diff options
Diffstat (limited to 'src/gallium/drivers/r600/r600_shader.c')
| -rw-r--r-- | src/gallium/drivers/r600/r600_shader.c | 1321 |
1 files changed, 507 insertions, 814 deletions
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c index d6455023a3..e7285d624e 100644 --- a/src/gallium/drivers/r600/r600_shader.c +++ b/src/gallium/drivers/r600/r600_shader.c @@ -28,60 +28,12 @@ #include "r600_pipe.h" #include "r600_asm.h" #include "r600_sq.h" +#include "r600_formats.h" #include "r600_opcodes.h" #include "r600d.h" #include <stdio.h> #include <errno.h> -static void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader) -{ - struct r600_pipe_state *rstate = &shader->rstate; - struct r600_shader *rshader = &shader->shader; - unsigned spi_vs_out_id[10]; - unsigned i, tmp; - - /* clear previous register */ - rstate->nregs = 0; - - /* so far never got proper semantic id from tgsi */ - /* FIXME better to move this in config things so they get emited - * only one time per cs - */ - for (i = 0; i < 10; i++) { - spi_vs_out_id[i] = 0; - } - for (i = 0; i < 32; i++) { - tmp = i << ((i & 3) * 8); - spi_vs_out_id[i / 4] |= tmp; - } - for (i = 0; i < 10; i++) { - r600_pipe_state_add_reg(rstate, - R_028614_SPI_VS_OUT_ID_0 + i * 4, - spi_vs_out_id[i], 0xFFFFFFFF, NULL); - } - - r600_pipe_state_add_reg(rstate, - R_0286C4_SPI_VS_OUT_CONFIG, - S_0286C4_VS_EXPORT_COUNT(rshader->noutput - 2), - 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, - R_028868_SQ_PGM_RESOURCES_VS, - S_028868_NUM_GPRS(rshader->bc.ngpr) | - S_028868_STACK_SIZE(rshader->bc.nstack), - 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, - R_0288D0_SQ_PGM_CF_OFFSET_VS, - 0x00000000, 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, - R_028858_SQ_PGM_START_VS, - r600_bo_offset(shader->bo) >> 8, 0xFFFFFFFF, shader->bo); - - r600_pipe_state_add_reg(rstate, - R_03E200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF, - 0xFFFFFFFF, NULL); - -} - int r600_find_vs_semantic_index(struct r600_shader *vs, struct r600_shader *ps, int id) { @@ -96,98 +48,7 @@ int r600_find_vs_semantic_index(struct r600_shader *vs, return 0; } -static void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader) -{ - struct r600_pipe_state *rstate = &shader->rstate; - struct r600_shader *rshader = &shader->shader; - unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1; - int pos_index = -1, face_index = -1; - - rstate->nregs = 0; - - for (i = 0; i < rshader->ninput; i++) { - if (rshader->input[i].name == TGSI_SEMANTIC_POSITION) - pos_index = i; - if (rshader->input[i].name == TGSI_SEMANTIC_FACE) - face_index = i; - } - - for (i = 0; i < rshader->noutput; i++) { - if (rshader->output[i].name == TGSI_SEMANTIC_POSITION) - r600_pipe_state_add_reg(rstate, - R_02880C_DB_SHADER_CONTROL, - S_02880C_Z_EXPORT_ENABLE(1), - S_02880C_Z_EXPORT_ENABLE(1), NULL); - if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL) - r600_pipe_state_add_reg(rstate, - R_02880C_DB_SHADER_CONTROL, - S_02880C_STENCIL_REF_EXPORT_ENABLE(1), - S_02880C_STENCIL_REF_EXPORT_ENABLE(1), NULL); - } - - exports_ps = 0; - num_cout = 0; - for (i = 0; i < rshader->noutput; i++) { - if (rshader->output[i].name == TGSI_SEMANTIC_POSITION || rshader->output[i].name == TGSI_SEMANTIC_STENCIL) - exports_ps |= 1; - else if (rshader->output[i].name == TGSI_SEMANTIC_COLOR) { - num_cout++; - } - } - exports_ps |= S_028854_EXPORT_COLORS(num_cout); - if (!exports_ps) { - /* always at least export 1 component per pixel */ - exports_ps = 2; - } - - spi_ps_in_control_0 = S_0286CC_NUM_INTERP(rshader->ninput) | - S_0286CC_PERSP_GRADIENT_ENA(1); - spi_input_z = 0; - if (pos_index != -1) { - spi_ps_in_control_0 |= (S_0286CC_POSITION_ENA(1) | - S_0286CC_POSITION_CENTROID(rshader->input[pos_index].centroid) | - S_0286CC_POSITION_ADDR(rshader->input[pos_index].gpr) | - S_0286CC_BARYC_SAMPLE_CNTL(1)); - spi_input_z |= 1; - } - - spi_ps_in_control_1 = 0; - if (face_index != -1) { - spi_ps_in_control_1 |= S_0286D0_FRONT_FACE_ENA(1) | - S_0286D0_FRONT_FACE_ADDR(rshader->input[face_index].gpr); - } - - r600_pipe_state_add_reg(rstate, R_0286CC_SPI_PS_IN_CONTROL_0, spi_ps_in_control_0, 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, R_0286D0_SPI_PS_IN_CONTROL_1, spi_ps_in_control_1, 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, R_0286D8_SPI_INPUT_Z, spi_input_z, 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, - R_028840_SQ_PGM_START_PS, - r600_bo_offset(shader->bo) >> 8, 0xFFFFFFFF, shader->bo); - r600_pipe_state_add_reg(rstate, - R_028850_SQ_PGM_RESOURCES_PS, - S_028868_NUM_GPRS(rshader->bc.ngpr) | - S_028868_STACK_SIZE(rshader->bc.nstack), - 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, - R_028854_SQ_PGM_EXPORTS_PS, - exports_ps, 0xFFFFFFFF, NULL); - r600_pipe_state_add_reg(rstate, - R_0288CC_SQ_PGM_CF_OFFSET_PS, - 0x00000000, 0xFFFFFFFF, NULL); - - if (rshader->uses_kill) { - /* only set some bits here, the other bits are set in the dsa state */ - r600_pipe_state_add_reg(rstate, - R_02880C_DB_SHADER_CONTROL, - S_02880C_KILL_ENABLE(1), - S_02880C_KILL_ENABLE(1), NULL); - } - r600_pipe_state_add_reg(rstate, - R_03E200_SQ_LOOP_CONST_0, 0x01000FFF, - 0xFFFFFFFF, NULL); -} - -int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader) +static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader) { struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx; struct r600_shader *rshader = &shader->shader; @@ -225,14 +86,23 @@ int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader) return 0; } -int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader); +static int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader); + int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader, const struct tgsi_token *tokens) { + static int dump_shaders = -1; struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx; int r; -//fprintf(stderr, "--------------------------------------------------------------\n"); -//tgsi_dump(tokens, 0); + /* Would like some magic "get_bool_option_once" routine. + */ + if (dump_shaders == -1) + dump_shaders = debug_get_bool_option("R600_DUMP_SHADERS", FALSE); + + if (dump_shaders) { + fprintf(stderr, "--------------------------------------------------------------\n"); + tgsi_dump(tokens, 0); + } shader->shader.family = r600_get_family(rctx->radeon); r = r600_shader_from_tgsi(tokens, &shader->shader); if (r) { @@ -244,8 +114,10 @@ int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *s R600_ERR("building bytecode failed !