wine/vkd3d
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/vkd3d.git synced 2025-01-28 13:05:02 -08:00
Code Issues Packages Projects Releases Wiki Activity

vkd3d-shader/hlsl: Implement the InterlockedCompareExchange() intrinsic.

Browse Source
This commit is contained in:
Shaun Ren 2024-12-20 15:33:28 -05:00 committed by Henri Verbeet
parent 3b19a4aaf3
commit 22ab08f4d8
Notes: Henri Verbeet 2025-01-20 16:18:51 +01:00 
Approved-by: Henri Verbeet (@hverbeet)
Approved-by: Elizabeth Figura (@zfigura)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/1330
5 changed files with 48 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
libs/vkd3d-shader/hlsl.c
View File

@ -3508,6 +3508,7 @@ static void dump_ir_interlocked(struct vkd3d_string_buffer *buffer, const struct
{
[HLSL_INTERLOCKED_ADD] = "add",
[HLSL_INTERLOCKED_AND] = "and",
[HLSL_INTERLOCKED_CMP_EXCH] = "cmp_exch",
};
VKD3D_ASSERT(interlocked->op < ARRAY_SIZE(op_names));
@ -3515,6 +3516,11 @@ static void dump_ir_interlocked(struct vkd3d_string_buffer *buffer, const struct
dump_deref(buffer, &interlocked->dst);
vkd3d_string_buffer_printf(buffer, ", coords = ");
dump_src(buffer, &interlocked->coords);
if (interlocked->cmp_value.node)
{
vkd3d_string_buffer_printf(buffer, ", cmp_value = ");
dump_src(buffer, &interlocked->cmp_value);
}
vkd3d_string_buffer_printf(buffer, ", value = ");
dump_src(buffer, &interlocked->value);
vkd3d_string_buffer_printf(buffer, ")");

1
libs/vkd3d-shader/hlsl.h
View File

@ -955,6 +955,7 @@ enum hlsl_interlocked_op
{
HLSL_INTERLOCKED_ADD,
HLSL_INTERLOCKED_AND,
HLSL_INTERLOCKED_CMP_EXCH,
};
/* Represents an interlocked operation.

33
libs/vkd3d-shader/hlsl.y
View File

@ -5378,7 +5378,7 @@ static bool intrinsic_GetRenderTargetSampleCount(struct hlsl_ctx *ctx,
static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op op,
const struct parse_initializer *params, const struct vkd3d_shader_location *loc, const char *name)
{
struct hlsl_ir_node *lhs, *coords, *val, *orig_val = NULL;
struct hlsl_ir_node *lhs, *coords, *val, *cmp_val = NULL, *orig_val = NULL;
struct hlsl_ir_node *interlocked, *void_ret;
struct hlsl_type *lhs_type, *val_type;
struct vkd3d_string_buffer *string;
@ -5388,7 +5388,7 @@ static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op
hlsl_error(ctx, loc, VKD3D_SHADER_ERROR_HLSL_INCOMPATIBLE_PROFILE,
"Interlocked functions can only be used in shader model 5.0 or higher.");
if (params->args_count != 2 && params->args_count != 3)
if (op != HLSL_INTERLOCKED_CMP_EXCH && params->args_count != 2 && params->args_count != 3)
{
hlsl_error(ctx, loc, VKD3D_SHADER_ERROR_HLSL_WRONG_PARAMETER_COUNT,
"Unexpected number of arguments to function '%s': expected 2 or 3, but got %u.",
@ -5399,6 +5399,19 @@ static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op
lhs = params->args[0];
lhs_type = lhs->data_type;
if (op == HLSL_INTERLOCKED_CMP_EXCH)
{
cmp_val = params->args[1];
val = params->args[2];
orig_val = params->args[3];
}
else
{
val = params->args[1];
if (params->args_count == 3)
orig_val = params->args[2];
}
if (lhs_type->class != HLSL_CLASS_SCALAR || (lhs_type->e.numeric.type != HLSL_TYPE_UINT
&& lhs_type->e.numeric.type != HLSL_TYPE_INT))
{
@ -5415,11 +5428,10 @@ static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op
/* Interlocked*() functions always take uint for the value parameters,
* except for InterlockedMax()/InterlockedMin(). */
val_type = hlsl_get_scalar_type(ctx, HLSL_TYPE_UINT);
if (!(val = add_implicit_conversion(ctx, params->instrs, params->args[1], val_type, loc)))
if (cmp_val && !(cmp_val = add_implicit_conversion(ctx, params->instrs, cmp_val, val_type, loc)))
return false;
if (!(val = add_implicit_conversion(ctx, params->instrs, val, val_type, loc)))
return false;
if (params->args_count == 3)
orig_val = params->args[2];
/* TODO: groupshared variables */
if (lhs->type == HLSL_IR_INDEX && hlsl_index_chain_has_resource_access(hlsl_ir_index(lhs)))
@ -5449,7 +5461,7 @@ static bool intrinsic_interlocked(struct hlsl_ctx *ctx, enum hlsl_interlocked_op
return false;
}
interlocked = hlsl_new_interlocked(ctx, op, orig_val ? lhs_type : NULL, &dst_deref, coords, NULL, val, loc);
interlocked = hlsl_new_interlocked(ctx, op, orig_val ? lhs_type : NULL, &dst_deref, coords, cmp_val, val, loc);
hlsl_cleanup_deref(&dst_deref);
if (!interlocked)
return false;
@ -5484,6 +5496,12 @@ static bool intrinsic_InterlockedAnd(struct hlsl_ctx *ctx,
return intrinsic_interlocked(ctx, HLSL_INTERLOCKED_AND, params, loc, "InterlockedAnd");
}
static bool intrinsic_InterlockedCompareExchange(struct hlsl_ctx *ctx,
const struct parse_initializer *params, const struct vkd3d_shader_location *loc)
{
return intrinsic_interlocked(ctx, HLSL_INTERLOCKED_CMP_EXCH, params, loc, "InterlockedCompareExchange");
}
static const struct intrinsic_function
{
const char *name;
@ -5499,6 +5517,7 @@ intrinsic_functions[] =
{"GetRenderTargetSampleCount", 0, true, intrinsic_GetRenderTargetSampleCount},
{"InterlockedAdd", -1, true, intrinsic_InterlockedAdd},
{"InterlockedAnd", -1, true, intrinsic_InterlockedAnd},
{"InterlockedCompareExchange", 4, true, intrinsic_InterlockedCompareExchange},
{"abs", 1, true, intrinsic_abs},
{"acos", 1, true, intrinsic_acos},
{"all", 1, true, intrinsic_all},

