wine/vkd3d
0
0
Fork 0
mirror of https://gitlab.winehq.org/wine/vkd3d.git synced 2025-12-15 08:03:30 -08:00
Code Issues Packages Projects Releases Wiki Activity

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

Browse Source
This commit is contained in:
Petrichor Park
2024-08-19 12:19:36 -05:00
committed by Henri Verbeet
parent 16cb69a324
commit e49beca0d5
Notes: Henri Verbeet 2025-09-22 11:46:20 +02:00 
Approved-by: Francisco Casas (@fcasas)
Approved-by: Elizabeth Figura (@zfigura)
Approved-by: Henri Verbeet (@hverbeet)
Merge-Request: https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/965
7 changed files with 107 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
libs/vkd3d-shader/hlsl_codegen.c
View File

@@ -11062,6 +11062,12 @@ static bool sm4_generate_vsir_instr_expr(struct hlsl_ctx *ctx,
sm4_generate_vsir_expr_with_two_destinations(ctx, program, VSIR_OP_SINCOS, expr, 1);
return true;
case HLSL_OP1_COUNTBITS:
VKD3D_ASSERT(hlsl_type_is_integer(dst_type));
VKD3D_ASSERT(hlsl_version_ge(ctx, 5, 0));
generate_vsir_instr_expr_single_instr_op(ctx, program, expr, VSIR_OP_COUNTBITS, 0, 0, true);
return true;
case HLSL_OP1_DSX:
VKD3D_ASSERT(type_is_float(dst_type));
generate_vsir_instr_expr_single_instr_op(ctx, program, expr, VSIR_OP_DSX, 0, 0, true);
@@ -14097,6 +14103,51 @@ static void loop_unrolling_execute(struct hlsl_ctx *ctx, struct hlsl_block *bloc
hlsl_transform_ir(ctx, resolve_loops, block, NULL);
}
static bool lower_countbits(struct hlsl_ctx *ctx, struct hlsl_ir_node *node, struct hlsl_block *block)
{
struct hlsl_ir_function_decl *func;
struct hlsl_ir_node *call, *rhs;
struct hlsl_ir_expr *expr;
struct hlsl_ir_var *lhs;
char *body;
/* Like vkd3d_popcount(). */
static const char template[] =
"typedef uint%u uintX;\n"
"uintX countbits(uintX v)\n"
"{\n"
" v -= (v >> 1) & 0x55555555;\n"
" v = (v & 0x33333333) + ((v >> 2) & 0x33333333);\n"
" return (((v + (v >> 4)) & 0x0f0f0f0f) * 0x01010101) >> 24;\n"
"}\n";
if (node->type != HLSL_IR_EXPR)
return false;
expr = hlsl_ir_expr(node);
if (expr->op != HLSL_OP1_COUNTBITS)
return false;
rhs = expr->operands[0].node;
if (!(body = hlsl_sprintf_alloc(ctx, template, hlsl_type_component_count(rhs->data_type))))
return false;
func = hlsl_compile_internal_function(ctx, "countbits", body);
vkd3d_free(body);
if (!func)
return false;
lhs = func->parameters.vars[0];
hlsl_block_add_simple_store(ctx, block, lhs, rhs);
if (!(call = hlsl_new_call(ctx, func, &node->loc)))
return false;
hlsl_block_add_instr(block, call);
hlsl_block_add_simple_load(ctx, block, func->return_var, &node->loc);
return true;
}
static bool lower_f16tof32(struct hlsl_ctx *ctx, struct hlsl_ir_node *node, struct hlsl_block *block)
{
struct hlsl_ir_function_decl *func;
@@ -14355,6 +14406,7 @@ static void process_entry_function(struct hlsl_ctx *ctx, struct list *semantic_v
if (hlsl_version_ge(ctx, 4, 0) && hlsl_version_lt(ctx, 5, 0))
{
lower_ir(ctx, lower_countbits, body);
lower_ir(ctx, lower_f16tof32, body);
lower_ir(ctx, lower_f32tof16, body);
}