\n"); return r; } -//r600_bc_dump(&shader->shader.bc); -//fprintf(stderr, "______________________________________________________________\n"); + if (dump_shaders) { + r600_bc_dump(&shader->shader.bc); + fprintf(stderr, "______________________________________________________________\n"); + } return r600_pipe_shader(ctx, shader); } @@ -262,6 +134,15 @@ void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader */ struct r600_shader_tgsi_instruction; +struct r600_shader_src { + unsigned sel; + unsigned swizzle[4]; + unsigned neg; + unsigned abs; + unsigned rel; + uint32_t value[4]; +}; + struct r600_shader_ctx { struct tgsi_shader_info info; struct tgsi_parse_context parse; @@ -269,10 +150,11 @@ struct r600_shader_ctx { unsigned type; unsigned file_offset[TGSI_FILE_COUNT]; unsigned temp_reg; + unsigned ar_reg; struct r600_shader_tgsi_instruction *inst_info; struct r600_bc *bc; struct r600_shader *shader; - u32 value[4]; + struct r600_shader_src src[3]; u32 *literals; u32 nliterals; u32 max_driver_temp_used; @@ -391,6 +273,7 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx) { struct tgsi_full_declaration *d = &ctx->parse.FullToken.FullDeclaration; unsigned i; + int r; switch (d->Declaration.File) { case TGSI_FILE_INPUT: @@ -422,6 +305,26 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx) case TGSI_FILE_SAMPLER: case TGSI_FILE_ADDRESS: break; + + case TGSI_FILE_SYSTEM_VALUE: + if (d->Semantic.Name == TGSI_SEMANTIC_INSTANCEID) { + struct r600_bc_alu alu; + memset(&alu, 0, sizeof(struct r600_bc_alu)); + + alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT); + alu.src[0].sel = 0; + alu.src[0].chan = 3; + + alu.dst.sel = 0; + alu.dst.chan = 3; + alu.dst.write = 1; + alu.last = 1; + + if ((r = r600_bc_add_alu(ctx->bc, &alu))) + return r; + break; + } + default: R600_ERR("unsupported file %d declaration\n", d->Declaration.File); return -EINVAL; @@ -481,9 +384,187 @@ static int evergreen_gpr_count(struct r600_shader_ctx *ctx) return ctx->num_interp_gpr; } -int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader) +static void tgsi_src(struct r600_shader_ctx *ctx, + const struct tgsi_full_src_register *tgsi_src, + struct r600_shader_src *r600_src) +{ + memset(r600_src, 0, sizeof(*r600_src)); + r600_src->swizzle[0] = tgsi_src->Register.SwizzleX; + r600_src->swizzle[1] = tgsi_src->Register.SwizzleY; + r600_src->swizzle[2] = tgsi_src->Register.SwizzleZ; + r600_src->swizzle[3] = tgsi_src->Register.SwizzleW; + r600_src->neg = tgsi_src->Register.Negate; + r600_src->abs = tgsi_src->Register.Absolute; + + if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) { + int index; + if ((tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleY) && + (tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleZ) && + (tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleW)) { + + index = tgsi_src->Register.Index * 4 + tgsi_src->Register.SwizzleX; + r600_bc_special_constants(ctx->literals[index], &r600_src->sel, &r600_src->neg); + if (r600_src->sel != V_SQ_ALU_SRC_LITERAL) + return; + } + index = tgsi_src->Register.Index; + r600_src->sel = V_SQ_ALU_SRC_LITERAL; + memcpy(r600_src->value, ctx->literals + index * 4, sizeof(r600_src->value)); + } else if (tgsi_src->Register.File == TGSI_FILE_SYSTEM_VALUE) { + /* assume we wan't TGSI_SEMANTIC_INSTANCEID here */ + r600_src->swizzle[0] = 3; + r600_src->swizzle[1] = 3; + r600_src->swizzle[2] = 3; + r600_src->swizzle[3] = 3; + r600_src->sel = 0; + } else { + if (tgsi_src->Register.Indirect) + r600_src->rel = V_SQ_REL_RELATIVE; + r600_src->sel = tgsi_src->Register.Index; + r600_src->sel += ctx->file_offset[tgsi_src->Register.File]; + } +} + +static int tgsi_fetch_rel_const(struct r600_shader_ctx *ctx, unsigned int offset, unsigned int dst_reg) +{ + struct r600_bc_vtx vtx; + unsigned int ar_reg; + int r; + + if (offset) { + struct r600_bc_alu alu; + + memset(&alu, 0, sizeof(alu)); + + alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD_INT); + alu.src[0].sel = ctx->ar_reg; + + alu.src[1].sel = V_SQ_ALU_SRC_LITERAL; + alu.src[1].value = offset; + + alu.dst.sel = dst_reg; + alu.dst.write = 1; + alu.last = 1; + + if ((r = r600_bc_add_alu(ctx->bc, &alu))) + return r; + + ar_reg = dst_reg; + } else { + ar_reg = ctx->ar_reg; + } + + memset(&vtx, 0, sizeof(vtx)); + vtx.fetch_type = 2; /* VTX_FETCH_NO_INDEX_OFFSET */ + vtx.src_gpr = ar_reg; + vtx.mega_fetch_count = 16; + vtx.dst_gpr = dst_reg; + vtx.dst_sel_x = 0; /* SEL_X */ + vtx.dst_sel_y = 1; /* SEL_Y */ + vtx.dst_sel_z = 2; /* SEL_Z */ + vtx.dst_sel_w = 3; /* SEL_W */ + vtx.data_format = FMT_32_32_32_32_FLOAT; + vtx.num_format_all = 2; /* NUM_FORMAT_SCALED */ + vtx.format_comp_all = 1; /* FORMAT_COMP_SIGNED */ + vtx.srf_mode_all = 1; /* SRF_MODE_NO_ZERO */ + + if ((r = r600_bc_add_vtx(ctx->bc, &vtx))) + return r; + + return 0; +} + +static int tgsi_split_constant(struct r600_shader_ctx *ctx) +{ + struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; + struct r600_bc_alu alu; + int i, j, k, nconst, r; + + for (i = 0, nconst = 0; i < inst->Instruction.NumSrcRegs; i++) { + if (inst->Src[i].Register.File == TGSI_FILE_CONSTANT) { + nconst++; + } + tgsi_src(ctx, &inst->Src[i], &ctx->src[i]); + } + for (i = 0, j = nconst - 1; i < inst->Instruction.NumSrcRegs; i++) { + if (inst->Src[i].Register.File != TGSI_FILE_CONSTANT) { + continue; + } + + if (ctx->src[i].rel) { + int treg = r600_get_temp(ctx); + if ((r = tgsi_fetch_rel_const(ctx, ctx->src[i].sel - 512, treg))) + return r; + + ctx->src[i].sel = treg; + ctx->src[i].rel = 0; + j--; + } else if (j > 0) { + int treg = r600_get_temp(ctx); + for (k = 0; k < 4; k++) { + memset(&alu, 0, sizeof(struct r600_bc_alu)); + alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); + alu.src[0].sel = ctx->src[i].sel; + alu.src[0].chan = k; + alu.src[0].rel = ctx->src[i].rel; + alu.dst.sel = treg; + alu.dst.chan = k; + alu.dst.write = 1; + if (k == 3) + alu.last = 1; + r = r600_bc_add_alu(ctx->bc, &alu); + if (r) + return r; + } + ctx->src[i].sel = treg; + ctx->src[i].rel =0; + j--; + } + } + return 0; +} + +/* need to move any immediate into a temp - for trig functions which use literal for PI stuff */ +static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx) +{ + struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; + struct r600_bc_alu alu; + int i, j, k, nliteral, r; + + for (i = 0, nliteral = 0; i < inst->Instruction.NumSrcRegs; i++) { + if (ctx->src[i].sel == V_SQ_ALU_SRC_LITERAL) { + nliteral++; + } + } + for (i = 0, j = nliteral - 1; i < inst->Instruction.NumSrcRegs; i++) { + if (j > 0 && ctx->src[i].sel == V_SQ_ALU_SRC_LITERAL) { + int treg = r600_get_temp(ctx); + for (k = 0; k < 4; k++) { + memset(&alu, 0, sizeof(struct r600_bc_alu)); + alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); + alu.src[0].sel = ctx->src[i].sel; + alu.src[0].chan = k; + alu.src[0].value = ctx->src[i].value[k]; + alu.dst.sel = treg; + alu.dst.chan = k; + alu.dst.write = 1; + if (k == 3) + alu.last = 1; + r = r600_bc_add_alu(ctx->bc, &alu); + if (r) + return r; + } + ctx->src[i].sel = treg; + j--; + } + } + return 0; +} + +static int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader) { struct tgsi_full_immediate *immediate; + struct tgsi_full_property *property; struct r600_shader_ctx ctx; struct r600_bc_output output[32]; unsigned output_done, noutput; @@ -506,7 +587,9 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s /* Values [0,127] correspond to GPR[0..127]. * Values [128,159] correspond to constant buffer bank 0 * Values [160,191] correspond to constant buffer bank 1 - * Values [256,511] correspond to cfile constants c[0..255]. + * Values [256,511] correspond to cfile constants c[0..255]. (Gone on EG) + * Values [256,287] correspond to constant buffer bank 2 (EG) + * Values [288,319] correspond to constant buffer bank 3 (EG) * Other special values are shown in the list below. * 244 ALU_SRC_1_DBL_L: special constant 1.0 double-float, LSW. (RV670+) * 245 ALU_SRC_1_DBL_M: special constant 1.0 