wine/vkd3d
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/vkd3d.git synced 2025-09-12 18:50:22 -07:00
Code Issues Packages Projects Releases Wiki Activity

vkd3d-shader/hlsl: Use a simple counter for indexable temps.

Browse Source
This commit is contained in:
Elizabeth Figura
2025-07-09 19:45:48 -05:00
committed by Henri Verbeet
parent 8db9465693
commit a44d79a13c
Notes: Henri Verbeet 2025-09-04 14:12:02 +02:00 
Approved-by: Francisco Casas (@fcasas)
Approved-by: Henri Verbeet (@hverbeet)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1713
2 changed files with 19 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
libs/vkd3d-shader/hlsl.h
View File

@@ -1187,8 +1187,8 @@ struct hlsl_ctx
} constant_defs; } constant_defs;
/* 'c' registers where the constants expected by SM2 sincos are stored. */ /* 'c' registers where the constants expected by SM2 sincos are stored. */
struct hlsl_reg d3dsincosconst1, d3dsincosconst2; struct hlsl_reg d3dsincosconst1, d3dsincosconst2;
/* Number of allocated SSA and temp IDs, used in translation to vsir. */ /* Number of allocated registers, used in translation to vsir. */
unsigned int ssa_count, temp_count; unsigned int ssa_count, temp_count, indexable_temp_count;
/* Number of threads to be executed (on the X, Y, and Z dimensions) in a single thread group in /* Number of threads to be executed (on the X, Y, and Z dimensions) in a single thread group in
* compute shader profiles. It is set using the numthreads() attribute in the entry point. */ * compute shader profiles. It is set using the numthreads() attribute in the entry point. */

