static nir_ssa_def *
build_atan2(nir_builder *b, nir_ssa_def *y, nir_ssa_def *x)
{
nir_ssa_def *zero = nir_imm_float(b, 0.0f);
/* If |x| >= 1.0e-8 * |y|: */
nir_ssa_def *condition =
nir_fge(b, nir_fabs(b, x),
nir_fmul(b, nir_imm_float(b, 1.0e-8f), nir_fabs(b, y)));
/* Then...call atan(y/x) and fix it up: */
nir_ssa_def *atan1 = build_atan(b, nir_fdiv(b, y, x));
nir_ssa_def *r_then =
nir_bcsel(b, nir_flt(b, x, zero),
nir_fadd(b, atan1,
nir_bcsel(b, nir_fge(b, y, zero),
nir_imm_float(b, M_PIf),
nir_imm_float(b, -M_PIf))),
atan1);
/* Else... */
nir_ssa_def *r_else =
nir_fmul(b, nir_fsign(b, y), nir_imm_float(b, M_PI_2f));
return nir_bcsel(b, condition, r_then, r_else);
}
static nir_ssa_def *
build_atan(nir_builder *b, nir_ssa_def *y_over_x)
{
nir_ssa_def *abs_y_over_x = nir_fabs(b, y_over_x);
nir_ssa_def *one = nir_imm_float(b, 1.0f);
/*
* range-reduction, first step:
*
* / y_over_x if |y_over_x| <= 1.0;
* x = <
* \ 1.0 / y_over_x otherwise
*/
nir_ssa_def *x = nir_fdiv(b, nir_fmin(b, abs_y_over_x, one),
nir_fmax(b, abs_y_over_x, one));
/*
* approximate atan by evaluating polynomial:
*
* x * 0.9999793128310355 - x^3 * 0.3326756418091246 +
* x^5 * 0.1938924977115610 - x^7 * 0.1173503194786851 +
* x^9 * 0.0536813784310406 - x^11 * 0.0121323213173444
*/
nir_ssa_def *x_2 = nir_fmul(b, x, x);
nir_ssa_def *x_3 = nir_fmul(b, x_2, x);
nir_ssa_def *x_5 = nir_fmul(b, x_3, x_2);
nir_ssa_def *x_7 = nir_fmul(b, x_5, x_2);
nir_ssa_def *x_9 = nir_fmul(b, x_7, x_2);
nir_ssa_def *x_11 = nir_fmul(b, x_9, x_2);
nir_ssa_def *polynomial_terms[] = {
nir_fmul(b, x, nir_imm_float(b, 0.9999793128310355f)),
nir_fmul(b, x_3, nir_imm_float(b, -0.3326756418091246f)),
nir_fmul(b, x_5, nir_imm_float(b, 0.1938924977115610f)),
nir_fmul(b, x_7, nir_imm_float(b, -0.1173503194786851f)),
nir_fmul(b, x_9, nir_imm_float(b, 0.0536813784310406f)),
nir_fmul(b, x_11, nir_imm_float(b, -0.0121323213173444f)),
};
nir_ssa_def *tmp =
build_fsum(b, polynomial_terms, ARRAY_SIZE(polynomial_terms));
/* range-reduction fixup */
tmp = nir_fadd(b, tmp,
nir_fmul(b,
nir_b2f(b, nir_flt(b, one, abs_y_over_x)),
nir_fadd(b, nir_fmul(b, tmp,
nir_imm_float(b, -2.0f)),
nir_imm_float(b, M_PI_2f))));
/* sign fixup */
return nir_fmul(b, tmp, nir_fsign(b, y_over_x));
}
static nir_shader *
build_resolve_compute_shader(struct radv_device *dev, bool is_integer, int samples)
{
nir_builder b;
char name[64];
nir_if *outer_if = NULL;
const struct glsl_type *sampler_type = glsl_sampler_type(GLSL_SAMPLER_DIM_MS,
false,
false,
GLSL_TYPE_FLOAT);
const struct glsl_type *img_type = glsl_sampler_type(GLSL_SAMPLER_DIM_2D,
false,
false,
GLSL_TYPE_FLOAT);
snprintf(name, 64, "meta_resolve_cs-%d-%s", samples, is_integer ? "int" : "float");
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_COMPUTE, NULL);
b.shader->info->name = ralloc_strdup(b.shader, name);
b.shader->info->cs.local_size[0] = 16;
b.shader->info->cs.local_size[1] = 16;
b.shader->info->cs.local_size[2] = 1;
nir_variable *input_img = nir_variable_create(b.shader, nir_var_uniform,
sampler_type, "s_tex");
input_img->data.descriptor_set = 0;
input_img->data.binding = 0;
nir_variable *output_img = nir_variable_create(b.shader, nir_var_uniform,
img_type, "out_img");
output_img->data.descriptor_set = 0;
output_img->data.binding = 1;
nir_ssa_def *invoc_id = nir_load_system_value(&b, nir_intrinsic_load_local_invocation_id, 0);
nir_ssa_def *wg_id = nir_load_system_value(&b, nir_intrinsic_load_work_group_id, 0);
nir_ssa_def *block_size = nir_imm_ivec4(&b,
b.shader->info->cs.local_size[0],
b.shader->info->cs.local_size[1],
b.shader->info->cs.local_size[2], 0);
nir_ssa_def *global_id = nir_iadd(&b, nir_imul(&b, wg_id, block_size), invoc_id);
nir_intrinsic_instr *src_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
src_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 0));
src_offset->num_components = 2;
nir_ssa_dest_init(&src_offset->instr, &src_offset->dest, 2, 32, "src_offset");
nir_builder_instr_insert(&b, &src_offset->instr);
nir_intrinsic_instr *dst_offset = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
dst_offset->src[0] = nir_src_for_ssa(nir_imm_int(&b, 8));
dst_offset->num_components = 2;
nir_ssa_dest_init(&dst_offset->instr, &dst_offset->dest, 2, 32, "dst_offset");
nir_builder_instr_insert(&b, &dst_offset->instr);
nir_ssa_def *img_coord = nir_iadd(&b, global_id, &src_offset->dest.ssa);
/* do a txf_ms on each sample */
nir_ssa_def *tmp;
nir_tex_instr *tex = nir_tex_instr_create(b.shader, 2);
tex->sampler_dim = GLSL_SAMPLER_DIM_MS;
tex->op = nir_texop_txf_ms;
tex->src[0].src_type = nir_tex_src_coord;
tex->src[0].src = nir_src_for_ssa(img_coord);
tex->src[1].src_type = nir_tex_src_ms_index;
tex->src[1].src = nir_src_for_ssa(nir_imm_int(&b, 0));
tex->dest_type = nir_type_float;
tex->is_array = false;
tex->coord_components = 2;
tex->texture = nir_deref_var_create(tex, input_img);
tex->sampler = NULL;
nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
nir_builder_instr_insert(&b, &tex->instr);
tmp = &tex->dest.ssa;
nir_variable *color =
nir_local_variable_create(b.impl, glsl_vec4_type(), "color");
if (!is_integer && samples > 1) {
nir_tex_instr *tex_all_same = nir_tex_instr_create(b.shader, 1);
tex_all_same->sampler_dim = GLSL_SAMPLER_DIM_MS;
tex_all_same->op = nir_texop_samples_identical;
tex_all_same->src[0].src_type = nir_tex_src_coord;
tex_all_same->src[0].src = nir_src_for_ssa(img_coord);
tex_all_same->dest_type = nir_type_float;
