vkd3d-shader/hlsl: Support multiple-register variables in object regsets.

Variables that contain more than one object (arrays or structs) require
the allocation of contiguous registers in the respective object
register spaces.
This commit is contained in:
Francisco Casas 2023-04-25 12:41:38 -04:00 committed by Alexandre Julliard
parent 9a8a440b9b
commit 69ff249ef4
Notes: Alexandre Julliard 2023-05-08 22:34:16 +02:00
Approved-by: Giovanni Mascellani (@giomasce)
Approved-by: Zebediah Figura (@zfigura)
Approved-by: Henri Verbeet (@hverbeet)
Approved-by: Alexandre Julliard (@julliard)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/159
6 changed files with 95 additions and 42 deletions

View File

@ -1323,19 +1323,15 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffe
if (!var->semantic.name && var->regs[regset].allocated)
{
put_u32(buffer, 0); /* name */
if (var->data_type->class == HLSL_CLASS_OBJECT
&& (var->data_type->base_type == HLSL_TYPE_SAMPLER
|| var->data_type->base_type == HLSL_TYPE_TEXTURE))
if (regset == HLSL_REGSET_NUMERIC)
{
assert(regset == HLSL_REGSET_SAMPLERS);
put_u32(buffer, vkd3d_make_u32(D3DXRS_SAMPLER, var->regs[regset].id));
put_u32(buffer, 1);
put_u32(buffer, vkd3d_make_u32(D3DXRS_FLOAT4, var->regs[regset].id));
put_u32(buffer, var->data_type->reg_size[regset] / 4);
}
else
{
assert(regset == HLSL_REGSET_NUMERIC);
put_u32(buffer, vkd3d_make_u32(D3DXRS_FLOAT4, var->regs[regset].id));
put_u32(buffer, var->data_type->reg_size[regset] / 4);
put_u32(buffer, vkd3d_make_u32(D3DXRS_SAMPLER, var->regs[regset].id));
put_u32(buffer, var->regs[regset].bind_count);
}
put_u32(buffer, 0); /* type */
put_u32(buffer, 0); /* FIXME: default value */

View File

@ -173,6 +173,9 @@ unsigned int hlsl_get_multiarray_size(const struct hlsl_type *type)
bool hlsl_type_is_resource(const struct hlsl_type *type)
{
if (type->class == HLSL_CLASS_ARRAY)
return hlsl_type_is_resource(type->e.array.type);
if (type->class == HLSL_CLASS_OBJECT)
{
switch (type->base_type)
@ -193,6 +196,9 @@ enum hlsl_regset hlsl_type_get_regset(const struct hlsl_type *type)
if (type->class <= HLSL_CLASS_LAST_NUMERIC)
return HLSL_REGSET_NUMERIC;
if (type->class == HLSL_CLASS_ARRAY)
return hlsl_type_get_regset(type->e.array.type);
if (type->class == HLSL_CLASS_OBJECT)
{
switch (type->base_type)
@ -210,8 +216,6 @@ enum hlsl_regset hlsl_type_get_regset(const struct hlsl_type *type)
vkd3d_unreachable();
}
}
else if (type->class == HLSL_CLASS_ARRAY)
return hlsl_type_get_regset(type->e.array.type);
vkd3d_unreachable();
}

View File

@ -240,16 +240,21 @@ struct hlsl_struct_field
size_t name_bytecode_offset;
};
/* Information of the register allocated for an instruction node or variable.
/* Information of the register(s) allocated for an instruction node or variable.
* These values are initialized at the end of hlsl_emit_bytecode(), after the compilation passes,
* just before writing the bytecode.
* For numeric registers, a writemask can be provided to indicate the reservation of only some of the
* 4 components.
* The type of register (register class) is implied from its use, so it is not stored in this
* struct. */
struct hlsl_reg
{
/* Index of the first register allocated. */
uint32_t id;
/* Number of registers to be allocated.
* Unlike the variable's type's regsize, it is not expressed in register components, but rather
* in whole registers, and may depend on which components are used within the shader. */
uint32_t bind_count;
/* For numeric registers, a writemask can be provided to indicate the reservation of only some
* of the 4 components. */
unsigned int writemask;
/* Whether the register has been allocated. */
bool allocated;