54
libs/vkd3d-shader/hlsl_codegen.c
View File

@@ -5931,11 +5931,7 @@ struct register_allocator
} *allocations; } *allocations;
size_t count, capacity; size_t count, capacity;
/* Indexable temps are allocated separately and always keep their index regardless of their /* Total number of registers allocated so far. */
* lifetime. */
uint32_t indexable_count;
/* Total number of registers allocated so far. Used to declare sm4 temp count. */
uint32_t reg_count; uint32_t reg_count;
/* Special flag so allocations that can share registers prioritize those /* Special flag so allocations that can share registers prioritize those
@@ -6099,18 +6095,14 @@ static struct hlsl_reg allocate_numeric_registers_for_type(struct hlsl_ctx *ctx,
unsigned int first_write, unsigned int last_read, const struct hlsl_type *type) unsigned int first_write, unsigned int last_read, const struct hlsl_type *type)
{ {
unsigned int reg_size = type->reg_size[HLSL_REGSET_NUMERIC]; unsigned int reg_size = type->reg_size[HLSL_REGSET_NUMERIC];
struct hlsl_reg ret;
/* FIXME: We could potentially pack structs or arrays more efficiently... */ /* FIXME: We could potentially pack structs or arrays more efficiently... */
if (type->class <= HLSL_CLASS_VECTOR) if (type->class <= HLSL_CLASS_VECTOR)
ret = allocate_register(ctx, allocator, first_write, last_read, return allocate_register(ctx, allocator, first_write, last_read,
type->e.numeric.dimx, type->e.numeric.dimx, 0, false, false); type->e.numeric.dimx, type->e.numeric.dimx, 0, false, false);
else else
ret = allocate_range(ctx, allocator, first_write, last_read, reg_size, 0, false); return allocate_range(ctx, allocator, first_write, last_read, reg_size, 0, false);
if (allocator->type == VKD3DSPR_TEMP)
ctx->temp_count = max(ctx->temp_count, ret.id + ret.allocation_size);
return ret;
} }
static const char *debug_register(struct hlsl_reg reg, const struct hlsl_type *type) static const char *debug_register(struct hlsl_reg reg, const struct hlsl_type *type)
@@ -6275,8 +6267,7 @@ static void calculate_resource_register_counts(struct hlsl_ctx *ctx)
} }
} }
static void allocate_instr_temp_register(struct hlsl_ctx *ctx, static void allocate_instr_temp_register(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr)
struct hlsl_ir_node *instr, struct register_allocator *allocator)
{ {
unsigned int dst_writemask = 0; unsigned int dst_writemask = 0;
bool is_per_component = false; bool is_per_component = false;
@@ -6322,8 +6313,6 @@ static void allocate_instr_temp_register(struct hlsl_ctx *ctx,
instr->reg.allocated = true; instr->reg.allocated = true;
instr->reg.type = VKD3DSPR_TEMP; instr->reg.type = VKD3DSPR_TEMP;
instr->reg.id = ctx->temp_count++; instr->reg.id = ctx->temp_count++;
record_allocation(ctx, allocator, ctx->temp_count - 1, VKD3DSP_WRITEMASK_ALL, 1, UINT_MAX, 0, false);
} }
else if (is_per_component) else if (is_per_component)
{ {
@@ -6332,8 +6321,6 @@ static void allocate_instr_temp_register(struct hlsl_ctx *ctx,
instr->reg.allocated = true; instr->reg.allocated = true;
instr->reg.type = VKD3DSPR_TEMP; instr->reg.type = VKD3DSPR_TEMP;
instr->reg.id = ctx->temp_count++; instr->reg.id = ctx->temp_count++;
record_allocation(ctx, allocator, ctx->temp_count - 1, VKD3DSP_WRITEMASK_ALL, 1, UINT_MAX, 0, false);
} }
else else
{ {
@@ -6348,8 +6335,7 @@ static void allocate_instr_temp_register(struct hlsl_ctx *ctx,
debug_register(instr->reg, instr->data_type), instr->index, instr->last_read); debug_register(instr->reg, instr->data_type), instr->index, instr->last_read);
} }
static void allocate_variable_temp_register(struct hlsl_ctx *ctx, static void allocate_variable_temp_register(struct hlsl_ctx *ctx, struct hlsl_ir_var *var)
struct hlsl_ir_var *var, struct register_allocator *allocator)
{ {
struct hlsl_reg *reg = &var->regs[HLSL_REGSET_NUMERIC]; struct hlsl_reg *reg = &var->regs[HLSL_REGSET_NUMERIC];
@@ -6360,7 +6346,7 @@ static void allocate_variable_temp_register(struct hlsl_ctx *ctx,
{ {
if (var->indexable) if (var->indexable)
{ {
reg->id = allocator->indexable_count++; reg->id = ctx->indexable_temp_count++;
reg->allocation_size = 1; reg->allocation_size = 1;
reg->writemask = 0; reg->writemask = 0;
reg->allocated = true; reg->allocated = true;
@@ -6376,9 +6362,6 @@ static void allocate_variable_temp_register(struct hlsl_ctx *ctx,
reg->writemask = vkd3d_write_mask_from_component_count(var->data_type->e.numeric.dimx); reg->writemask = vkd3d_write_mask_from_component_count(var->data_type->e.numeric.dimx);
reg->allocated = true; reg->allocated = true;