17
libs/vkd3d-shader/hlsl_codegen.c
View File

@ -10008,9 +10008,11 @@ static bool sm4_generate_vsir_instr_interlocked(struct hlsl_ctx *ctx,
{
[HLSL_INTERLOCKED_ADD] = VKD3DSIH_IMM_ATOMIC_IADD,
[HLSL_INTERLOCKED_AND] = VKD3DSIH_IMM_ATOMIC_AND,
[HLSL_INTERLOCKED_CMP_EXCH] = VKD3DSIH_IMM_ATOMIC_CMP_EXCH,
};
struct hlsl_ir_node *coords = interlocked->coords.node, *value = interlocked->value.node;
struct hlsl_ir_node *cmp_value = interlocked->cmp_value.node, *value = interlocked->value.node;
struct hlsl_ir_node *coords = interlocked->coords.node;
struct hlsl_ir_node *instr = &interlocked->node;
bool is_imm = interlocked->node.reg.allocated;
struct vkd3d_shader_dst_param *dst_param;
@ -10018,7 +10020,8 @@ static bool sm4_generate_vsir_instr_interlocked(struct hlsl_ctx *ctx,
enum vkd3d_shader_opcode opcode;
opcode = is_imm ? imm_opcodes[interlocked->op] : opcodes[interlocked->op];
if (!(ins = generate_vsir_add_program_instruction(ctx, program, &instr->loc, opcode, is_imm ? 2 : 1, 2)))
if (!(ins = generate_vsir_add_program_instruction(ctx, program, &instr->loc, opcode,
is_imm ? 2 : 1, cmp_value ? 3 : 2)))
return false;
if (is_imm)
@ -10030,7 +10033,15 @@ static bool sm4_generate_vsir_instr_interlocked(struct hlsl_ctx *ctx,
dst_param->reg.dimension = VSIR_DIMENSION_NONE;
vsir_src_from_hlsl_node(&ins->src[0], ctx, coords, VKD3DSP_WRITEMASK_ALL);
vsir_src_from_hlsl_node(&ins->src[1], ctx, value, VKD3DSP_WRITEMASK_ALL);
if (cmp_value)
{
vsir_src_from_hlsl_node(&ins->src[1], ctx, cmp_value, VKD3DSP_WRITEMASK_ALL);
vsir_src_from_hlsl_node(&ins->src[2], ctx, value, VKD3DSP_WRITEMASK_ALL);
}
else
{
vsir_src_from_hlsl_node(&ins->src[1], ctx, value, VKD3DSP_WRITEMASK_ALL);
}
return true;
}

1
libs/vkd3d-shader/tpf.c
View File

@ -4071,6 +4071,7 @@ static void tpf_handle_instruction(struct tpf_compiler *tpf, const struct vkd3d_
case VKD3DSIH_IMAX:
case VKD3DSIH_IMIN:
case VKD3DSIH_IMM_ATOMIC_AND:
case VKD3DSIH_IMM_ATOMIC_CMP_EXCH:
case VKD3DSIH_IMM_ATOMIC_IADD:
case VKD3DSIH_IMUL:
case VKD3DSIH_INE: