fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-06 17:30:20 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

glsl: add half float support for common functions

Browse source 
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18540>
This commit is contained in:
Timothy Arceri 2022-07-01 20:44:31 +10:00 • committed by Marge Bot
parent eea1c1fa7b
commit c386d56915
2 changed files with 86 additions and 21 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

101
src/compiler/glsl/builtin_functions.cpp
View file

@ -2159,14 +2159,14 @@ builtin_builder::create_builtins()
FHF(log2)
FDHF(sqrt)
FDHF(inversesqrt)
FI64(abs)
FI64(sign)
FD(floor)
FD130(trunc)
FD130GS4(round)
FD130(roundEven)
FD(ceil)
FD(fract)
FI64HF(abs)
FI64HF(sign)
FDHF(floor)
FDHF130(trunc)
FDHF130GS4(round)
FDHF130(roundEven)
FDHF(ceil)
FDHF(fract)
add_function("truncate",
_truncate(gpu_shader4, &glsl_type_builtin_float),
@ -2186,6 +2186,15 @@ builtin_builder::create_builtins()
_mod(always_available, &glsl_type_builtin_vec3, &glsl_type_builtin_vec3),
_mod(always_available, &glsl_type_builtin_vec4, &glsl_type_builtin_vec4),
_mod(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_float16_t),
_mod(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_float16_t),
_mod(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_float16_t),
_mod(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_float16_t),
_mod(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_f16vec2),
_mod(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_f16vec3),
_mod(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_f16vec4),
_mod(fp64, &glsl_type_builtin_double, &glsl_type_builtin_double),
_mod(fp64, &glsl_type_builtin_dvec2, &glsl_type_builtin_double),
_mod(fp64, &glsl_type_builtin_dvec3, &glsl_type_builtin_double),
@ -2196,11 +2205,11 @@ builtin_builder::create_builtins()
_mod(fp64, &glsl_type_builtin_dvec4, &glsl_type_builtin_dvec4),
NULL);
FD130(modf)
FDHF130(modf)
FIUD2_MIXED(min)
FIUD2_MIXED(max)
FIUD2_MIXED(clamp)
FIUDHF2_MIXED(min)
FIUDHF2_MIXED(max)
FIUDHF2_MIXED(clamp)
add_function("mix",
_mix_lrp(always_available, &glsl_type_builtin_float, &glsl_type_builtin_float),
@ -2212,6 +2221,15 @@ builtin_builder::create_builtins()
_mix_lrp(always_available, &glsl_type_builtin_vec3, &glsl_type_builtin_vec3),
_mix_lrp(always_available, &glsl_type_builtin_vec4, &glsl_type_builtin_vec4),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_float16_t),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_float16_t),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_float16_t),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_float16_t),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_f16vec2),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_f16vec3),
_mix_lrp(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_f16vec4),
_mix_lrp(fp64, &glsl_type_builtin_double, &glsl_type_builtin_double),
_mix_lrp(fp64, &glsl_type_builtin_dvec2, &glsl_type_builtin_double),
_mix_lrp(fp64, &glsl_type_builtin_dvec3, &glsl_type_builtin_double),
@ -2226,6 +2244,11 @@ builtin_builder::create_builtins()
_mix_sel(v130, &glsl_type_builtin_vec3, &glsl_type_builtin_bvec3),
_mix_sel(v130, &glsl_type_builtin_vec4, &glsl_type_builtin_bvec4),
_mix_sel(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_bool),
_mix_sel(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_bvec2),
_mix_sel(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_bvec3),