for (unsigned int i = 0; i < reg->allocation_size; ++i)
record_allocation(ctx, allocator, ctx->temp_count + i, VKD3DSP_WRITEMASK_ALL, 1, UINT_MAX, 0, false);
ctx->temp_count += reg->allocation_size; ctx->temp_count += reg->allocation_size;
TRACE("Allocated %s to %s (liveness %u-%u).\n", var->name, TRACE("Allocated %s to %s (liveness %u-%u).\n", var->name,
@@ -6387,8 +6370,7 @@ static void allocate_variable_temp_register(struct hlsl_ctx *ctx,
} }
} }
static void allocate_temp_registers_recurse(struct hlsl_ctx *ctx, static void allocate_temp_registers_recurse(struct hlsl_ctx *ctx, struct hlsl_block *block)
struct hlsl_block *block, struct register_allocator *allocator)
{ {
struct hlsl_ir_node *instr; struct hlsl_ir_node *instr;
@@ -6398,15 +6380,15 @@ static void allocate_temp_registers_recurse(struct hlsl_ctx *ctx,
if (ctx->profile->major_version >= 4 && instr->type == HLSL_IR_CONSTANT) if (ctx->profile->major_version >= 4 && instr->type == HLSL_IR_CONSTANT)
continue; continue;
allocate_instr_temp_register(ctx, instr, allocator); allocate_instr_temp_register(ctx, instr);
switch (instr->type) switch (instr->type)
{ {
case HLSL_IR_IF: case HLSL_IR_IF:
{ {
struct hlsl_ir_if *iff = hlsl_ir_if(instr); struct hlsl_ir_if *iff = hlsl_ir_if(instr);
allocate_temp_registers_recurse(ctx, &iff->then_block, allocator); allocate_temp_registers_recurse(ctx, &iff->then_block);
allocate_temp_registers_recurse(ctx, &iff->else_block, allocator); allocate_temp_registers_recurse(ctx, &iff->else_block);
break; break;
} }
@@ -6415,21 +6397,21 @@ static void allocate_temp_registers_recurse(struct hlsl_ctx *ctx,
struct hlsl_ir_load *load = hlsl_ir_load(instr); struct hlsl_ir_load *load = hlsl_ir_load(instr);
/* We need to at least allocate a variable for undefs. /* We need to at least allocate a variable for undefs.
* FIXME: We should probably find a way to remove them instead. */ * FIXME: We should probably find a way to remove them instead. */
allocate_variable_temp_register(ctx, load->src.var, allocator); allocate_variable_temp_register(ctx, load->src.var);
break; break;
} }
case HLSL_IR_LOOP: case HLSL_IR_LOOP:
{ {
struct hlsl_ir_loop *loop = hlsl_ir_loop(instr); struct hlsl_ir_loop *loop = hlsl_ir_loop(instr);
allocate_temp_registers_recurse(ctx, &loop->body, allocator); allocate_temp_registers_recurse(ctx, &loop->body);
break; break;
} }
case HLSL_IR_STORE: case HLSL_IR_STORE:
{ {
struct hlsl_ir_store *store = hlsl_ir_store(instr); struct hlsl_ir_store *store = hlsl_ir_store(instr);
allocate_variable_temp_register(ctx, store->lhs.var, allocator); allocate_variable_temp_register(ctx, store->lhs.var);
break; break;
} }
@@ -6440,7 +6422,7 @@ static void allocate_temp_registers_recurse(struct hlsl_ctx *ctx,
LIST_FOR_EACH_ENTRY(c, &s->cases, struct hlsl_ir_switch_case, entry) LIST_FOR_EACH_ENTRY(c, &s->cases, struct hlsl_ir_switch_case, entry)
{ {
allocate_temp_registers_recurse(ctx, &c->body, allocator); allocate_temp_registers_recurse(ctx, &c->body);
} }
break; break;
} }
@@ -6756,10 +6738,11 @@ static void allocate_const_registers(struct hlsl_ctx *ctx, struct hlsl_block *bo
* does not handle constants. */ * does not handle constants. */
static void allocate_temp_registers(struct hlsl_ctx *ctx, struct hlsl_block *body, struct list *semantic_vars) static void allocate_temp_registers(struct hlsl_ctx *ctx, struct hlsl_block *body, struct list *semantic_vars)
{ {
struct register_allocator allocator = {.type = VKD3DSPR_TEMP};
struct hlsl_scope *scope; struct hlsl_scope *scope;
struct hlsl_ir_var *var; struct hlsl_ir_var *var;
ctx->indexable_temp_count = 0;
/* Reset variable temp register allocations. */ /* Reset variable temp register allocations. */
LIST_FOR_EACH_ENTRY(scope, &ctx->scopes, struct hlsl_scope, entry) LIST_FOR_EACH_ENTRY(scope, &ctx->scopes, struct hlsl_scope, entry)
{ {
@@ -6777,16 +6760,13 @@ static void allocate_temp_registers(struct hlsl_ctx *ctx, struct hlsl_block *bod
{ {
if (var->is_output_semantic) if (var->is_output_semantic)
{ {
record_allocation(ctx, &allocator, 0, VKD3DSP_WRITEMASK_ALL,
var->first_write, UINT_MAX, 0, false);
ctx->temp_count = 1; ctx->temp_count = 1;
break; break;
} }
} }
} }
allocate_temp_registers_recurse(ctx, body, &allocator); allocate_temp_registers_recurse(ctx, body);
vkd3d_free(allocator.allocations);
} }
static enum vkd3d_shader_interpolation_mode sm4_get_interpolation_mode(struct hlsl_type *type, static enum vkd3d_shader_interpolation_mode sm4_get_interpolation_mode(struct hlsl_type *type,