double-float, MSW. (RV670+) @@ -540,15 +623,18 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s ctx.file_offset[TGSI_FILE_TEMPORARY] = ctx.file_offset[TGSI_FILE_OUTPUT] + ctx.info.file_count[TGSI_FILE_OUTPUT]; - ctx.file_offset[TGSI_FILE_CONSTANT] = 128; + /* Outside the GPR range. This will be translated to one of the + * kcache banks later. */ + ctx.file_offset[TGSI_FILE_CONSTANT] = 512; - ctx.file_offset[TGSI_FILE_IMMEDIATE] = 253; - ctx.temp_reg = ctx.file_offset[TGSI_FILE_TEMPORARY] + + ctx.file_offset[TGSI_FILE_IMMEDIATE] = V_SQ_ALU_SRC_LITERAL; + ctx.ar_reg = ctx.file_offset[TGSI_FILE_TEMPORARY] + ctx.info.file_count[TGSI_FILE_TEMPORARY]; + ctx.temp_reg = ctx.ar_reg + 1; ctx.nliterals = 0; ctx.literals = NULL; - + shader->fs_write_all = FALSE; while (!tgsi_parse_end_of_tokens(&ctx.parse)) { tgsi_parse_token(&ctx.parse); switch (ctx.parse.FullToken.Token.Type) { @@ -577,7 +663,12 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s ctx.max_driver_temp_used = 0; /* reserve first tmp for everyone */ r600_get_temp(&ctx); + opcode = ctx.parse.FullToken.FullInstruction.Instruction.Opcode; + if ((r = tgsi_split_constant(&ctx))) + goto out_err; + if ((r = tgsi_split_literal_constant(&ctx))) + goto out_err; if (ctx.bc->chiprev == CHIPREV_EVERGREEN) ctx.inst_info = &eg_shader_tgsi_instruction[opcode]; else @@ -585,9 +676,13 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s r = ctx.inst_info->process(&ctx); if (r) goto out_err; - r = r600_bc_add_literal(ctx.bc, ctx.value); - if (r) - goto out_err; + break; + case TGSI_TOKEN_TYPE_PROPERTY: + property = &ctx.parse.FullToken.FullProperty; + if (property->Property.PropertyName == TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS) { + if (property->u[0].Data == 1) + shader->fs_write_all = TRUE; + } break; default: R600_ERR("unsupported token type %d\n", ctx.parse.FullToken.Token.Type); @@ -605,6 +700,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s output[i].swizzle_y = 1; output[i].swizzle_z = 2; output[i].swizzle_w = 3; + output[i].burst_count = 1; output[i].barrier = 1; output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM; output[i].array_base = i - pos0; @@ -668,6 +764,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s output[i].swizzle_y = 1; output[i].swizzle_z = 2; output[i].swizzle_w = 3; + output[i].burst_count = 1; output[i].barrier = 1; output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM; output[i].array_base = 0; @@ -684,6 +781,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s output[0].swizzle_y = 7; output[0].swizzle_z = 7; output[0].swizzle_w = 7; + output[0].burst_count = 1; output[0].barrier = 1; output[0].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL; output[0].array_base = 0; @@ -726,33 +824,22 @@ static int tgsi_end(struct r600_shader_ctx *ctx) return 0; } -static int tgsi_src(struct r600_shader_ctx *ctx, - const struct tgsi_full_src_register *tgsi_src, - struct r600_bc_alu_src *r600_src) +static void r600_bc_src(struct r600_bc_alu_src *bc_src, + const struct r600_shader_src *shader_src, + unsigned chan) { - int index; - memset(r600_src, 0, sizeof(struct r600_bc_alu_src)); - r600_src->sel = tgsi_src->Register.Index; - if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) { - r600_src->sel = 0; - index = tgsi_src->Register.Index; - ctx->value[0] = ctx->literals[index * 4 + 0]; - ctx->value[1] = ctx->literals[index * 4 + 1]; - ctx->value[2] = ctx->literals[index * 4 + 2]; - ctx->value[3] = ctx->literals[index * 4 + 3]; - } - if (tgsi_src->Register.Indirect) - r600_src->rel = V_SQ_REL_RELATIVE; - r600_src->neg = tgsi_src->Register.Negate; - r600_src->abs = tgsi_src->Register.Absolute; - r600_src->sel += ctx->file_offset[tgsi_src->Register.File]; - return 0; + bc_src->sel = shader_src->sel; + bc_src->chan = shader_src->swizzle[chan]; + bc_src->neg = shader_src->neg; + bc_src->abs = shader_src->abs; + bc_src->rel = shader_src->rel; + bc_src->value = shader_src->value[bc_src->chan]; } -static int tgsi_dst(struct r600_shader_ctx *ctx, - const struct tgsi_full_dst_register *tgsi_dst, - unsigned swizzle, - struct r600_bc_alu_dst *r600_dst) +static void tgsi_dst(struct r600_shader_ctx *ctx, + const struct tgsi_full_dst_register *tgsi_dst, + unsigned swizzle, + struct r600_bc_alu_dst *r600_dst) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; @@ -765,146 +852,42 @@ static int tgsi_dst(struct r600_shader_ctx *ctx, if (inst->Instruction.Saturate) { r600_dst->clamp = 1; } - return 0; } -static unsigned tgsi_chan(const struct tgsi_full_src_register *tgsi_src, unsigned swizzle) +static int tgsi_last_instruction(unsigned writemask) { - switch (swizzle) { - case 0: - return tgsi_src->Register.SwizzleX; - case 1: - return tgsi_src->Register.SwizzleY; - case 2: - return tgsi_src->Register.SwizzleZ; - case 3: - return tgsi_src->Register.SwizzleW; - default: - return 0; - } -} + int i, lasti = 0; -static int tgsi_split_constant(struct r600_shader_ctx *ctx, struct r600_bc_alu_src r600_src[3]) -{ - struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu alu; - int i, j, k, nconst, r; - - for (i = 0, nconst = 0; i < inst->Instruction.NumSrcRegs; i++) { - if (inst->Src[i].Register.File == TGSI_FILE_CONSTANT) { - nconst++; - } - r = tgsi_src(ctx, &inst->Src[i], &r600_src[i]); - if (r) { - return r; - } - } - for (i = 0, j = nconst - 1; i < inst->Instruction.NumSrcRegs; i++) { - if (j > 0 && inst->Src[i].Register.File == TGSI_FILE_CONSTANT) { - int treg = r600_get_temp(ctx); - for (k = 0; k < 4; k++) { - memset(&alu, 0, sizeof(struct r600_bc_alu)); - alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); - alu.src[0].sel = r600_src[i].sel; - alu.src[0].chan = k; - alu.src[0].rel = r600_src[i].rel; - alu.dst.sel = treg; - alu.dst.chan = k; - alu.dst.write = 1; - if (k == 3) - alu.last = 1; - r = r600_bc_add_alu(ctx->bc, &alu); - if (r) - return r; - } - r600_src[i].sel = treg; - r600_src[i].rel =0; - j--; - } - } - return 0; -} - -/* need to move any immediate into a temp - for trig functions which use literal for PI stuff */ -static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_bc_alu_src r600_src[3]) -{ - struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu alu; - int i, j, k, nliteral, r; - - for (i = 0, nliteral = 0; i < inst->Instruction.NumSrcRegs; i++) { - if (inst->Src[i].Register.File == TGSI_FILE_IMMEDIATE) { - nliteral++; - } - } - for (i = 0, j = nliteral - 1; i < inst->Instruction.NumSrcRegs; i++) { - if (j > 0 && inst->Src[i].Register.File == TGSI_FILE_IMMEDIATE) { - int treg = r600_get_temp(ctx); - for (k = 0; k < 4; k++) { - memset(&alu, 0, sizeof(struct r600_bc_alu)); - alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); - alu.src[0].sel = r600_src[i].sel; - alu.src[0].chan = k; - alu.dst.sel = treg; - alu.dst.chan = k; - alu.dst.write = 1; - if (k == 3) - alu.last = 1; - r = r600_bc_add_alu(ctx->bc, &alu); - if (r) - return r; - } - r = r600_bc_add_literal(ctx->bc, &ctx->literals[inst->Src[i].Register.Index * 4]); - if (r) - return r; - r600_src[i].sel = treg; - j--; + for (i = 0; i < 4; i++) { + if (writemask & (1 << i)) { + lasti = i; } } - return 0; + return lasti; } static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; int i, j, r; - int lasti = 0; + int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask); - for (i = 0; i < 4; i++) { - if (inst->Dst[0].Register.WriteMask & (1 << i)) { - lasti = i; - } - } - - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; for (i = 0; i < lasti + 1; i++) { if (!