_mix_sel(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_bvec4),
_mix_sel(fp64, &glsl_type_builtin_double, &glsl_type_builtin_bool),
_mix_sel(fp64, &glsl_type_builtin_dvec2, &glsl_type_builtin_bvec2),
_mix_sel(fp64, &glsl_type_builtin_dvec3, &glsl_type_builtin_bvec3),
@ -2274,6 +2297,15 @@ builtin_builder::create_builtins()
_step(fp64, &glsl_type_builtin_dvec2, &glsl_type_builtin_dvec2),
_step(fp64, &glsl_type_builtin_dvec3, &glsl_type_builtin_dvec3),
_step(fp64, &glsl_type_builtin_dvec4, &glsl_type_builtin_dvec4),
_step(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_float16_t),
_step(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_f16vec2),
_step(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_f16vec3),
_step(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_f16vec4),
_step(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_f16vec2),
_step(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_f16vec3),
_step(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_f16vec4),
NULL);
add_function("smoothstep",
@ -2293,10 +2325,19 @@ builtin_builder::create_builtins()
_smoothstep(fp64, &glsl_type_builtin_dvec2, &glsl_type_builtin_dvec2),
_smoothstep(fp64, &glsl_type_builtin_dvec3, &glsl_type_builtin_dvec3),
_smoothstep(fp64, &glsl_type_builtin_dvec4, &glsl_type_builtin_dvec4),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_float16_t),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_f16vec2),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_f16vec3),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_float16_t, &glsl_type_builtin_f16vec4),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_f16vec2, &glsl_type_builtin_f16vec2),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_f16vec3, &glsl_type_builtin_f16vec3),
_smoothstep(gpu_shader_half_float, &glsl_type_builtin_f16vec4, &glsl_type_builtin_f16vec4),
NULL);
FD130(isnan)
FD130(isinf)
FDHF130(isnan)
FDHF130(isinf)
F(floatBitsToInt)
F(floatBitsToUint)
@ -4917,7 +4958,7 @@ builtin_builder::create_builtins()
IU(bitCount)
IU(findLSB)
IU(findMSB)
FDGS5(fma)
FDHFGS5(fma)
add_function("ldexp",
_ldexp(&glsl_type_builtin_float, &glsl_type_builtin_int),
@ -4928,6 +4969,10 @@ builtin_builder::create_builtins()
_ldexp(&glsl_type_builtin_dvec2, &glsl_type_builtin_ivec2),
_ldexp(&glsl_type_builtin_dvec3, &glsl_type_builtin_ivec3),
_ldexp(&glsl_type_builtin_dvec4, &glsl_type_builtin_ivec4),
_ldexp(&glsl_type_builtin_float16_t, &glsl_type_builtin_int),
_ldexp(&glsl_type_builtin_f16vec2, &glsl_type_builtin_ivec2),
_ldexp(&glsl_type_builtin_f16vec3, &glsl_type_builtin_ivec3),
_ldexp(&glsl_type_builtin_f16vec4, &glsl_type_builtin_ivec4),
NULL);
add_function("frexp",
@ -4939,6 +4984,10 @@ builtin_builder::create_builtins()
_frexp(&glsl_type_builtin_dvec2, &glsl_type_builtin_ivec2),
_frexp(&glsl_type_builtin_dvec3, &glsl_type_builtin_ivec3),
_frexp(&glsl_type_builtin_dvec4, &glsl_type_builtin_ivec4),
_frexp(&glsl_type_builtin_float16_t, &glsl_type_builtin_int),
_frexp(&glsl_type_builtin_f16vec2, &glsl_type_builtin_ivec2),
_frexp(&glsl_type_builtin_f16vec3, &glsl_type_builtin_ivec3),
_frexp(&glsl_type_builtin_f16vec4, &glsl_type_builtin_ivec4),
NULL);
add_function("uaddCarry",
_uaddCarry(&glsl_type_builtin_uint),
@ -6356,6 +6405,8 @@ builtin_builder::_step(builtin_available_predicate avail, const glsl_type *edge_
/* Both are floats */
if (glsl_type_is_double(edge_type))
body.emit(assign(t, f2d(b2f(gequal(x, edge)))));
else if (glsl_type_is_float_16(edge_type))