View File

@ -2586,8 +2586,10 @@ static void allocate_register_reservations(struct hlsl_ctx *ctx)
{
var->regs[regset].allocated = true;
var->regs[regset].id = var->reg_reservation.reg_index;
TRACE("Allocated reserved %s to %c%u.\n", var->name, var->reg_reservation.reg_type,
var->reg_reservation.reg_index);
var->regs[regset].bind_count = var->data_type->reg_size[regset];
TRACE("Allocated reserved %s to %c%u-%c%u.\n", var->name, var->reg_reservation.reg_type,
var->reg_reservation.reg_index, var->reg_reservation.reg_type,
var->reg_reservation.reg_index + var->regs[regset].bind_count);
}
}
}
@ -2829,6 +2831,7 @@ static struct hlsl_reg allocate_register(struct hlsl_ctx *ctx, struct register_a
record_allocation(ctx, allocator, reg_idx, writemask, first_write, last_read);
ret.id = reg_idx;
ret.bind_count = 1;
ret.writemask = hlsl_combine_writemasks(writemask, (1u << component_count) - 1);
ret.allocated = true;
return ret;
@ -2864,6 +2867,7 @@ static struct hlsl_reg allocate_range(struct hlsl_ctx *ctx, struct register_allo
record_allocation(ctx, allocator, reg_idx + i, VKD3DSP_WRITEMASK_ALL, first_write, last_read);
ret.id = reg_idx;
ret.bind_count = align(reg_size, 4) / 4;
ret.allocated = true;
return ret;
}
@ -3183,6 +3187,7 @@ static void allocate_semantic_register(struct hlsl_ctx *ctx, struct hlsl_ir_var
{
var->regs[HLSL_REGSET_NUMERIC].allocated = true;
var->regs[HLSL_REGSET_NUMERIC].id = (*counter)++;
var->regs[HLSL_REGSET_NUMERIC].bind_count = 1;
var->regs[HLSL_REGSET_NUMERIC].writemask = (1 << var->data_type->dimx) - 1;
TRACE("Allocated %s to %s.\n", var->name, debug_register(output ? 'o' : 'v',
var->regs[HLSL_REGSET_NUMERIC], var->data_type));
@ -3361,6 +3366,7 @@ static void allocate_buffers(struct hlsl_ctx *ctx)
}
buffer->reg.id = buffer->reservation.reg_index;
buffer->reg.bind_count = 1;
buffer->reg.allocated = true;
TRACE("Allocated reserved %s to cb%u.\n", buffer->name, index);
}
@ -3370,6 +3376,7 @@ static void allocate_buffers(struct hlsl_ctx *ctx)
++index;
buffer->reg.id = index;
buffer->reg.bind_count = 1;
buffer->reg.allocated = true;
TRACE("Allocated %s to cb%u.\n", buffer->name, index);
++index;
@ -3391,13 +3398,17 @@ static const struct hlsl_ir_var *get_allocated_object(struct hlsl_ctx *ctx, enum
uint32_t index)
{
const struct hlsl_ir_var *var;
unsigned int start, count;
LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, const struct hlsl_ir_var, extern_entry)
{
if (!var->regs[regset].allocated)
continue;
if (index == var->regs[regset].id)
start = var->regs[regset].id;
count = var->regs[regset].bind_count;
if (start <= index && index < start + count)
return var;
}
return NULL;
@ -3408,7 +3419,6 @@ static void allocate_objects(struct hlsl_ctx *ctx, enum hlsl_regset regset)
char regset_name = get_regset_name(regset);
struct hlsl_ir_var *var;
uint32_t min_index = 0;
uint32_t index;
if (regset == HLSL_REGSET_UAVS)
{
@ -3420,19 +3430,17 @@ static void allocate_objects(struct hlsl_ctx *ctx, enum hlsl_regset regset)
}
}
index = min_index;
LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry)
{
if (!var->last_read || !var->data_type->reg_size[regset])
unsigned int count = var->regs[regset].bind_count;
if (count == 0)
continue;
if (var->regs[regset].allocated)
{
const struct hlsl_ir_var *reserved_object;
unsigned int index = var->regs[regset].id;
reserved_object = get_allocated_object(ctx, regset, index);
const struct hlsl_ir_var *reserved_object, *last_reported = NULL;
unsigned int index, i;
if (var->regs[regset].id < min_index)
{
@ -3440,28 +3448,44 @@ static void allocate_objects(struct hlsl_ctx *ctx, enum hlsl_regset regset)
hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_OVERLAPPING_RESERVATIONS,
"UAV index (%u) must be higher than the maximum render target index (%u).",
var->regs[regset].id, min_index - 1);
}
else if (reserved_object && reserved_object != var)
{
hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_OVERLAPPING_RESERVATIONS,
"Multiple objects bound to %c%u.", regset_name, index);
hlsl_note(ctx, &reserved_object->loc, VKD3D_SHADER_LOG_ERROR,
"Object '%s' is already bound to %c%u.", reserved_object->name,
regset_name, index);
continue;
}
var->regs[regset].id = var->reg_reservation.reg_index;
var->regs[regset].allocated = true;
TRACE("Allocated reserved %s to %c%u.\n", var->name, regset_name, var->regs[regset].id);
for (i = 0; i < count; ++i)
{
index = var->regs[regset].id + i;
reserved_object = get_allocated_object(ctx, regset, index);
if (reserved_object && reserved_object != var && reserved_object != last_reported)
{
hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_OVERLAPPING_RESERVATIONS,
"Multiple variables bound to %c%u.", regset_name, index);
hlsl_note(ctx, &reserved_object->loc, VKD3D_SHADER_LOG_ERROR,
"Variable '%s' is already bound to %c%u.", reserved_object->name,
regset_name, index);
last_reported = reserved_object;
}
}
}
else
{
while (get_allocated_object(ctx, regset, index))
unsigned int index = min_index;
unsigned int available = 0;
while (available < count)
{
if (get_allocated_object(ctx, regset, index))
available = 0;
else
++available;
++index;
}
index -= count;
var->regs[regset].id = index;
var->regs[regset].allocated = true;
TRACE("Allocated object to %c%u.\n", regset_name, index);
TRACE("Allocated variable %s to %c%u-%c%u.\n", var->name, regset_name, index, regset_name,
index + count);
++index;
}
}
@ -3787,6 +3811,19 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry
rb_for_each_entry(&ctx->functions, dump_function, ctx);
allocate_register_reservations(ctx);
/* For now, request all the registers for each variable, as long as it is used. */
LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry)
{
unsigned int k;
for (k = 0; k <= HLSL_REGSET_LAST_OBJECT; ++k)
{
if (!var->regs[k].allocated)
var->regs[k].bind_count = var->last_read ? var->data_type->reg_size[k] : 0;
}
}
allocate_temp_registers(ctx, entry_func);
if (profile->major_version < 4)
{