(inst->Dst[0].Register.WriteMask & (1 << i))) continue; memset(&alu, 0, sizeof(struct r600_bc_alu)); - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.inst = ctx->inst_info->r600_opcode; if (!swap) { for (j = 0; j < inst->Instruction.NumSrcRegs; j++) { - alu.src[j] = r600_src[j]; - alu.src[j].chan = tgsi_chan(&inst->Src[j], i); + r600_bc_src(&alu.src[j], &ctx->src[j], i); } } else { - alu.src[0] = r600_src[1]; - alu.src[0].chan = tgsi_chan(&inst->Src[1], i); - - alu.src[1] = r600_src[0]; - alu.src[1].chan = tgsi_chan(&inst->Src[0], i); + r600_bc_src(&alu.src[0], &ctx->src[1], i); + r600_bc_src(&alu.src[1], &ctx->src[0], i); } /* handle some special cases */ switch (ctx->inst_info->tgsi_opcode) { @@ -942,25 +925,15 @@ static int tgsi_op2_swap(struct r600_shader_ctx *ctx) * r700 - normalize by dividing by 2PI * see fdo bug 27901 */ -static int tgsi_setup_trig(struct r600_shader_ctx *ctx, - struct r600_bc_alu_src r600_src[3]) +static int tgsi_setup_trig(struct r600_shader_ctx *ctx) { - struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; + static float half_inv_pi = 1.0 /(3.1415926535 * 2); + static float double_pi = 3.1415926535 * 2; + static float neg_pi = -3.1415926535; + int r; - uint32_t lit_vals[4]; struct r600_bc_alu alu; - memset(lit_vals, 0, 4*4); - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; - - lit_vals[0] = fui(1.0 /(3.1415926535 * 2)); - lit_vals[1] = fui(0.5f); - memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD); alu.is_op3 = 1; @@ -969,20 +942,17 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx, alu.dst.sel = ctx->temp_reg; alu.dst.write = 1; - alu.src[0] = r600_src[0]; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.src[1].sel = V_SQ_ALU_SRC_LITERAL; alu.src[1].chan = 0; - alu.src[2].sel = V_SQ_ALU_SRC_LITERAL; - alu.src[2].chan = 1; + alu.src[1].value = *(uint32_t *)&half_inv_pi; + alu.src[2].sel = V_SQ_ALU_SRC_0_5; + alu.src[2].chan = 0; alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, lit_vals); - if (r) - return r; memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT); @@ -998,14 +968,6 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx, if (r) return r; - if (ctx->bc->chiprev == CHIPREV_R600) { - lit_vals[0] = fui(3.1415926535897f * 2.0f); - lit_vals[1] = fui(-3.1415926535897f); - } else { - lit_vals[0] = fui(1.0f); - lit_vals[1] = fui(-0.5f); - } - memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD); alu.is_op3 = 1; @@ -1020,26 +982,32 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx, alu.src[1].sel = V_SQ_ALU_SRC_LITERAL; alu.src[1].chan = 0; alu.src[2].sel = V_SQ_ALU_SRC_LITERAL; - alu.src[2].chan = 1; + alu.src[2].chan = 0; + + if (ctx->bc->chiprev == CHIPREV_R600) { + alu.src[1].value = *(uint32_t *)&double_pi; + alu.src[2].value = *(uint32_t *)&neg_pi; + } else { + alu.src[1].sel = V_SQ_ALU_SRC_1; + alu.src[2].sel = V_SQ_ALU_SRC_0_5; + alu.src[2].neg = 1; + } + alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, lit_vals); - if (r) - return r; return 0; } static int tgsi_trig(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; int i, r; - int lasti = 0; + int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask); - r = tgsi_setup_trig(ctx, r600_src); + r = tgsi_setup_trig(ctx); if (r) return r; @@ -1057,10 +1025,6 @@ static int tgsi_trig(struct r600_shader_ctx *ctx) return r; /* replicate result */ - for (i = 0; i < 4; i++) { - if (inst->Dst[0].Register.WriteMask & (1 << i)) - lasti = i; - } for (i = 0; i < lasti + 1; i++) { if (!(inst->Dst[0].Register.WriteMask & (1 << i))) continue; @@ -1069,9 +1033,7 @@ static int tgsi_trig(struct r600_shader_ctx *ctx) alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); alu.src[0].sel = ctx->temp_reg; - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); if (i == lasti) alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); @@ -1084,7 +1046,6 @@ static int tgsi_trig(struct r600_shader_ctx *ctx) static int tgsi_scs(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; int r; @@ -1092,7 +1053,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) * X or Y components of the destination vector. */ if (likely(inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_XY)) { - r = tgsi_setup_trig(ctx, r600_src); + r = tgsi_setup_trig(ctx); if (r) return r; } @@ -1101,9 +1062,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS); - r = tgsi_dst(ctx, &inst->Dst[0], 0, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 0, &alu.dst); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = 0; @@ -1117,9 +1076,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN); - r = tgsi_dst(ctx, &inst->Dst[0], 1, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 1, &alu.dst); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = 0; @@ -1135,9 +1092,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); - r = tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst); alu.src[0].sel = V_SQ_ALU_SRC_0; alu.src[0].chan = 0; @@ -1147,10 +1102,6 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* dst.w = 1.0; */ @@ -1159,9 +1110,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); - r = tgsi_dst(ctx, &inst->Dst[0], 3, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 3, &alu.dst); alu.src[0].sel = V_SQ_ALU_SRC_1; alu.src[0].chan = 0; @@ -1171,10 +1120,6 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } return 0; @@ -1182,7 +1127,6 @@ static int tgsi_scs(struct r600_shader_ctx *ctx) static int tgsi_kill(struct r600_shader_ctx *ctx) { - struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; int i, r; @@ -1198,10 +1142,7 @@ static int tgsi_kill(struct r600_shader_ctx *ctx) alu.src[1].sel = V_SQ_ALU_SRC_1; alu.src[1].neg = 1; } else { - r = tgsi_src(ctx, &inst->Src[0], &alu.src[1]); - if (r) - return r; - alu.src[1].chan = tgsi_chan(&inst->Src[0], i); + r600_bc_src(&alu.src[1], &ctx->src[0], i); } if (i == 3) { alu.last = 1; @@ -1210,9 +1151,6 @@ static int tgsi_kill(struct r600_shader_ctx *ctx) if (r) return r; } - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* kill must be last in ALU */ ctx->bc->force_add_cf = 1; @@ -1224,24 +1162,14 @@ static int tgsi_lit(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; - struct r600_bc_alu_src r600_src[3]; int r; - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; - /* dst.x, <- 1.0 */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); alu.src[0].sel = V_SQ_ALU_SRC_1; /*1.0*/ alu.src[0].chan = 0; - r = tgsi_dst(ctx, &inst->Dst[0], 0, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 0, &alu.dst); alu.dst.write = (inst->Dst[0].Register.WriteMask >> 0) & 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) @@ -1250,12 +1178,10 @@ static int tgsi_lit(struct r600_shader_ctx *ctx) /* dst.y = max(src.x, 0.0) */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX); - alu.src[0] = r600_src[0]; + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.src[1].sel = V_SQ_ALU_SRC_0; /*0.0*/ alu.src[1].chan = 0; - r = tgsi_dst(ctx, &inst->Dst[0], 1, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 1, &alu.dst); alu.dst.write = (inst->Dst[0].Register.WriteMask >> 1) & 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) @@ -1266,19 +1192,13 @@ static int tgsi_lit(struct r600_shader_ctx *ctx) alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); alu.src[0].sel = V_SQ_ALU_SRC_1; alu.src[0].chan = 0; - r = tgsi_dst(ctx, &inst->Dst[0], 3, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 3, &alu.dst); alu.dst.write = (inst->Dst[0].Register.WriteMask >> 3) & 1; alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - if (inst->Dst[0].Register.WriteMask & (1 << 2)) { int chan; @@ -1287,33 +1207,24 @@ static int tgsi_lit(struct r600_shader_ctx *ctx) /* dst.z = log(src.y) */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED); - alu.src[0] = r600_src[0]; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 1); - r = tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst); - if (r) - return r; + r600_bc_src(&alu.src[0], &ctx->src[0], 1); + tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst); alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - chan = alu.dst.chan; sel = alu.dst.sel; /* tmp.x = amd MUL_LIT(src.w, dst.z, src.x ) */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MUL_LIT); - alu.src[0] = r600_src[0]; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 3); + r600_bc_src(&alu.src[0], &ctx->src[0], 3); alu.src[1].sel = sel; alu.src[1].chan = chan; - alu.src[2] = r600_src[0]; - alu.src[2].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[2], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.chan = 0; alu.dst.write = 1; @@ -1323,17 +1234,12 @@ static int tgsi_lit(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* dst.z = exp(tmp.x) */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = 0; - r = tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst); alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) @@ -1357,10 +1263,7 @@ static int tgsi_rsq(struct r600_shader_ctx *ctx) alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIPSQRT_CLAMPED); for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { - r = tgsi_src(ctx, &inst->Src[i], &alu.src[i]); - if (r) - return r; - alu.src[i].chan = tgsi_chan(&inst->Src[i], 0); + r600_bc_src(&alu.src[i], &ctx->src[i], 0); alu.src[i].abs = 1; } alu.dst.sel = ctx->temp_reg; @@ -1369,9 +1272,6 @@ static int tgsi_rsq(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* replicate result */ return tgsi_helper_tempx_replicate(ctx); } @@ -1387,9 +1287,7 @@ static int tgsi_helper_tempx_replicate(struct r600_shader_ctx *ctx) alu.src[0].sel = ctx->temp_reg; alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); alu.dst.chan = i; - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.dst.write = (inst->Dst[0].Register.WriteMask >> i) & 1; if (i == 3) alu.last = 1; @@ -1409,10 +1307,7 @@ static int tgsi_trans_srcx_replicate(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = ctx->inst_info->r600_opcode; for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { - r = tgsi_src(ctx, &inst->Src[i], &alu.src[i]); - if (r) - return r; - alu.src[i].chan = tgsi_chan(&inst->Src[i], 0); + r600_bc_src(&alu.src[i], &ctx->src[i], 0); } alu.dst.sel = ctx->temp_reg; alu.dst.write = 1; @@ -1420,42 +1315,29 @@ static int tgsi_trans_srcx_replicate(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* replicate result */ return tgsi_helper_tempx_replicate(ctx); } static int tgsi_pow(struct r600_shader_ctx *ctx) { - struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; int r; /* LOG2(a) */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.write = 1; alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc,ctx->value); - if (r) - return r; /* b * LOG2(a) */ memset(&alu, 0, sizeof(struct r600_bc_alu)); - alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL_IEEE); - r = tgsi_src(ctx, &inst->Src[1], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[1], 0); + alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL); + r600_bc_src(&alu.src[0], &ctx->src[1], 0); alu.src[1].sel = ctx->temp_reg; alu.dst.sel = ctx->temp_reg; alu.dst.write = 1; @@ -1463,9 +1345,6 @@ static int tgsi_pow(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc,ctx->value); - if (r) - return r; /* POW(a,b) = EXP2(b * LOG2(a))*/ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE); @@ -1476,9 +1355,6 @@ static int tgsi_pow(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc,ctx->value); - if (r) - return r; return tgsi_helper_tempx_replicate(ctx); } @@ -1486,16 +1362,8 @@ static int tgsi_ssg(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; - struct r600_bc_alu_src r600_src[3]; int i, r; - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; - /* tmp = (src > 0 ? 1 : src) */ for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); @@ -1505,31 +1373,23 @@ static int tgsi_ssg(struct r600_shader_ctx *ctx) alu.dst.sel = ctx->temp_reg; alu.dst.chan = i; - alu.src[0] = r600_src[0]; - alu.src[0].chan = tgsi_chan(&inst->Src[0], i); - + r600_bc_src(&alu.src[0], &ctx->src[0], i); alu.src[1].sel = V_SQ_ALU_SRC_1; + r600_bc_src(&alu.src[2], &ctx->src[0], i); - alu.src[2] = r600_src[0]; - alu.src[2].chan = tgsi_chan(&inst->Src[0], i); if (i == 3) alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; } - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* dst = (-tmp > 0 ? -1 : tmp) */ for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGT); alu.is_op3 = 1; - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = i; @@ -1555,9 +1415,6 @@ static int tgsi_helper_copy(struct r600_shader_ctx *ctx, struct tgsi_full_instru struct r600_bc_alu alu; int i, r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); if (!(inst->Dst[0].Register.WriteMask & (1 << i))) { @@ -1565,9 +1422,7 @@ static int tgsi_helper_copy(struct r600_shader_ctx *ctx, struct tgsi_full_instru alu.dst.chan = i; } else { alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = i; } @@ -1584,61 +1439,50 @@ static int tgsi_helper_copy(struct r600_shader_ctx *ctx, struct tgsi_full_instru static int tgsi_op3(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; int i, j, r; + int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask); + + for (i = 0; i < lasti + 1; i++) { + if (!