body.emit(assign(t, f2f16(b2f(gequal(x, edge)))));
else
body.emit(assign(t, b2f(gequal(x, edge))));
} else if (edge_type->vector_elements == 1) {
@ -6363,6 +6414,8 @@ builtin_builder::_step(builtin_available_predicate avail, const glsl_type *edge_
for (int i = 0; i < x_type->vector_elements; i++) {
if (glsl_type_is_double(edge_type))
body.emit(assign(t, f2d(b2f(gequal(swizzle(x, i, 1), edge))), 1 << i));
else if (glsl_type_is_float_16(edge_type))
body.emit(assign(t, f2f16(b2f(gequal(swizzle(x, i, 1), edge))), 1 << i));
else
body.emit(assign(t, b2f(gequal(swizzle(x, i, 1), edge)), 1 << i));
}
@ -6372,6 +6425,9 @@ builtin_builder::_step(builtin_available_predicate avail, const glsl_type *edge_
if (glsl_type_is_double(edge_type))
body.emit(assign(t, f2d(b2f(gequal(swizzle(x, i, 1), swizzle(edge, i, 1)))),
1 << i));
else if (glsl_type_is_float_16(edge_type))
body.emit(assign(t, f2f16(b2f(gequal(swizzle(x, i, 1), swizzle(edge, i, 1)))),
1 << i));
else
body.emit(assign(t, b2f(gequal(swizzle(x, i, 1), swizzle(edge, i, 1))),
1 << i));
@ -6427,6 +6483,9 @@ builtin_builder::_isinf(builtin_available_predicate avail, const glsl_type *type
ir_constant_data infinities;
for (int i = 0; i < type->vector_elements; i++) {
switch (type->base_type) {
case GLSL_TYPE_FLOAT16:
infinities.f16[i] = _mesa_float_to_half(INFINITY);
break;
case GLSL_TYPE_FLOAT:
infinities.f[i] = INFINITY;
break;
@ -7947,7 +8006,11 @@ builtin_builder::_ldexp(const glsl_type *x_type, const glsl_type *exp_type)
{
ir_variable *x = in_highp_var(x_type, "x");
ir_variable *y = in_highp_var(exp_type, "y");
MAKE_SIG(x_type, glsl_type_is_double(x_type) ? fp64 : gpu_shader5_or_es31_or_integer_functions, 2, x, y);
builtin_available_predicate avail = glsl_type_is_double(x_type) ? fp64 :
(glsl_type_is_float_16(x_type) ?
gpu_shader_half_float : gpu_shader5_or_es31_or_integer_functions);
MAKE_SIG(x_type, avail, 2, x, y);
sig->return_precision = GLSL_PRECISION_HIGH;
body.emit(ret(expr(ir_binop_ldexp, x, y)));
return sig;
@ -7958,8 +8021,10 @@ builtin_builder::_frexp(const glsl_type *x_type, const glsl_type *exp_type)
{
ir_variable *x = in_highp_var(x_type, "x");
ir_variable *exponent = out_var(exp_type, "exp");
MAKE_SIG(x_type, glsl_type_is_double(x_type) ? fp64 : gpu_shader5_or_es31_or_integer_functions,
2, x, exponent);
builtin_available_predicate avail = glsl_type_is_double(x_type) ? fp64 :
(glsl_type_is_float_16(x_type) ?
gpu_shader_half_float : gpu_shader5_or_es31_or_integer_functions);
MAKE_SIG(x_type, avail, 2, x, exponent);
sig->return_precision = GLSL_PRECISION_HIGH;
body.emit(assign(exponent, expr(ir_unop_frexp_exp, x)));

6
src/compiler/glsl/ir_validate.cpp
View file

@ -736,10 +736,10 @@ ir_validate::visit_leave(ir_expression *ir)
break;
case ir_unop_frexp_sig:
assert(glsl_type_is_float_32_64(ir->operands[0]->type));
assert(glsl_type_is_float_16_32_64(ir->operands[0]->type));
break;
case ir_unop_frexp_exp:
assert(glsl_type_is_float_32_64(ir->operands[0]->type));
assert(glsl_type_is_float_16_32_64(ir->operands[0]->type));
assert(ir->type->base_type == GLSL_TYPE_INT);
break;
case ir_unop_subroutine_to_int:
@ -889,7 +889,7 @@ ir_validate::visit_leave(ir_expression *ir)
case ir_binop_ldexp:
assert(ir->operands[0]->type == ir->type);
assert(glsl_type_is_float_32_64(ir->operands[0]->type));
assert(glsl_type_is_float_16_32_64(ir->operands[0]->type));
assert(ir->operands[1]->type->base_type == GLSL_TYPE_INT);
assert(glsl_get_components(ir->operands[0]->type) ==
glsl_get_components(ir->operands[1]->type));