i965: Use unreachable() instead of unconditional assert().
[mesa.git] / src / mesa / drivers / dri / i965 / brw_lower_texture_gradients.cpp
1 /*
2 * Copyright © 2012 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23
24 /**
25 * \file brw_lower_texture_gradients.cpp
26 */
27
28 #include "glsl/ir.h"
29 #include "glsl/ir_builder.h"
30 #include "program/prog_instruction.h"
31 #include "brw_context.h"
32
33 using namespace ir_builder;
34
35 class lower_texture_grad_visitor : public ir_hierarchical_visitor {
36 public:
37 lower_texture_grad_visitor(bool has_sample_d_c)
38 : has_sample_d_c(has_sample_d_c)
39 {
40 progress = false;
41 }
42
43 ir_visitor_status visit_leave(ir_texture *ir);
44
45
46 bool progress;
47 bool has_sample_d_c;
48
49 private:
50 void emit(ir_variable *, ir_rvalue *);
51 };
52
53 /**
54 * Emit a variable declaration and an assignment to initialize it.
55 */
56 void
57 lower_texture_grad_visitor::emit(ir_variable *var, ir_rvalue *value)
58 {
59 base_ir->insert_before(var);
60 base_ir->insert_before(assign(var, value));
61 }
62
63 static const glsl_type *
64 txs_type(const glsl_type *type)
65 {
66 unsigned dims;
67 switch (type->sampler_dimensionality) {
68 case GLSL_SAMPLER_DIM_1D:
69 dims = 1;
70 break;
71 case GLSL_SAMPLER_DIM_2D:
72 case GLSL_SAMPLER_DIM_RECT:
73 case GLSL_SAMPLER_DIM_CUBE:
74 dims = 2;
75 break;
76 case GLSL_SAMPLER_DIM_3D:
77 dims = 3;
78 break;
79 default:
80 unreachable("Should not get here: invalid sampler dimensionality");
81 }
82
83 if (type->sampler_array)
84 dims++;
85
86 return glsl_type::get_instance(GLSL_TYPE_INT, dims, 1);
87 }
88
89 ir_visitor_status
90 lower_texture_grad_visitor::visit_leave(ir_texture *ir)
91 {
92 /* Only lower textureGrad with shadow samplers */
93 if (ir->op != ir_txd || !ir->shadow_comparitor)
94 return visit_continue;
95
96 /* Lower textureGrad() with samplerCubeShadow even if we have the sample_d_c
97 * message. GLSL provides gradients for the 'r' coordinate. Unfortunately:
98 *
99 * From the Ivybridge PRM, Volume 4, Part 1, sample_d message description:
100 * "The r coordinate contains the faceid, and the r gradients are ignored
101 * by hardware."
102 *
103 * We likely need to do a similar treatment for samplerCube and
104 * samplerCubeArray, but we have insufficient testing for that at the moment.
105 */
106 bool need_lowering = !has_sample_d_c ||
107 ir->sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_CUBE;
108
109 if (!need_lowering)
110 return visit_continue;
111
112 void *mem_ctx = ralloc_parent(ir);
113
114 const glsl_type *grad_type = ir->lod_info.grad.dPdx->type;
115
116 /* Use textureSize() to get the width and height of LOD 0; swizzle away
117 * the depth/number of array slices.
118 */
119 ir_texture *txs = new(mem_ctx) ir_texture(ir_txs);
120 txs->set_sampler(ir->sampler->clone(mem_ctx, NULL),
121 txs_type(ir->sampler->type));
122 txs->lod_info.lod = new(mem_ctx) ir_constant(0);
123 ir_variable *size =
124 new(mem_ctx) ir_variable(grad_type, "size", ir_var_temporary);
125 if (ir->sampler->type->sampler_dimensionality == GLSL_SAMPLER_DIM_CUBE) {
126 base_ir->insert_before(size);
127 base_ir->insert_before(assign(size, expr(ir_unop_i2f, txs), WRITEMASK_XY));
128 base_ir->insert_before(assign(size, new(mem_ctx) ir_constant(1.0f), WRITEMASK_Z));
129 } else {
130 emit(size, expr(ir_unop_i2f,
131 swizzle_for_size(txs, grad_type->vector_elements)));
132 }
133
134 /* Scale the gradients by width and height. Effectively, the incoming
135 * gradients are s'(x,y), t'(x,y), and r'(x,y) from equation 3.19 in the
136 * GL 3.0 spec; we want u'(x,y), which is w_t * s'(x,y).
137 */
138 ir_variable *dPdx =
139 new(mem_ctx) ir_variable(grad_type, "dPdx", ir_var_temporary);
140 emit(dPdx, mul(size, ir->lod_info.grad.dPdx));
141
142 ir_variable *dPdy =
143 new(mem_ctx) ir_variable(grad_type, "dPdy", ir_var_temporary);
144 emit(dPdy, mul(size, ir->lod_info.grad.dPdy));
145
146 /* Calculate rho from equation 3.20 of the GL 3.0 specification. */
147 ir_rvalue *rho;
148 if (dPdx->type->is_scalar()) {
149 rho = expr(ir_binop_max, expr(ir_unop_abs, dPdx),
150 expr(ir_unop_abs, dPdy));
151 } else {
152 rho = expr(ir_binop_max, expr(ir_unop_sqrt, dot(dPdx, dPdx)),
153 expr(ir_unop_sqrt, dot(dPdy, dPdy)));
154 }
155
156 /* lambda_base = log2(rho). We're ignoring GL state biases for now. */
157 ir->op = ir_txl;
158 ir->lod_info.lod = expr(ir_unop_log2, rho);
159
160 progress = true;
161 return visit_continue;
162 }
163
164 extern "C" {
165
166 bool
167 brw_lower_texture_gradients(struct brw_context *brw,
168 struct exec_list *instructions)
169 {
170 bool has_sample_d_c = brw->gen >= 8 || brw->is_haswell;
171 lower_texture_grad_visitor v(has_sample_d_c);
172
173 visit_list_elements(&v, instructions);
174
175 return v.progress;
176 }
177
178 }