Signed-off-by: Giovanni Mascellani gmascellani@codeweavers.com Signed-off-by: Zebediah Figura zfigura@codeweavers.com --- v8: Address comments in https://source.winehq.org/patches/data/219953
libs/vkd3d-shader/hlsl.c | 21 ++- libs/vkd3d-shader/hlsl.h | 1 + libs/vkd3d-shader/hlsl_codegen.c | 222 ++++++++++++++++++++++++++++++- 3 files changed, 238 insertions(+), 6 deletions(-)
diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 1eee4278..f59e7403 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -1097,6 +1097,20 @@ const char *debug_hlsl_writemask(unsigned int writemask) return vkd3d_dbg_sprintf(".%s", string); }
+const char *debug_hlsl_swizzle(unsigned int swizzle, unsigned int count) +{ + static const char components[] = {'x', 'y', 'z', 'w'}; + char string[5]; + unsigned int i = 0; + + for (i = 0; i < count; ++i) + { + string[i] = components[(swizzle >> (2 * i)) % 4]; + } + string[count] = '\0'; + return vkd3d_dbg_sprintf(".%s", string); +} + static void dump_ir_constant(struct vkd3d_string_buffer *buffer, const struct hlsl_ir_constant *constant) { struct hlsl_type *type = constant->node.data_type; @@ -1277,18 +1291,15 @@ static void dump_ir_swizzle(struct vkd3d_string_buffer *buffer, const struct hls unsigned int i;
dump_src(buffer, &swizzle->val); - vkd3d_string_buffer_printf(buffer, "."); if (swizzle->val.node->data_type->dimy > 1) { + vkd3d_string_buffer_printf(buffer, "."); for (i = 0; i < swizzle->node.data_type->dimx; ++i) vkd3d_string_buffer_printf(buffer, "_m%u%u", (swizzle->swizzle >> i * 8) & 0xf, (swizzle->swizzle >> (i * 8 + 4)) & 0xf); } else { - static const char c[] = {'x', 'y', 'z', 'w'}; - - for (i = 0; i < swizzle->node.data_type->dimx; ++i) - vkd3d_string_buffer_printf(buffer, "%c", c[(swizzle->swizzle >> i * 2) & 0x3]); + vkd3d_string_buffer_printf(buffer, "%s", debug_hlsl_swizzle(swizzle->swizzle, swizzle->node.data_type->dimx)); } }
diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 365ef980..bca59b65 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -666,6 +666,7 @@ static inline struct hlsl_type *hlsl_get_numeric_type(const struct hlsl_ctx *ctx const char *debug_hlsl_expr_op(enum hlsl_ir_expr_op op); const char *debug_hlsl_type(struct hlsl_ctx *ctx, const struct hlsl_type *type); const char *debug_hlsl_writemask(unsigned int writemask); +const char *debug_hlsl_swizzle(unsigned int swizzle, unsigned int count);
struct vkd3d_string_buffer *hlsl_type_to_string(struct hlsl_ctx *ctx, const struct hlsl_type *type); struct vkd3d_string_buffer *hlsl_modifiers_to_string(struct hlsl_ctx *ctx, unsigned int modifiers); diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 4a7e003f..4e1397f6 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -237,6 +237,221 @@ static void replace_node(struct hlsl_ir_node *old, struct hlsl_ir_node *new) hlsl_free_instr(old); }
+struct copy_propagation_value +{ + struct hlsl_ir_node *node; + unsigned int component; +}; + +struct copy_propagation_var_def +{ + struct rb_entry entry; + struct hlsl_ir_var *var; + struct copy_propagation_value values[]; +}; + +struct copy_propagation_state +{ + struct rb_tree variables; +}; + +static int copy_propagation_variable_compare(const void *key, const struct rb_entry *entry) +{ + struct copy_propagation_var_def *variable = RB_ENTRY_VALUE(entry, struct copy_propagation_var_def, entry); + uintptr_t key_int = (uintptr_t)key, entry_int = (uintptr_t)variable->var; + + return (key_int > entry_int) - (key_int < entry_int); +} + +static void copy_propagation_variable_destroy(struct rb_entry *entry, void *context) +{ + struct copy_propagation_var_def *variable = RB_ENTRY_VALUE(entry, struct copy_propagation_var_def, entry); + + vkd3d_free(variable); +} + +static struct copy_propagation_var_def *copy_propagation_get_variable(struct copy_propagation_state *state, + struct hlsl_ir_var *var) +{ + struct rb_entry *entry = rb_get(&state->variables, var); + + if (entry) + return RB_ENTRY_VALUE(entry, struct copy_propagation_var_def, entry); + else + return NULL; +} + +static struct copy_propagation_var_def *copy_propagation_create_variable(struct hlsl_ctx *ctx, + struct copy_propagation_state *state, struct hlsl_ir_var *var) +{ + struct rb_entry *entry = rb_get(&state->variables, var); + struct copy_propagation_var_def *variable; + int res; + + if (entry) + return RB_ENTRY_VALUE(entry, struct copy_propagation_var_def, entry); + + variable = hlsl_alloc(ctx, offsetof(struct copy_propagation_var_def, values[var->data_type->reg_size])); + if (!variable) + return NULL; + + variable->var = var; + + res = rb_put(&state->variables, var, &variable->entry); + assert(!res); + + return variable; +} + +static void copy_propagation_invalidate_whole_variable(struct copy_propagation_var_def *variable) +{ + TRACE("Invalidate variable %s.