Merge remote branch 'origin/master' into nv50-compiler
[mesa.git] / src / mesa / drivers / dri / i965 / brw_eu.c
1 /*
2 Copyright (C) Intel Corp. 2006. All Rights Reserved.
3 Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
4 develop this 3D driver.
5
6 Permission is hereby granted, free of charge, to any person obtaining
7 a copy of this software and associated documentation files (the
8 "Software"), to deal in the Software without restriction, including
9 without limitation the rights to use, copy, modify, merge, publish,
10 distribute, sublicense, and/or sell copies of the Software, and to
11 permit persons to whom the Software is furnished to do so, subject to
12 the following conditions:
13
14 The above copyright notice and this permission notice (including the
15 next paragraph) shall be included in all copies or substantial
16 portions of the Software.
17
18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25
26 **********************************************************************/
27 /*
28 * Authors:
29 * Keith Whitwell <keith@tungstengraphics.com>
30 */
31
32
33 #include "brw_context.h"
34 #include "brw_defines.h"
35 #include "brw_eu.h"
36
37
38
39 /* How does predicate control work when execution_size != 8? Do I
40 * need to test/set for 0xffff when execution_size is 16?
41 */
42 void brw_set_predicate_control_flag_value( struct brw_compile *p, GLuint value )
43 {
44 p->current->header.predicate_control = BRW_PREDICATE_NONE;
45
46 if (value != 0xff) {
47 if (value != p->flag_value) {
48 brw_push_insn_state(p);
49 brw_MOV(p, brw_flag_reg(), brw_imm_uw(value));
50 p->flag_value = value;
51 brw_pop_insn_state(p);
52 }
53
54 p->current->header.predicate_control = BRW_PREDICATE_NORMAL;
55 }
56 }
57
58 void brw_set_predicate_control( struct brw_compile *p, GLuint pc )
59 {
60 p->current->header.predicate_control = pc;
61 }
62
63 void brw_set_conditionalmod( struct brw_compile *p, GLuint conditional )
64 {
65 p->current->header.destreg__conditionalmod = conditional;
66 }
67
68 void brw_set_access_mode( struct brw_compile *p, GLuint access_mode )
69 {
70 p->current->header.access_mode = access_mode;
71 }
72
73 void brw_set_compression_control( struct brw_compile *p, GLboolean compression_control )
74 {
75 p->current->header.compression_control = compression_control;
76 }
77
78 void brw_set_mask_control( struct brw_compile *p, GLuint value )
79 {
80 p->current->header.mask_control = value;
81 }
82
83 void brw_set_saturate( struct brw_compile *p, GLuint value )
84 {
85 p->current->header.saturate = value;
86 }
87
88 void brw_set_acc_write_control(struct brw_compile *p, GLuint value)
89 {
90 if (p->brw->intel.gen >= 6)
91 p->current->header.acc_wr_control = value;
92 }
93
94 void brw_push_insn_state( struct brw_compile *p )
95 {
96 assert(p->current != &p->stack[BRW_EU_MAX_INSN_STACK-1]);
97 memcpy(p->current+1, p->current, sizeof(struct brw_instruction));
98 p->current++;
99 }
100
101 void brw_pop_insn_state( struct brw_compile *p )
102 {
103 assert(p->current != p->stack);
104 p->current--;
105 }
106
107
108 /***********************************************************************
109 */
110 void brw_init_compile( struct brw_context *brw, struct brw_compile *p )
111 {
112 p->brw = brw;
113 p->nr_insn = 0;
114 p->current = p->stack;
115 memset(p->current, 0, sizeof(p->current[0]));
116
117 /* Some defaults?
118 */
119 brw_set_mask_control(p, BRW_MASK_ENABLE); /* what does this do? */
120 brw_set_saturate(p, 0);
121 brw_set_compression_control(p, BRW_COMPRESSION_NONE);
122 brw_set_predicate_control_flag_value(p, 0xff);
123 }
124
125
126 const GLuint *brw_get_program( struct brw_compile *p,
127 GLuint *sz )
128 {
129 GLuint i;
130
131 for (i = 0; i < 8; i++)
132 brw_NOP(p);
133
134 *sz = p->nr_insn * sizeof(struct brw_instruction);
135 return (const GLuint *)p->store;
136 }
137
138
139
140 /**
141 * Subroutine calls require special attention.