tex_all_same->is_array = false;
tex_all_same->coord_components = 2;
tex_all_same->texture = nir_deref_var_create(tex_all_same, input_img);
tex_all_same->sampler = NULL;
nir_ssa_dest_init(&tex_all_same->instr, &tex_all_same->dest, 1, 32, "tex");
nir_builder_instr_insert(&b, &tex_all_same->instr);
nir_ssa_def *all_same = nir_ine(&b, &tex_all_same->dest.ssa, nir_imm_int(&b, 0));
nir_if *if_stmt = nir_if_create(b.shader);
if_stmt->condition = nir_src_for_ssa(all_same);
nir_cf_node_insert(b.cursor, &if_stmt->cf_node);
b.cursor = nir_after_cf_list(&if_stmt->then_list);
for (int i = 1; i < samples; i++) {
nir_tex_instr *tex_add = nir_tex_instr_create(b.shader, 2);
tex_add->sampler_dim = GLSL_SAMPLER_DIM_MS;
tex_add->op = nir_texop_txf_ms;
tex_add->src[0].src_type = nir_tex_src_coord;
tex_add->src[0].src = nir_src_for_ssa(img_coord);
tex_add->src[1].src_type = nir_tex_src_ms_index;
tex_add->src[1].src = nir_src_for_ssa(nir_imm_int(&b, i));
tex_add->dest_type = nir_type_float;
tex_add->is_array = false;
tex_add->coord_components = 2;
tex_add->texture = nir_deref_var_create(tex_add, input_img);
tex_add->sampler = NULL;
nir_ssa_dest_init(&tex_add->instr, &tex_add->dest, 4, 32, "tex");
nir_builder_instr_insert(&b, &tex_add->instr);
tmp = nir_fadd(&b, tmp, &tex_add->dest.ssa);
}
tmp = nir_fdiv(&b, tmp, nir_imm_float(&b, samples));
nir_store_var(&b, color, tmp, 0xf);
b.cursor = nir_after_cf_list(&if_stmt->else_list);
outer_if = if_stmt;
}
nir_store_var(&b, color, &tex->dest.ssa, 0xf);
if (outer_if)
b.cursor = nir_after_cf_node(&outer_if->cf_node);
nir_ssa_def *newv = nir_load_var(&b, color);
nir_ssa_def *coord = nir_iadd(&b, global_id, &dst_offset->dest.ssa);
nir_intrinsic_instr *store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_image_store);
store->src[0] = nir_src_for_ssa(coord);
store->src[1] = nir_src_for_ssa(nir_ssa_undef(&b, 1, 32));
store->src[2] = nir_src_for_ssa(newv);
store->variables[0] = nir_deref_var_create(store, output_img);
nir_builder_instr_insert(&b, &store->instr);
return b.shader;
}
static void
handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint,
const uint32_t *w, unsigned count)
{
struct nir_builder *nb = &b->nb;
const struct glsl_type *dest_type =
vtn_value(b, w[1], vtn_value_type_type)->type->type;
struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_ssa);
val->ssa = vtn_create_ssa_value(b, dest_type);
/* Collect the various SSA sources */
unsigned num_inputs = count - 5;
nir_ssa_def *src[3] = { NULL, };
for (unsigned i = 0; i < num_inputs; i++)
src[i] = vtn_ssa_value(b, w[i + 5])->def;
switch (entrypoint) {
case GLSLstd450Radians:
val->ssa->def = nir_fmul(nb, src[0], nir_imm_float(nb, 0.01745329251));
return;
case GLSLstd450Degrees:
