Signed-off-by: Zebediah Figura zfigura@codeweavers.com --- libs/vkd3d-shader/hlsl.c | 4 +- libs/vkd3d-shader/hlsl.h | 9 ++ libs/vkd3d-shader/hlsl_codegen.c | 184 +++++++++++++++++++++++++++++++ 3 files changed, 195 insertions(+), 2 deletions(-)
diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 53b9db36..67ca2f2c 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -890,7 +890,7 @@ static void dump_deref(struct vkd3d_string_buffer *buffer, const struct hlsl_der } }
-static const char *debug_writemask(DWORD writemask) +const char *debug_hlsl_writemask(unsigned int writemask) { static const char components[] = {'x', 'y', 'z', 'w'}; char string[5]; @@ -914,7 +914,7 @@ static void dump_ir_assignment(struct vkd3d_string_buffer *buffer, const struct vkd3d_string_buffer_printf(buffer, "= ("); dump_deref(buffer, &assign->lhs); if (assign->writemask != VKD3DSP_WRITEMASK_ALL) - vkd3d_string_buffer_printf(buffer, "%s", debug_writemask(assign->writemask)); + vkd3d_string_buffer_printf(buffer, "%s", debug_hlsl_writemask(assign->writemask)); vkd3d_string_buffer_printf(buffer, " "); dump_src(buffer, &assign->rhs); vkd3d_string_buffer_printf(buffer, ")"); diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index cf12f63f..fa7709ef 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -137,6 +137,13 @@ struct hlsl_struct_field unsigned int reg_offset; };
+struct hlsl_reg +{ + uint32_t id; + unsigned char writemask; + bool allocated; +}; + enum hlsl_ir_node_type { HLSL_IR_ASSIGNMENT = 0, @@ -208,6 +215,7 @@ struct hlsl_ir_var struct list scope_entry, param_entry;
unsigned int first_write, last_read; + struct hlsl_reg reg;
uint32_t is_input_varying : 1; uint32_t is_output_varying : 1; @@ -493,6 +501,7 @@ static inline void hlsl_src_remove(struct hlsl_src *src) }
const char *debug_hlsl_type(const struct hlsl_type *type) DECLSPEC_HIDDEN; +const char *debug_hlsl_writemask(unsigned int writemask) DECLSPEC_HIDDEN;
struct vkd3d_string_buffer *hlsl_type_to_string(struct vkd3d_string_buffer_cache *string_buffers, const struct hlsl_type *type) DECLSPEC_HIDDEN; diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index dbd591af..420e720f 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -612,6 +612,188 @@ static void compute_liveness(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl compute_liveness_recurse(entry_func->body, 0, 0); }
+struct liveness +{ + size_t size; + struct + { + /* 0 if not live yet. */ + unsigned int last_read; + } *regs; +}; + +static unsigned char get_available_writemask(struct liveness *liveness, + unsigned int first_write, unsigned int index, unsigned int components) +{ + unsigned char i, writemask = 0, count = 0; + + for (i = 0; i < 4; ++i) + { + if (liveness->regs[index + i].last_read <= first_write) + { + writemask |= 1 << i; + if (++count == components) + return writemask; + } + } + + return 0; +} + +static bool resize_liveness(struct liveness *liveness, size_t new_count) +{ + size_t old_capacity = liveness->size; + + if (!vkd3d_array_reserve((void **)&liveness->regs, &liveness->size, new_count, sizeof(*liveness->regs))) + return false; + + if (liveness->size > old_capacity) + memset(liveness->regs + old_capacity, 0, (liveness->size - old_capacity) * sizeof(*liveness->regs)); + return true; +} + +static struct hlsl_reg allocate_register(struct liveness *liveness, + unsigned int first_write, unsigned int last_read, unsigned char components) +{ + struct hlsl_reg ret = {.allocated = true}; + unsigned char writemask, i; + unsigned int regnum; + + for (regnum = 0; regnum < liveness->size; regnum += 4) + { + if ((writemask = get_available_writemask(liveness, first_write, regnum, components))) + break; + } + if (regnum == liveness->size) + { + if (!resize_liveness(liveness, regnum + 4)) + return ret; + writemask = (1 << components) - 1; + } + for (i = 0; i < 4; ++i) + { + if (writemask & (1 << i)) + liveness->regs[regnum + i].last_read = last_read; + } + ret.id = regnum / 4; + ret.writemask = writemask; + return ret; +} + +static bool is_range_available(struct liveness *liveness, unsigned int first_write, + unsigned int index, unsigned int elements) +{ + unsigned int i; + + for (i = 0; i < elements; i += 4) + { + if (!get_available_writemask(liveness, first_write, index + i, 4)) + return false; + } + return true; +} + +/* "elements" is the total number of consecutive whole registers needed. */ +static struct hlsl_reg allocate_range(struct liveness *liveness, + unsigned int first_write, unsigned int last_read, unsigned int elements) +{ + const unsigned int components = elements * 4; + struct hlsl_reg ret = {.allocated = true}; + unsigned int i, regnum; + + for (regnum = 0; regnum < liveness->size; regnum += 4) + { + if (is_range_available(liveness, first_write, regnum, min(components, liveness->size - regnum))) + break; + } + if (!resize_liveness(liveness, regnum + components)) + return ret; + + for (i = 0; i < components; ++i) + liveness->regs[regnum + i].last_read = last_read; + ret.id = regnum / 4; + return ret; +} + +static const char *debug_register(char class, struct hlsl_reg reg, const struct hlsl_type *type) +{ + if (type->reg_size > 4) + return vkd3d_dbg_sprintf("%c%u-%c%u", class, reg.id, class, + reg.id + type->reg_size - 1); + return vkd3d_dbg_sprintf("%c%u%s", class, reg.id, debug_hlsl_writemask(reg.writemask)); +} + +static void allocate_variable_temp_register(struct hlsl_ir_var *var, struct liveness *liveness) +{ + if (var->is_input_varying || var->is_output_varying || var->is_uniform) + return; + + if (!var->reg.allocated && var->last_read) + { + if (var->data_type->reg_size > 1) + var->reg = allocate_range(liveness, var->first_write, + var->last_read, var->data_type->reg_size); + else + var->reg = allocate_register(liveness, var->first_write, + var->last_read, var->data_type->dimx); + TRACE("Allocated %s to %s (liveness %u-%u).\n", debug_register('r', var->reg, var->data_type), + var->name, var->first_write, var->last_read); + } +} + +static void allocate_temp_registers_recurse(struct list *instrs, struct liveness *liveness) +{ + struct hlsl_ir_node *instr; + + LIST_FOR_EACH_ENTRY(instr, instrs, struct hlsl_ir_node, entry) + { + switch (instr->type) + { + case HLSL_IR_ASSIGNMENT: + { + struct hlsl_ir_assignment *assignment = hlsl_ir_assignment(instr); + allocate_variable_temp_register(assignment->lhs.var, liveness); + break; + } + + case HLSL_IR_IF: + { + struct hlsl_ir_if *iff = hlsl_ir_if(instr); + allocate_temp_registers_recurse(&iff->then_instrs, liveness); + allocate_temp_registers_recurse(&iff->else_instrs, liveness); + break; + } + + case HLSL_IR_LOAD: + { + struct hlsl_ir_load *load = hlsl_ir_load(instr); + allocate_variable_temp_register(load->src.var, liveness); + break; + } + + case HLSL_IR_LOOP: + { + struct hlsl_ir_loop *loop = hlsl_ir_loop(instr); + allocate_temp_registers_recurse(&loop->body, liveness); + break; + } + + default: + break; + } + } +} + +/* Simple greedy temporary register allocation pass that just assigns a unique + * index to all (simultaneously live) variables or intermediate values. Agnostic + * as to how many registers are actually available for the current backend, and + * does not handle constants. */ +static void allocate_temp_registers(struct hlsl_ir_function_decl *entry_func) +{ + struct liveness liveness = {0}; + allocate_temp_registers_recurse(entry_func->body, &liveness); +} + int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func) { struct hlsl_ir_var *var; @@ -649,6 +831,8 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun if (TRACE_ON()) rb_for_each_entry(&ctx->functions, dump_function, NULL);
+ allocate_temp_registers(entry_func); + if (ctx->failed) return VKD3D_ERROR_INVALID_SHADER; return VKD3D_ERROR_NOT_IMPLEMENTED;