View File

@ -2595,6 +2595,9 @@ static void write_sm4_type(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *b
static D3D_SHADER_INPUT_TYPE sm4_resource_type(const struct hlsl_type *type)
{
if (type->class == HLSL_CLASS_ARRAY)
return sm4_resource_type(type->e.array.type);
switch (type->base_type)
{
case HLSL_TYPE_SAMPLER:
@ -2610,6 +2613,9 @@ static D3D_SHADER_INPUT_TYPE sm4_resource_type(const struct hlsl_type *type)
static D3D_RESOURCE_RETURN_TYPE sm4_resource_format(const struct hlsl_type *type)
{
if (type->class == HLSL_CLASS_ARRAY)
return sm4_resource_format(type->e.array.type);
switch (type->e.resource_format->base_type)
{
case HLSL_TYPE_DOUBLE:
@ -2634,6 +2640,9 @@ static D3D_RESOURCE_RETURN_TYPE sm4_resource_format(const struct hlsl_type *type
static D3D_SRV_DIMENSION sm4_rdef_resource_dimension(const struct hlsl_type *type)
{
if (type->class == HLSL_CLASS_ARRAY)
return sm4_rdef_resource_dimension(type->e.array.type);
switch (type->sampler_dim)
{
case HLSL_SAMPLER_DIM_1D:
@ -3037,7 +3046,9 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r
if (var->is_uniform)
{
if (data_type->class == HLSL_CLASS_OBJECT && data_type->base_type == HLSL_TYPE_TEXTURE)
enum hlsl_regset regset = hlsl_type_get_regset(data_type);
if (regset == HLSL_REGSET_TEXTURES)
{
reg->type = VKD3D_SM4_RT_RESOURCE;
reg->dim = VKD3D_SM4_DIMENSION_VEC4;
@ -3047,7 +3058,7 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r
reg->idx_count = 1;
*writemask = VKD3DSP_WRITEMASK_ALL;
}
else if (data_type->class == HLSL_CLASS_OBJECT && data_type->base_type == HLSL_TYPE_UAV)
else if (regset == HLSL_REGSET_UAVS)
{
reg->type = VKD3D_SM5_RT_UAV;
reg->dim = VKD3D_SM4_DIMENSION_VEC4;
@ -3057,7 +3068,7 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r
reg->idx_count = 1;
*writemask = VKD3DSP_WRITEMASK_ALL;
}
else if (data_type->class == HLSL_CLASS_OBJECT && data_type->base_type == HLSL_TYPE_SAMPLER)
else if (regset == HLSL_REGSET_SAMPLERS)
{
reg->type = VKD3D_SM4_RT_SAMPLER;
reg->dim = VKD3D_SM4_DIMENSION_NONE;

View File

@ -80,4 +80,4 @@ float4 main() : sv_target
[test]
draw quad
todo probe all rgba (4.0, 4.0, 4.0, 99.0)
probe all rgba (4.0, 4.0, 4.0, 99.0)