val->ssa->def = nir_fmul(nb, src[0], nir_imm_float(nb, 57.2957795131));
return;
case GLSLstd450Tan:
val->ssa->def = nir_fdiv(nb, nir_fsin(nb, src[0]),
nir_fcos(nb, src[0]));
return;
case GLSLstd450Modf: {
nir_ssa_def *sign = nir_fsign(nb, src[0]);
nir_ssa_def *abs = nir_fabs(nb, src[0]);
val->ssa->def = nir_fmul(nb, sign, nir_ffract(nb, abs));
nir_store_deref_var(nb, vtn_nir_deref(b, w[6]),
nir_fmul(nb, sign, nir_ffloor(nb, abs)), 0xf);
return;
}
case GLSLstd450ModfStruct: {
nir_ssa_def *sign = nir_fsign(nb, src[0]);
nir_ssa_def *abs = nir_fabs(nb, src[0]);
assert(glsl_type_is_struct(val->ssa->type));
val->ssa->elems[0]->def = nir_fmul(nb, sign, nir_ffract(nb, abs));
val->ssa->elems[1]->def = nir_fmul(nb, sign, nir_ffloor(nb, abs));
return;
}
case GLSLstd450Step:
val->ssa->def = nir_sge(nb, src[1], src[0]);
return;
case GLSLstd450Length:
val->ssa->def = build_length(nb, src[0]);
return;
case GLSLstd450Distance:
val->ssa->def = build_length(nb, nir_fsub(nb, src[0], src[1]));
return;
case GLSLstd450Normalize:
val->ssa->def = nir_fdiv(nb, src[0], build_length(nb, src[0]));
return;
case GLSLstd450Exp:
val->ssa->def = build_exp(nb, src[0]);
return;
case GLSLstd450Log:
val->ssa->def = build_log(nb, src[0]);
return;
case GLSLstd450FClamp:
val->ssa->def = build_fclamp(nb, src[0], src[1], src[2]);
return;
case GLSLstd450UClamp:
val->ssa->def = nir_umin(nb, nir_umax(nb, src[0], src[1]), src[2]);
return;
case GLSLstd450SClamp:
val->ssa->def = nir_imin(nb, nir_imax(nb, src[0], src[1]), src[2]);
return;
case GLSLstd450Cross: {
unsigned yzx[4] = { 1, 2, 0, 0 };
unsigned zxy[4] = { 2, 0, 1, 0 };
val->ssa->def =
nir_fsub(nb, nir_fmul(nb, nir_swizzle(nb, src[0], yzx, 3, true),
nir_swizzle(nb, src[1], zxy, 3, true)),
nir_fmul(nb, nir_swizzle(nb, src[0], zxy, 3, true),
nir_swizzle(nb, src[1], yzx, 3, true)));
return;
}
case GLSLstd450SmoothStep: {
/* t = clamp((x - edge0) / (edge1 - edge0), 0, 1) */
nir_ssa_def *t =
build_fclamp(nb, nir_fdiv(nb, nir_fsub(nb, src[2], src[0]),
nir_fsub(nb, src[1], src[0])),
nir_imm_float(nb, 0.0), nir_imm_float(nb, 1.0));
/* result = t * t * (3 - 2 * t) */
val->ssa->def =
nir_fmul(nb, t, nir_fmul(nb, t,
nir_fsub(nb, nir_imm_float(nb, 3.0),
nir_fmul(nb, nir_imm_float(nb, 2.0), t))));
return;
}
case GLSLstd450FaceForward:
val->ssa->def =
nir_bcsel(nb, nir_flt(nb, nir_fdot(nb, src[2], src[1]),
nir_imm_float(nb, 0.0)),
src[0], nir_fneg(nb, src[0]));
return;
case GLSLstd450Reflect:
/* I - 2 * dot(N, I) * N */
val->ssa->def =
nir_fsub(nb, src[0], nir_fmul(nb, nir_imm_float(nb, 2.0),
nir_fmul(nb, nir_fdot(nb, src[0], src[1]),
src[1])));
return;
case GLSLstd450Refract: {
nir_ssa_def *I = src[0];
nir_ssa_def *N = src[1];
nir_ssa_def *eta = src[2];