(inst->Dst[0].Register.WriteMask & (1 << i))) + continue; - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; - /* do it in 2 step as op3 doesn't support writemask */ - for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = ctx->inst_info->r600_opcode; for (j = 0; j < inst->Instruction.NumSrcRegs; j++) { - alu.src[j] = r600_src[j]; - alu.src[j].chan = tgsi_chan(&inst->Src[j], i); + r600_bc_src(&alu.src[j], &ctx->src[j], i); } - alu.dst.sel = ctx->temp_reg; + + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.dst.chan = i; alu.dst.write = 1; alu.is_op3 = 1; - if (i == 3) { + if (i == lasti) { alu.last = 1; } r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; } - return tgsi_helper_copy(ctx, inst); + return 0; } static int tgsi_dp(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; int i, j, r; - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = ctx->inst_info->r600_opcode; for (j = 0; j < inst->Instruction.NumSrcRegs; j++) { - alu.src[j] = r600_src[j]; - alu.src[j].chan = tgsi_chan(&inst->Src[j], i); + r600_bc_src(&alu.src[j], &ctx->src[j], i); } - alu.dst.sel = ctx->temp_reg; + + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.dst.chan = i; - alu.dst.write = 1; + alu.dst.write = (inst->Dst[0].Register.WriteMask >> i) & 1; /* handle some special cases */ switch (ctx->inst_info->tgsi_opcode) { case TGSI_OPCODE_DP2: @@ -1670,19 +1514,21 @@ static int tgsi_dp(struct r600_shader_ctx *ctx) if (r) return r; } - return tgsi_helper_copy(ctx, inst); + return 0; } static int tgsi_tex(struct r600_shader_ctx *ctx) { + static float one_point_five = 1.5f; struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_tex tex; struct r600_bc_alu alu; unsigned src_gpr; int r, i; int opcode; - boolean src_not_temp = inst->Src[0].Register.File != TGSI_FILE_TEMPORARY; - uint32_t lit_vals[4]; + boolean src_not_temp = + inst->Src[0].Register.File != TGSI_FILE_TEMPORARY && + inst->Src[0].Register.File != TGSI_FILE_INPUT; src_gpr = ctx->file_offset[inst->Src[0].Register.File] + inst->Src[0].Register.Index; @@ -1690,11 +1536,8 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) /* Add perspective divide */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; + r600_bc_src(&alu.src[0], &ctx->src[0], 3); - alu.src[0].chan = tgsi_chan(&inst->Src[0], 3); alu.dst.sel = ctx->temp_reg; alu.dst.chan = 3; alu.last = 1; @@ -1708,10 +1551,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = 3; - r = tgsi_src(ctx, &inst->Src[0], &alu.src[1]); - if (r) - return r; - alu.src[1].chan = tgsi_chan(&inst->Src[0], i); + r600_bc_src(&alu.src[1], &ctx->src[0], i); alu.dst.sel = ctx->temp_reg; alu.dst.chan = i; alu.dst.write = 1; @@ -1735,43 +1575,15 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) } if (inst->Texture.Texture == TGSI_TEXTURE_CUBE) { - int src_chan, src2_chan; + static const unsigned src0_swizzle[] = {2, 2, 0, 1}; + static const unsigned src1_swizzle[] = {1, 0, 2, 2}; /* tmp1.xyzw = CUBE(R0.zzxy, R0.yxzz) */ for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE); - switch (i) { - case 0: - src_chan = 2; - src2_chan = 1; - break; - case 1: - src_chan = 2; - src2_chan = 0; - break; - case 2: - src_chan = 0; - src2_chan = 2; - break; - case 3: - src_chan = 1; - src2_chan = 2; - break; - default: - assert(0); - src_chan = 0; - src2_chan = 0; - break; - } - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], src_chan); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[1]); - if (r) - return r; - alu.src[1].chan = tgsi_chan(&inst->Src[0], src2_chan); + r600_bc_src(&alu.src[0], &ctx->src[0], src0_swizzle[i]); + r600_bc_src(&alu.src[1], &ctx->src[0], src1_swizzle[i]); alu.dst.sel = ctx->temp_reg; alu.dst.chan = i; if (i == 3) @@ -1811,6 +1623,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) alu.src[2].sel = V_SQ_ALU_SRC_LITERAL; alu.src[2].chan = 0; + alu.src[2].value = *(uint32_t *)&one_point_five; alu.dst.sel = ctx->temp_reg; alu.dst.chan = 0; @@ -1831,6 +1644,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) alu.src[2].sel = V_SQ_ALU_SRC_LITERAL; alu.src[2].chan = 0; + alu.src[2].value = *(uint32_t *)&one_point_five; alu.dst.sel = ctx->temp_reg; alu.dst.chan = 1; @@ -1841,11 +1655,6 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) if (r) return r; - lit_vals[0] = fui(1.5f); - - r = r600_bc_add_literal(ctx->bc, lit_vals); - if (r) - return r; src_not_temp = FALSE; src_gpr = ctx->temp_reg; } @@ -1854,8 +1663,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV); - alu.src[0].sel = src_gpr; - alu.src[0].chan = tgsi_chan(&inst->Src[0], i); + r600_bc_src(&alu.src[0], &ctx->src[0], i); alu.dst.sel = ctx->temp_reg; alu.dst.chan = i; if (i == 3) @@ -1876,7 +1684,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) memset(&tex, 0, sizeof(struct r600_bc_tex)); tex.inst = opcode; tex.sampler_id = ctx->file_offset[inst->Src[1].Register.File] + inst->Src[1].Register.Index; - tex.resource_id = tex.sampler_id; + tex.resource_id = tex.sampler_id + R600_MAX_CONST_BUFFERS; tex.src_gpr = src_gpr; tex.dst_gpr = ctx->file_offset[inst->Dst[0].Register.File] + inst->Dst[0].Register.Index; tex.dst_sel_x = (inst->Dst[0].Register.WriteMask & 1) ? 0 : 7; @@ -1902,6 +1710,12 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) tex.coord_type_w = 1; } + if (inst->Texture.Texture == TGSI_TEXTURE_1D_ARRAY) { + tex.coord_type_z = 0; + tex.src_sel_z = 1; + } else if (inst->Texture.Texture == TGSI_TEXTURE_2D_ARRAY) + tex.coord_type_z = 0; + if (inst->Texture.Texture == TGSI_TEXTURE_SHADOW1D || inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D) tex.src_sel_w = 2; @@ -1916,29 +1730,48 @@ static int tgsi_tex(struct r600_shader_ctx *ctx) static int tgsi_lrp(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; + int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask); unsigned i; int r; - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; + /* optimize if it's just an equal balance */ + if (ctx->src[0].sel == V_SQ_ALU_SRC_0_5) { + for (i = 0; i < lasti + 1; i++) { + if (!(inst->Dst[0].Register.WriteMask & (1 << i))) + continue; + + memset(&alu, 0, sizeof(struct r600_bc_alu)); + alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD); + r600_bc_src(&alu.src[0], &ctx->src[1], i); + r600_bc_src(&alu.src[1], &ctx->src[2], i); + alu.omod = 3; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); + alu.dst.chan = i; + if (i == lasti) { + alu.last = 1; + } + r = r600_bc_add_alu(ctx->bc, &alu); + if (r) + return r; + } + return 0; + } + /* 1 - src0 */ - for (i = 0; i < 4; i++) { + for (i = 0; i < lasti + 1; i++) { + if (!(inst->Dst[0].Register.WriteMask & (1 << i))) + continue; + memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD); alu.src[0].sel = V_SQ_ALU_SRC_1; alu.src[0].chan = 0; - alu.src[1] = r600_src[0]; - alu.src[1].chan = tgsi_chan(&inst->Src[0], i); + r600_bc_src(&alu.src[1], &ctx->src[0], i); alu.src[1].neg = 1; alu.dst.sel = ctx->temp_reg; alu.dst.chan = i; - if (i == 3) { + if (i == lasti) { alu.last = 1; } alu.dst.write = 1; @@ -1946,21 +1779,20 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx) if (r) return r; } - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* (1 - src0) * src2 */ - for (i = 0; i < 4; i++) { + for (i = 0; i < lasti + 1; i++) { + if (!(inst->Dst[0].Register.WriteMask & (1 << i))) + continue; + memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = i; - alu.src[1] = r600_src[2]; - alu.src[1].chan = tgsi_chan(&inst->Src[2], i); + r600_bc_src(&alu.src[1], &ctx->src[2], i); alu.dst.sel = ctx->temp_reg; alu.dst.chan = i; - if (i == 3) { + if (i == lasti) { alu.last = 1; } alu.dst.write = 1; @@ -1968,88 +1800,66 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx) if (r) return r; } - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; /* src0 * src1 + (1 - src0) * src2 */ - for (i = 0; i < 4; i++) { + for (i = 0; i < lasti + 1; i++) { + if (!