From d44a516142b094cf3cf2fb8471b0fb5429e1d5ce Mon Sep 17 00:00:00 2001 From: Alistair Leslie-Hughes Date: Thu, 8 Jun 2023 14:12:14 +1000 Subject: [PATCH] Updated vkd3d-latest patchset --- ...-ebf7573571d4bfcd3f38846886106f204fd.patch | 336 ++++++++++++++++++ 1 file changed, 336 insertions(+) create mode 100644 patches/vkd3d-latest/0005--Updated-vkd3d-to-ebf7573571d4bfcd3f38846886106f204fd.patch diff --git a/patches/vkd3d-latest/0005--Updated-vkd3d-to-ebf7573571d4bfcd3f38846886106f204fd.patch b/patches/vkd3d-latest/0005--Updated-vkd3d-to-ebf7573571d4bfcd3f38846886106f204fd.patch new file mode 100644 index 00000000..4875240d --- /dev/null +++ b/patches/vkd3d-latest/0005--Updated-vkd3d-to-ebf7573571d4bfcd3f38846886106f204fd.patch @@ -0,0 +1,336 @@ +From 65a602939e9306618d1ec33d19b217b869638beb Mon Sep 17 00:00:00 2001 +From: Alistair Leslie-Hughes +Date: Thu, 8 Jun 2023 14:11:22 +1000 +Subject: [PATCH] Updated vkd3d to ebf7573571d4bfcd3f38846886106f204fd7f0e8 + +--- + libs/vkd3d/libs/vkd3d-shader/hlsl.c | 13 ++ + libs/vkd3d/libs/vkd3d-shader/hlsl.h | 2 + + libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c | 209 ++++++++++++++++---- + 3 files changed, 191 insertions(+), 33 deletions(-) + +diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl.c b/libs/vkd3d/libs/vkd3d-shader/hlsl.c +index 152ec6275eb..f1edfa5e625 100644 +--- a/libs/vkd3d/libs/vkd3d-shader/hlsl.c ++++ b/libs/vkd3d/libs/vkd3d-shader/hlsl.c +@@ -1287,6 +1287,19 @@ struct hlsl_ir_load *hlsl_new_load_index(struct hlsl_ctx *ctx, const struct hlsl + return load; + } + ++struct hlsl_ir_load *hlsl_new_load_parent(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, ++ const struct vkd3d_shader_location *loc) ++{ ++ /* This deref can only exists temporarily because it is not the real owner of its members. */ ++ struct hlsl_deref tmp_deref; ++ ++ assert(deref->path_len >= 1); ++ ++ tmp_deref = *deref; ++ tmp_deref.path_len = deref->path_len - 1; ++ return hlsl_new_load_index(ctx, &tmp_deref, NULL, loc); ++} ++ + struct hlsl_ir_load *hlsl_new_var_load(struct hlsl_ctx *ctx, struct hlsl_ir_var *var, + const struct vkd3d_shader_location *loc) + { +diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl.h b/libs/vkd3d/libs/vkd3d-shader/hlsl.h +index 6b79c582f55..14037a4a96f 100644 +--- a/libs/vkd3d/libs/vkd3d-shader/hlsl.h ++++ b/libs/vkd3d/libs/vkd3d-shader/hlsl.h +@@ -1129,6 +1129,8 @@ struct hlsl_ir_load *hlsl_new_var_load(struct hlsl_ctx *ctx, struct hlsl_ir_var + const struct vkd3d_shader_location *loc); + struct hlsl_ir_load *hlsl_new_load_index(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, + struct hlsl_ir_node *idx, const struct vkd3d_shader_location *loc); ++struct hlsl_ir_load *hlsl_new_load_parent(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, ++ const struct vkd3d_shader_location *loc); + struct hlsl_ir_node *hlsl_new_load_component(struct hlsl_ctx *ctx, struct hlsl_block *block, + const struct hlsl_deref *deref, unsigned int comp, const struct vkd3d_shader_location *loc); + +diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c +index 2b6c595a15d..72ab27d3b80 100644 +--- a/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c ++++ b/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c +@@ -123,15 +123,14 @@ static struct hlsl_ir_node *new_offset_instr_from_deref(struct hlsl_ctx *ctx, st + } + + /* TODO: remove when no longer needed, only used for transform_deref_paths_into_offsets() */ +-static void replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_deref *deref, ++static bool replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_deref *deref, + struct hlsl_ir_node *instr) + { + const struct hlsl_type *type; + struct hlsl_ir_node *offset; + struct hlsl_block block; + +- if (!deref->var) +- return; ++ assert(deref->var); + + /* register offsets shouldn't be used before this point is reached. */ + assert(!deref->offset.node); +@@ -143,45 +142,19 @@ static void replace_deref_path_with_offset(struct hlsl_ctx *ctx, struct hlsl_der + if (type->class == HLSL_CLASS_STRUCT || type->class == HLSL_CLASS_ARRAY) + { + hlsl_cleanup_deref(deref); +- return; ++ return true; + } + + deref->offset_regset = hlsl_type_get_regset(type); + + if (!(offset = new_offset_instr_from_deref(ctx, &block, deref, &instr->loc))) +- return; ++ return false; + list_move_before(&instr->entry, &block.instrs); + + hlsl_cleanup_deref(deref); + hlsl_src_from_node(&deref->offset, offset); +-} + +-/* TODO: remove when no longer needed. */ +-static bool transform_deref_paths_into_offsets(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +-{ +- switch(instr->type) +- { +- case HLSL_IR_LOAD: +- replace_deref_path_with_offset(ctx, &hlsl_ir_load(instr)->src, instr); +- return true; +- +- case HLSL_IR_STORE: +- replace_deref_path_with_offset(ctx, &hlsl_ir_store(instr)->lhs, instr); +- return true; +- +- case HLSL_IR_RESOURCE_LOAD: +- replace_deref_path_with_offset(ctx, &hlsl_ir_resource_load(instr)->resource, instr); +- replace_deref_path_with_offset(ctx, &hlsl_ir_resource_load(instr)->sampler, instr); +- return true; +- +- case HLSL_IR_RESOURCE_STORE: +- replace_deref_path_with_offset(ctx, &hlsl_ir_resource_store(instr)->resource, instr); +- return true; +- +- default: +- return false; +- } +- return false; ++ return true; + } + + /* Split uniforms into two variables representing the constant and temp +@@ -600,6 +573,44 @@ bool hlsl_transform_ir(struct hlsl_ctx *ctx, bool (*func)(struct hlsl_ctx *ctx, + return progress; + } + ++static bool transform_instr_derefs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) ++{ ++ bool res; ++ bool (*func)(struct hlsl_ctx *ctx, struct hlsl_deref *, struct hlsl_ir_node *) = context; ++ ++ switch(instr->type) ++ { ++ case HLSL_IR_LOAD: ++ res = func(ctx, &hlsl_ir_load(instr)->src, instr); ++ return res; ++ ++ case HLSL_IR_STORE: ++ res = func(ctx, &hlsl_ir_store(instr)->lhs, instr); ++ return res; ++ ++ case HLSL_IR_RESOURCE_LOAD: ++ res = func(ctx, &hlsl_ir_resource_load(instr)->resource, instr); ++ if (hlsl_ir_resource_load(instr)->sampler.var) ++ res |= func(ctx, &hlsl_ir_resource_load(instr)->sampler, instr); ++ return