nir_ssa_def *n_dot_i = nir_fdot(nb, N, I);
nir_ssa_def *one = nir_imm_float(nb, 1.0);
nir_ssa_def *zero = nir_imm_float(nb, 0.0);
/* k = 1.0 - eta * eta * (1.0 - dot(N, I) * dot(N, I)) */
nir_ssa_def *k =
nir_fsub(nb, one, nir_fmul(nb, eta, nir_fmul(nb, eta,
nir_fsub(nb, one, nir_fmul(nb, n_dot_i, n_dot_i)))));
nir_ssa_def *result =
nir_fsub(nb, nir_fmul(nb, eta, I),
nir_fmul(nb, nir_fadd(nb, nir_fmul(nb, eta, n_dot_i),
nir_fsqrt(nb, k)), N));
/* XXX: bcsel, or if statement? */
val->ssa->def = nir_bcsel(nb, nir_flt(nb, k, zero), zero, result);
return;
}
case GLSLstd450Sinh:
/* 0.5 * (e^x - e^(-x)) */
val->ssa->def =
nir_fmul(nb, nir_imm_float(nb, 0.5f),
nir_fsub(nb, build_exp(nb, src[0]),
build_exp(nb, nir_fneg(nb, src[0]))));
return;
case GLSLstd450Cosh:
/* 0.5 * (e^x + e^(-x)) */
val->ssa->def =
nir_fmul(nb, nir_imm_float(nb, 0.5f),
nir_fadd(nb, build_exp(nb, src[0]),
build_exp(nb, nir_fneg(nb, src[0]))));
return;
case GLSLstd450Tanh:
/* (0.5 * (e^x - e^(-x))) / (0.5 * (e^x + e^(-x))) */
val->ssa->def =
nir_fdiv(nb, nir_fmul(nb, nir_imm_float(nb, 0.5f),
nir_fsub(nb, build_exp(nb, src[0]),
build_exp(nb, nir_fneg(nb, src[0])))),
nir_fmul(nb, nir_imm_float(nb, 0.5f),
nir_fadd(nb, build_exp(nb, src[0]),
build_exp(nb, nir_fneg(nb, src[0])))));
return;
case GLSLstd450Asinh:
val->ssa->def = nir_fmul(nb, nir_fsign(nb, src[0]),
build_log(nb, nir_fadd(nb, nir_fabs(nb, src[0]),
nir_fsqrt(nb, nir_fadd(nb, nir_fmul(nb, src[0], src[0]),
nir_imm_float(nb, 1.0f))))));
return;
case GLSLstd450Acosh:
val->ssa->def = build_log(nb, nir_fadd(nb, src[0],
nir_fsqrt(nb, nir_fsub(nb, nir_fmul(nb, src[0], src[0]),
nir_imm_float(nb, 1.0f)))));
return;
case GLSLstd450Atanh: {
nir_ssa_def *one = nir_imm_float(nb, 1.0);
val->ssa->def = nir_fmul(nb, nir_imm_float(nb, 0.5f),
build_log(nb, nir_fdiv(nb, nir_fadd(nb, one, src[0]),
nir_fsub(nb, one, src[0]))));
return;
}
case GLSLstd450Asin:
val->ssa->def = build_asin(nb, src[0], 0.086566724, -0.03102955);
return;
case GLSLstd450Acos:
val->ssa->def = nir_fsub(nb, nir_imm_float(nb, M_PI_2f),
build_asin(nb, src[0], 0.08132463, -0.02363318));
return;
case GLSLstd450Atan:
val->ssa->def = build_atan(nb, src[0]);
return;
case GLSLstd450Atan2:
val->ssa->def = build_atan2(nb, src[0], src[1]);
return;
case GLSLstd450Frexp: {
nir_ssa_def *exponent;
val->ssa->def = build_frexp(nb, src[0], &exponent);
nir_store_deref_var(nb, vtn_nir_deref(b, w[6]), exponent, 0xf);
return;
}
case GLSLstd450FrexpStruct: {
assert(glsl_type_is_struct(val->ssa->type));
val->ssa->elems[0]->def = build_frexp(nb, src[0],
&val->ssa->elems[1]->def);
return;
}
default:
val->ssa->def =
nir_build_alu(&b->nb, vtn_nir_alu_op_for_spirv_glsl_opcode(entrypoint),
src[0], src[1], src[2], NULL);
return;
}
}
