTEX]; + unsigned instruction_count = code->length / 4; CS_LOCALS(r300); if (!r300screen->caps->has_tcl) { @@ -395,10 +483,10 @@ void r300_emit_vertex_shader(struct r300_context* r300, return; } - if (constants->count) { - BEGIN_CS(14 + (vs->instruction_count * 4) + (constants->count * 4)); + if (code->constants.Count) { + BEGIN_CS(14 + code->length + (code->constants.Count * 4)); } else { - BEGIN_CS(11 + (vs->instruction_count * 4)); + BEGIN_CS(11 + code->length); } /* R300_VAP_PVS_CODE_CNTL_0 @@ -408,30 +496,27 @@ void r300_emit_vertex_shader(struct r300_context* r300, * XXX these could be optimized to select better values... */ OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3); OUT_CS(R300_PVS_FIRST_INST(0) | - R300_PVS_XYZW_VALID_INST(vs->instruction_count - 1) | - R300_PVS_LAST_INST(vs->instruction_count - 1)); - OUT_CS(R300_PVS_MAX_CONST_ADDR(constants->count - 1)); - OUT_CS(vs->instruction_count - 1); + R300_PVS_XYZW_VALID_INST(instruction_count - 1) | + R300_PVS_LAST_INST(instruction_count - 1)); + OUT_CS(R300_PVS_MAX_CONST_ADDR(code->constants.Count - 1)); + OUT_CS(instruction_count - 1); OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0); - OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, vs->instruction_count * 4); - for (i = 0; i < vs->instruction_count; i++) { - OUT_CS(vs->instructions[i].inst0); - OUT_CS(vs->instructions[i].inst1); - OUT_CS(vs->instructions[i].inst2); - OUT_CS(vs->instructions[i].inst3); - } + OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length); + for (i = 0; i < code->length; i++) + OUT_CS(code->body.d[i]); - if (constants->count) { + if (code->constants.Count) { OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, (r300screen->caps->is_r500 ? R500_PVS_CONST_START : R300_PVS_CONST_START)); - OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, constants->count * 4); - for (i = 0; i < constants->count; i++) { - OUT_CS_32F(constants->constants[i][0]); - OUT_CS_32F(constants->constants[i][1]); - OUT_CS_32F(constants->constants[i][2]); - OUT_CS_32F(constants->constants[i][3]); + OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->constants.Count * 4); + for (i = 0; i < code->constants.Count; i++) { + const float * data = get_shader_constant(r300, &code->constants.Constants[i], constants); + OUT_CS_32F(data[0]); + OUT_CS_32F(data[1]); + OUT_CS_32F(data[2]); + OUT_CS_32F(data[3]); } } @@ -443,6 +528,12 @@ void r300_emit_vertex_shader(struct r300_context* r300, END_CS; } +void r300_emit_vertex_shader(struct r300_context* r300, + struct r300_vertex_shader* vs) +{ + r300_emit_vertex_program_code(r300, &vs->code, &r300->shader_constants[PIPE_SHADER_VERTEX]); +} + void r300_emit_viewport_state(struct r300_context* r300, struct r300_viewport_state* viewport) { @@ -564,11 +655,9 @@ validate: if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) { if (r300screen->caps->is_r500) { - r500_emit_fragment_shader(r300, - (struct r5xx_fragment_shader*)r300->fs); + r500_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]); } else { - r300_emit_fragment_shader(r300, - (struct r3xx_fragment_shader*)r300->fs); + r300_emit_fragment_program_code(r300, &r300->fs->code, &r300->shader_constants[PIPE_SHADER_FRAGMENT]); } r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER; } |