142 * Mesa instructions may be expanded into multiple hardware instructions
143 * so the prog_instruction::BranchTarget field can't be used as an index
144 * into the hardware instructions.
145 *
146 * The BranchTarget field isn't needed, however. Mesa's GLSL compiler
147 * emits CAL and BGNSUB instructions with labels that can be used to map
148 * subroutine calls to actual subroutine code blocks.
149 *
150 * The structures and function here implement patching of CAL instructions
151 * so they jump to the right subroutine code...
152 */
153
154
155 /**
156 * For each OPCODE_BGNSUB we create one of these.
157 */
158 struct brw_glsl_label
159 {
160 const char *name; /**< the label string */
161 GLuint position; /**< the position of the brw instruction for this label */
162 struct brw_glsl_label *next; /**< next in linked list */
163 };
164
165
166 /**
167 * For each OPCODE_CAL we create one of these.
168 */
169 struct brw_glsl_call
170 {
171 GLuint call_inst_pos; /**< location of the CAL instruction */
172 const char *sub_name; /**< name of subroutine to call */
173 struct brw_glsl_call *next; /**< next in linked list */
174 };
175
176
177 /**
178 * Called for each OPCODE_BGNSUB.
179 */
180 void
181 brw_save_label(struct brw_compile *c, const char *name, GLuint position)
182 {
183 struct brw_glsl_label *label = CALLOC_STRUCT(brw_glsl_label);
184 label->name = name;
185 label->position = position;
186 label->next = c->first_label;
187 c->first_label = label;
188 }
189
190
191 /**
192 * Called for each OPCODE_CAL.
193 */
194 void
195 brw_save_call(struct brw_compile *c, const char *name, GLuint call_pos)
196 {
197 struct brw_glsl_call *call = CALLOC_STRUCT(brw_glsl_call);
198 call->call_inst_pos = call_pos;
199 call->sub_name = name;
200 call->next = c->first_call;
201 c->first_call = call;
202 }
203
204
205 /**
206 * Lookup a label, return label's position/offset.
207 */
208 static GLuint
209 brw_lookup_label(struct brw_compile *c, const char *name)
210 {
211 const struct brw_glsl_label *label;
212 for (label = c->first_label; label; label = label->next) {
213 if (strcmp(name, label->name) == 0) {
214 return label->position;
215 }
216 }
217 abort(); /* should never happen */
218 return ~0;
219 }
220
221
222 /**
223 * When we're done generating code, this function is called to resolve
224 * subroutine calls.
225 */
226 void
227 brw_resolve_cals(struct brw_compile *c)
228 {
229 const struct brw_glsl_call *call;
230
231 for (call = c->first_call; call; call = call->next) {
232 const GLuint sub_loc = brw_lookup_label(c, call->sub_name);
233 struct brw_instruction *brw_call_inst = &c->store[call->call_inst_pos];
234 struct brw_instruction *brw_sub_inst = &c->store[sub_loc];
235 GLint offset = brw_sub_inst - brw_call_inst;
236
237 /* patch brw_inst1 to point to brw_inst2 */
238 brw_set_src1(brw_call_inst, brw_imm_d(offset * 16));
239 }
240
241 /* free linked list of calls */
242 {
243 struct brw_glsl_call *call, *next;
244 for (call = c->first_call; call; call = next) {
245 next = call->next;
246 free(call);
247 }
248 c->first_call = NULL;
249 }
250
251 /* free linked list of labels */
252 {
253 struct brw_glsl_label *label, *next;
254 for (label = c->first_label; label; label = next) {
255 next = label->next;
256 free(label);
257 }
258 c->first_label = NULL;
259 }
260 }