\n", variable->var->name); + memset(variable->values, 0, sizeof(*variable->values) * variable->var->data_type->reg_size); +} + +static void copy_propagation_set_value(struct copy_propagation_var_def *variable, unsigned int offset, + unsigned char writemask, struct hlsl_ir_node *node) +{ + unsigned int i, j = 0; + + for (i = 0; i < 4; ++i) + { + if (writemask & (1u << i)) + { + TRACE("Variable %s[%u] is written by instruction %p%s.\n", + variable->var->name, offset + i, node, debug_hlsl_writemask(1u << i)); + variable->values[offset + i].node = node; + variable->values[offset + i].component = j++; + } + } +} + +static struct hlsl_ir_node *copy_propagation_compute_replacement(struct copy_propagation_var_def *variable, + unsigned int offset, unsigned int count, unsigned int *swizzle) +{ + struct hlsl_ir_node *node = NULL; + unsigned int i; + + assert(offset + count <= variable->var->data_type->reg_size); + + *swizzle = 0; + + for (i = 0; i < count; ++i) + { + if (!node) + node = variable->values[offset + i].node; + else if (node != variable->values[offset + i].node) + return NULL; + *swizzle |= variable->values[offset + i].component << (2 * i); + } + + return node; +} + +static bool copy_propagation_analyze_load(struct hlsl_ctx *ctx, struct hlsl_ir_load *load, + struct copy_propagation_state *state) +{ + struct hlsl_ir_node *node = &load->node, *new_node; + struct copy_propagation_var_def *variable; + struct hlsl_type *type = node->data_type; + struct hlsl_ir_swizzle *swizzle_node; + struct hlsl_deref *src = &load->src; + struct hlsl_ir_var *var = src->var; + unsigned int offset, swizzle; + + if (type->type != HLSL_CLASS_SCALAR && type->type != HLSL_CLASS_VECTOR) + return false; + + if (!hlsl_offset_from_deref(src, &offset)) + return false; + + variable = copy_propagation_get_variable(state, var); + if (!variable) + return false; + + new_node = copy_propagation_compute_replacement(variable, offset, type->dimx, &swizzle); + + TRACE("Load from %s[%d-%d] reconstructed as instruction %p%s.\n", + var->name, offset, offset + type->dimx, new_node, debug_hlsl_swizzle(swizzle, 4)); + + if (!new_node) + return false; + + if (!(swizzle_node = hlsl_new_swizzle(ctx, swizzle, type->dimx, new_node, &node->loc))) + return false; + list_add_before(&node->entry, &swizzle_node->node.entry); + + replace_node(node, &swizzle_node->node); + + return true; +} + +static void copy_propagation_record_store(struct hlsl_ctx *ctx, struct hlsl_ir_store *store, + struct copy_propagation_state *state) +{ + struct copy_propagation_var_def *variable; + struct hlsl_deref *lhs = &store->lhs; + struct hlsl_ir_var *var = lhs->var; + unsigned int offset; + + variable = copy_propagation_create_variable(ctx, state, var); + if (!variable) + return; + + if (hlsl_offset_from_deref(lhs, &offset)) + copy_propagation_set_value(variable, offset, store->writemask, store->rhs.node); + else + copy_propagation_invalidate_whole_variable(variable); +} + +static bool copy_propagation_transform_block(struct hlsl_ctx *ctx, struct hlsl_block *block, + struct copy_propagation_state *state) +{ + struct hlsl_ir_node *instr, *next; + bool progress = false; + + LIST_FOR_EACH_ENTRY_SAFE(instr, next, &block->instrs, struct hlsl_ir_node, entry) + { + switch (instr->type) + { + case HLSL_IR_LOAD: + progress |= copy_propagation_analyze_load(ctx, hlsl_ir_load(instr), state); + break; + + case HLSL_IR_STORE: + copy_propagation_record_store(ctx, hlsl_ir_store(instr), state); + break; + + case HLSL_IR_IF: + FIXME("Copy propagation doesn't support conditionals yet, leaving.\n"); + return progress; + + case HLSL_IR_LOOP: + FIXME("Copy propagation doesn't support loops yet, leaving.\n"); + return progress; + + default: + break; + } + } + + return progress; +} + +static bool copy_propagation_execute(struct hlsl_ctx *ctx, struct hlsl_block *block) +{ + struct copy_propagation_state state; + bool progress; + + rb_init(&state.variables, copy_propagation_variable_compare); + + progress = copy_propagation_transform_block(ctx, block, &state); + + rb_destroy(&state.variables, copy_propagation_variable_destroy, NULL); + + return progress; +} + static bool is_vec1(const struct hlsl_type *type) { return (type->type == HLSL_CLASS_SCALAR) || (type->type == HLSL_CLASS_VECTOR && type->dimx == 1); @@ -1390,7 +1605,12 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun progress |= transform_ir(ctx, split_struct_copies, body, NULL); } while (progress); - while (transform_ir(ctx, fold_constants, body, NULL)); + do + { + progress = transform_ir(ctx, fold_constants, body, NULL); + progress |= copy_propagation_execute(ctx, body); + } + while (progress); transform_ir(ctx, remove_trivial_swizzles, body, NULL);
if (ctx->profile->major_version < 4)