(inst->Dst[0].Register.WriteMask & (1 << i))) + continue; + memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD); alu.is_op3 = 1; - alu.src[0] = r600_src[0]; - alu.src[0].chan = tgsi_chan(&inst->Src[0], i); - alu.src[1] = r600_src[1]; - alu.src[1].chan = tgsi_chan(&inst->Src[1], i); + r600_bc_src(&alu.src[0], &ctx->src[0], i); + r600_bc_src(&alu.src[1], &ctx->src[1], i); alu.src[2].sel = ctx->temp_reg; alu.src[2].chan = i; - alu.dst.sel = ctx->temp_reg; + + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.dst.chan = i; - if (i == 3) { + if (i == lasti) { alu.last = 1; } r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; } - return tgsi_helper_copy(ctx, inst); + return 0; } static int tgsi_cmp(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; struct r600_bc_alu alu; - int use_temp = 0; int i, r; + int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask); - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; - - if (inst->Dst[0].Register.WriteMask != 0xf) - use_temp = 1; + for (i = 0; i < lasti + 1; i++) { + if (!(inst->Dst[0].Register.WriteMask & (1 << i))) + continue; - for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE); - alu.src[0] = r600_src[0]; - alu.src[0].chan = tgsi_chan(&inst->Src[0], i); - - alu.src[1] = r600_src[2]; - alu.src[1].chan = tgsi_chan(&inst->Src[2], i); - - alu.src[2] = r600_src[1]; - alu.src[2].chan = tgsi_chan(&inst->Src[1], i); - - if (use_temp) - alu.dst.sel = ctx->temp_reg; - else { - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; - } + r600_bc_src(&alu.src[0], &ctx->src[0], i); + r600_bc_src(&alu.src[1], &ctx->src[2], i); + r600_bc_src(&alu.src[2], &ctx->src[1], i); + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.dst.chan = i; alu.dst.write = 1; alu.is_op3 = 1; - if (i == 3) + if (i == lasti) alu.last = 1; r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; } - if (use_temp) - return tgsi_helper_copy(ctx, inst); return 0; } static int tgsi_xpd(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3]; + static const unsigned int src0_swizzle[] = {2, 0, 1}; + static const unsigned int src1_swizzle[] = {1, 2, 0}; struct r600_bc_alu alu; uint32_t use_temp = 0; int i, r; @@ -2057,45 +1867,15 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx) if (inst->Dst[0].Register.WriteMask != 0xf) use_temp = 1; - r = tgsi_split_constant(ctx, r600_src); - if (r) - return r; - r = tgsi_split_literal_constant(ctx, r600_src); - if (r) - return r; - for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL); - - alu.src[0] = r600_src[0]; - switch (i) { - case 0: - alu.src[0].chan = tgsi_chan(&inst->Src[0], 2); - break; - case 1: - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); - break; - case 2: - alu.src[0].chan = tgsi_chan(&inst->Src[0], 1); - break; - case 3: + if (i < 3) { + r600_bc_src(&alu.src[0], &ctx->src[0], src0_swizzle[i]); + r600_bc_src(&alu.src[1], &ctx->src[1], src1_swizzle[i]); + } else { alu.src[0].sel = V_SQ_ALU_SRC_0; alu.src[0].chan = i; - } - - alu.src[1] = r600_src[1]; - switch (i) { - case 0: - alu.src[1].chan = tgsi_chan(&inst->Src[1], 1); - break; - case 1: - alu.src[1].chan = tgsi_chan(&inst->Src[1], 2); - break; - case 2: - alu.src[1].chan = tgsi_chan(&inst->Src[1], 0); - break; - case 3: alu.src[1].sel = V_SQ_ALU_SRC_0; alu.src[1].chan = i; } @@ -2109,44 +1889,18 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } for (i = 0; i < 4; i++) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD); - alu.src[0] = r600_src[0]; - switch (i) { - case 0: - alu.src[0].chan = tgsi_chan(&inst->Src[0], 1); - break; - case 1: - alu.src[0].chan = tgsi_chan(&inst->Src[0], 2); - break; - case 2: - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); - break; - case 3: + if (i < 3) { + r600_bc_src(&alu.src[0], &ctx->src[0], src1_swizzle[i]); + r600_bc_src(&alu.src[1], &ctx->src[1], src0_swizzle[i]); + } else { alu.src[0].sel = V_SQ_ALU_SRC_0; alu.src[0].chan = i; - } - - alu.src[1] = r600_src[1]; - switch (i) { - case 0: - alu.src[1].chan = tgsi_chan(&inst->Src[1], 2); - break; - case 1: - alu.src[1].chan = tgsi_chan(&inst->Src[1], 0); - break; - case 2: - alu.src[1].chan = tgsi_chan(&inst->Src[1], 1); - break; - case 3: alu.src[1].sel = V_SQ_ALU_SRC_0; alu.src[1].chan = i; } @@ -2157,11 +1911,8 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx) if (use_temp) alu.dst.sel = ctx->temp_reg; - else { - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; - } + else + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); alu.dst.chan = i; alu.dst.write = 1; alu.is_op3 = 1; @@ -2170,10 +1921,6 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } if (use_temp) return tgsi_helper_copy(ctx, inst); @@ -2183,7 +1930,6 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx) static int tgsi_exp(struct r600_shader_ctx *ctx) { struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; - struct r600_bc_alu_src r600_src[3] = { { 0 } }; struct r600_bc_alu alu; int r; @@ -2192,11 +1938,7 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.chan = 0; @@ -2206,10 +1948,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = 0; @@ -2221,10 +1959,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* result.y = tmp - floor(tmp); */ @@ -2232,11 +1966,7 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT); - alu.src[0] = r600_src[0]; - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; // r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); @@ -2250,19 +1980,13 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* result.z = RoughApprox2ToX(tmp);*/ if ((inst->Dst[0].Register.WriteMask >> 2) & 0x1) { memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.write = 1; @@ -2273,9 +1997,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* result.w = 1.0;*/ @@ -2293,9 +2014,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } return tgsi_helper_copy(ctx, inst); } @@ -2311,11 +2029,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.chan = 0; @@ -2325,10 +2039,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR); alu.src[0].sel = ctx->temp_reg; alu.src[0].chan = 0; @@ -2341,10 +2051,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* result.y = src.x / (2 ^ floor(log2(src.x))); */ @@ -2352,11 +2058,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.chan = 1; @@ -2367,10 +2069,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR); @@ -2386,10 +2084,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE); @@ -2405,10 +2099,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE); @@ -2424,19 +2114,11 @@ static int tgsi_log(struct r600_shader_ctx *ctx) if (r) return r; - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; - memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.src[1].sel = ctx->temp_reg; alu.src[1].chan = 1; @@ -2449,10 +2131,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* result.z = log2(src);*/ @@ -2460,11 +2138,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_IEEE); - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.dst.sel = ctx->temp_reg; alu.dst.write = 1; @@ -2474,10 +2148,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } /* result.w = 1.0; */ @@ -2496,10 +2166,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx) r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; - - r = r600_bc_add_literal(ctx->bc, ctx->value); - if (r) - return r; } return tgsi_helper_copy(ctx, inst); @@ -2510,6 +2176,7 @@ static int tgsi_eg_arl(struct r600_shader_ctx *ctx) struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; int r; + memset(&alu, 0, sizeof(struct r600_bc_alu)); switch (inst->Instruction.Opcode) { @@ -2524,26 +2191,26 @@ static int tgsi_eg_arl(struct r600_shader_ctx *ctx) return -1; } - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.last = 1; - alu.dst.chan = 0; - alu.dst.sel = ctx->temp_reg; + alu.dst.sel = ctx->ar_reg; alu.dst.write = 1; - r = r600_bc_add_alu_type(ctx->bc, &alu, CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU)); + r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; + + /* TODO: Note that the MOVA can be avoided if we never use AR for + * indexing non-CB registers in the current ALU clause. Similarly, we + * need to load AR from ar_reg again if we started a new clause + * between ARL and AR usage. The easy way to do that is to remove + * the MOVA here, and load it for the first AR access after ar_reg + * has been modified in each clause. */ memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT; - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].sel = ctx->temp_reg; + alu.src[0].sel = ctx->ar_reg; alu.src[0].chan = 0; alu.last = 1; - r = r600_bc_add_alu_type(ctx->bc, &alu, CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU)); + r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; return 0; @@ -2554,29 +2221,51 @@ static int tgsi_r600_arl(struct r600_shader_ctx *ctx) struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; int r; - memset(&alu, 0, sizeof(struct r600_bc_alu)); switch (inst->Instruction.Opcode) { case TGSI_OPCODE_ARL: - alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_FLOOR; + memset(&alu, 0, sizeof(alu)); + alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR; + r600_bc_src(&alu.src[0], &ctx->src[0], 0); + alu.dst.sel = ctx->ar_reg; + alu.dst.write = 1; + alu.last = 1; + + if ((r = r600_bc_add_alu(ctx->bc, &alu))) + return r; + + memset(&alu, 0, sizeof(alu)); + alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT; + alu.src[0].sel = ctx->ar_reg; + alu.dst.sel = ctx->ar_reg; + alu.dst.write = 1; + alu.last = 1; + + if ((r = r600_bc_add_alu(ctx->bc, &alu))) + return r; break; case TGSI_OPCODE_ARR: - alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA; + memset(&alu, 0, sizeof(alu)); + alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT; + r600_bc_src(&alu.src[0], &ctx->src[0], 0); + alu.dst.sel = ctx->ar_reg; + alu.dst.write = 1; + alu.last = 1; + + if ((r = r600_bc_add_alu(ctx->bc, &alu))) + return r; break; default: assert(0); return -1; } - - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); - + memset(&alu, 0, sizeof(alu)); + alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT; + alu.src[0].sel = ctx->ar_reg; alu.last = 1; - r = r600_bc_add_alu_type(ctx->bc, &alu, CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU)); + r = r600_bc_add_alu(ctx->bc, &alu); if (r) return r; ctx->bc->cf_last->r6xx_uses_waterfall = 1; @@ -2593,26 +2282,18 @@ static int tgsi_opdst(struct r600_shader_ctx *ctx) memset(&alu, 0, sizeof(struct r600_bc_alu)); alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL); - r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); - if (r) - return r; + tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst); if (i == 0 || i == 3) { alu.src[0].sel = V_SQ_ALU_SRC_1; } else { - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], i); + r600_bc_src(&alu.src[0], &ctx->src[0], i); } - if (i == 0 || i == 2) { + if (i == 0 || i == 2) { alu.src[1].sel = V_SQ_ALU_SRC_1; } else { - r = tgsi_src(ctx, &inst->Src[1], &alu.src[1]); - if (r) - return r; - alu.src[1].chan = tgsi_chan(&inst->Src[1], i); + r600_bc_src(&alu.src[1], &ctx->src[1], i); } if (i == 3) alu.last = 1; @@ -2625,7 +2306,6 @@ static int tgsi_opdst(struct r600_shader_ctx *ctx) static int emit_logic_pred(struct r600_shader_ctx *ctx, int opcode) { - struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction; struct r600_bc_alu alu; int r; @@ -2637,10 +2317,7 @@ static int emit_logic_pred(struct r600_shader_ctx *ctx, int opcode) alu.dst.write = 1; alu.dst.chan = 0; - r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]); - if (r) - return r; - alu.src[0].chan = tgsi_chan(&inst->Src[0], 0); + r600_bc_src(&alu.src[0], &ctx->src[0], 0); alu.src[1].sel = V_SQ_ALU_SRC_0; alu.src[1].chan = 0; @@ -2654,9 +2331,25 @@ static int emit_logic_pred(struct r600_shader_ctx *ctx, int opcode) static int pops(struct r600_shader_ctx *ctx, int pops) { - r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP)); - ctx->bc->cf_last->pop_count = pops; - ctx->bc->cf_last->cf_addr = ctx->bc->cf_last->id + 2; + int alu_pop = 3; + if (ctx->bc->cf_last) { + if (ctx->bc->cf_last->inst == CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU) << 3) + alu_pop = 0; + else if (ctx->bc->cf_last->inst == CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER) << 3) + alu_pop = 1; + } + alu_pop += pops; + if (alu_pop == 1) { + ctx->bc->cf_last->inst = CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER) << 3; + ctx->bc->force_add_cf = 1; + } else if (alu_pop == 2) { + ctx->bc->cf_last->inst = CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER) << 3; + ctx->bc->force_add_cf = 1; + } else { + r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP)); + ctx->bc->cf_last->pop_count = pops; + ctx->bc->cf_last->cf_addr = ctx->bc->cf_last->id + 2; + } return 0; } @@ -3002,7 +2695,7 @@ static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = { {TGSI_OPCODE_CEIL, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, {TGSI_OPCODE_I2F, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, {TGSI_OPCODE_NOT, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, - {TGSI_OPCODE_TRUNC, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_trans_srcx_replicate}, + {TGSI_OPCODE_TRUNC, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_op2}, {TGSI_OPCODE_SHL, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, /* gap */ {88, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, @@ -3075,7 +2768,7 @@ static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = { {TGSI_OPCODE_RCP, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE, tgsi_trans_srcx_replicate}, {TGSI_OPCODE_RSQ, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIPSQRT_IEEE, tgsi_trans_srcx_replicate}, {TGSI_OPCODE_EXP, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_exp}, - {TGSI_OPCODE_LOG, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, + {TGSI_OPCODE_LOG, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_log}, {TGSI_OPCODE_MUL, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL, tgsi_op2}, {TGSI_OPCODE_ADD, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD, tgsi_op2}, {TGSI_OPCODE_DP3, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_DOT4, tgsi_dp}, @@ -3160,7 +2853,7 @@ static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = { {TGSI_OPCODE_CEIL, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, {TGSI_OPCODE_I2F, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, {TGSI_OPCODE_NOT, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, - {TGSI_OPCODE_TRUNC, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_trans_srcx_replicate}, + {TGSI_OPCODE_TRUNC, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_op2}, {TGSI_OPCODE_SHL, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, /* gap */ {88, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported}, |