res; ++ ++ case HLSL_IR_RESOURCE_STORE: ++ res = func(ctx, &hlsl_ir_resource_store(instr)->resource, instr); ++ return res; ++ ++ default: ++ return false; ++ } ++ return false; ++} ++ ++static bool transform_derefs(struct hlsl_ctx *ctx, ++ bool (*func)(struct hlsl_ctx *ctx, struct hlsl_deref *, struct hlsl_ir_node *), ++ struct hlsl_block *block) ++{ ++ return hlsl_transform_ir(ctx, transform_instr_derefs, block, func); ++} ++ + struct recursive_call_ctx + { + const struct hlsl_ir_function_decl **backtrace; +@@ -1981,6 +1992,81 @@ static bool remove_trivial_swizzles(struct hlsl_ctx *ctx, struct hlsl_ir_node *i + return true; + } + ++static bool lower_nonconstant_vector_derefs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) ++{ ++ struct hlsl_ir_node *idx; ++ struct hlsl_deref *deref; ++ struct hlsl_type *type; ++ unsigned int i; ++ ++ if (instr->type != HLSL_IR_LOAD) ++ return false; ++ ++ deref = &hlsl_ir_load(instr)->src; ++ assert(deref->var); ++ ++ if (deref->path_len == 0) ++ return false; ++ ++ type = deref->var->data_type; ++ for (i = 0; i < deref->path_len - 1; ++i) ++ type = hlsl_get_element_type_from_path_index(ctx, type, deref->path[i].node); ++ ++ idx = deref->path[deref->path_len - 1].node; ++ ++ if (type->class == HLSL_CLASS_VECTOR && idx->type != HLSL_IR_CONSTANT) ++ { ++ struct hlsl_ir_node *eq, *swizzle, *dot, *operands[HLSL_MAX_OPERANDS] = {0}; ++ struct hlsl_ir_load *vector_load; ++ struct hlsl_ir_constant *c; ++ enum hlsl_ir_expr_op op; ++ ++ if (!(vector_load = hlsl_new_load_parent(ctx, deref, &instr->loc))) ++ return false; ++ list_add_before(&instr->entry, &vector_load->node.entry); ++ ++ if (!(swizzle = hlsl_new_swizzle(ctx, HLSL_SWIZZLE(X, X, X, X), type->dimx, idx, &instr->loc))) ++ return false; ++ list_add_before(&instr->entry, &swizzle->entry); ++ ++ if (!(c = hlsl_new_constant(ctx, hlsl_get_vector_type(ctx, HLSL_TYPE_UINT, type->dimx), &instr->loc))) ++ return false; ++ c->value.u[0].u = 0; ++ c->value.u[1].u = 1; ++ c->value.u[2].u = 2; ++ c->value.u[3].u = 3; ++ list_add_before(&instr->entry, &c->node.entry); ++ ++ operands[0] = swizzle; ++ operands[1] = &c->node; ++ if (!(eq = hlsl_new_expr(ctx, HLSL_OP2_EQUAL, operands, ++ hlsl_get_vector_type(ctx, HLSL_TYPE_BOOL, type->dimx), &instr->loc))) ++ return false; ++ list_add_before(&instr->entry, &eq->entry); ++ ++ if (!(eq = hlsl_new_cast(ctx, eq, type, &instr->loc))) ++ return false; ++ list_add_before(&instr->entry, &eq->entry); ++ ++ op = HLSL_OP2_DOT; ++ if (type->dimx == 1) ++ op = type->base_type == HLSL_TYPE_BOOL ? HLSL_OP2_LOGIC_AND : HLSL_OP2_MUL; ++ ++ /* Note: We may be creating a DOT for bool vectors here, which we need to lower to ++ * LOGIC_OR + LOGIC_AND. */ ++ operands[0] = &vector_load->node; ++ operands[1] = eq; ++ if (!(dot = hlsl_new_expr(ctx, op, operands, instr->data_type, &instr->loc))) ++ return false; ++ list_add_before(&instr->entry, &dot->entry); ++ hlsl_replace_node(instr, dot); ++ ++ return true; ++ } ++ ++ return false; ++} ++ + /* Lower DIV to RCP + MUL. */ + static bool lower_division(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) + { +@@ -2378,6 +2464,58 @@ static bool lower_int_abs(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void + return true; + } + ++static bool lower_int_dot(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) ++{ ++ struct hlsl_ir_node *arg1, *arg2, *mult, *comps[4] = {0}, *res; ++ struct hlsl_type *type = instr->data_type; ++ struct hlsl_ir_expr *expr; ++ unsigned int i, dimx; ++ bool is_bool; ++ ++ if (instr->type != HLSL_IR_EXPR) ++ return false; ++ expr = hlsl_ir_expr(instr); ++ ++ if (expr->op != HLSL_OP2_DOT) ++ return false; ++ ++ if (type->base_type == HLSL_TYPE_INT || type->base_type == HLSL_TYPE_UINT ++ || type->base_type == HLSL_TYPE_BOOL) ++ { ++ arg1 = expr->operands[0].node; ++ arg2 = expr->operands[1].node; ++ assert(arg1->data_type->dimx == arg2->data_type->dimx); ++ dimx = arg1->data_type->dimx; ++ is_bool = type->base_type == HLSL_TYPE_BOOL; ++ ++ if (!(mult = hlsl_new_binary_expr(ctx, is_bool ? HLSL_OP2_LOGIC_AND : HLSL_OP2_MUL, arg1, arg2))) ++ return false; ++ list_add_before(&instr->entry, &mult->entry); ++ ++ for (i = 0; i < dimx; ++i) ++ { ++ unsigned int s = hlsl_swizzle_from_writemask(1 << i); ++ ++ if (!(comps[i] = hlsl_new_swizzle(ctx, s, 1, mult, &instr->loc))) ++ return false; ++ list_add_before(&instr->entry, &comps[i]->entry); ++ } ++ ++ res = comps[0]; ++ for (i = 1; i < dimx; ++i) ++ { ++ if (!(res = hlsl_new_binary_expr(ctx, is_bool ? HLSL_OP2_LOGIC_OR : HLSL_OP2_ADD, res, comps[i]))) ++ return false; ++ list_add_before(&instr->entry, &res->entry); ++ } ++ ++ hlsl_replace_node(instr, res); ++ return true; ++ } ++ ++ return false; ++} ++ + static bool lower_float_modulus(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) + { + struct hlsl_ir_node *arg1, *arg2, *mul1, *neg1, *ge, *neg2, *div, *mul2, *frc, *cond; +@@ -3925,6 +4063,7 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry + + hlsl_transform_ir(ctx, lower_narrowing_casts, body, NULL); + hlsl_transform_ir(ctx, lower_casts_to_bool, body, NULL); ++ hlsl_transform_ir(ctx, lower_int_dot, body, NULL); + hlsl_transform_ir(ctx, lower_int_division, body, NULL); + hlsl_transform_ir(ctx, lower_int_modulus, body, NULL); + hlsl_transform_ir(ctx, lower_int_abs, body, NULL); +@@ -3939,6 +4078,10 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry + } + while (progress); + ++ hlsl_transform_ir(ctx, lower_nonconstant_vector_derefs, body, NULL); ++ hlsl_transform_ir(ctx, lower_casts_to_bool, body, NULL); ++ hlsl_transform_ir(ctx, lower_int_dot, body, NULL); ++ + if (profile->major_version < 4) + { + hlsl_transform_ir(ctx, lower_division, body, NULL); +@@ -3956,7 +4099,7 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry + hlsl_transform_ir(ctx, track_object_components_usage, body, NULL); + + /* TODO: move forward, remove when no longer needed */ +- hlsl_transform_ir(ctx, transform_deref_paths_into_offsets, body, NULL); ++ transform_derefs(ctx, replace_deref_path_with_offset, body); + while (hlsl_transform_ir(ctx, hlsl_fold_constant_exprs, body, NULL)); + + do +-- +2.40.1 +