1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 */
23 #include "r600_asm.h"
24 #include "r600_opcodes.h"
25 #include "r600_shader_common.h"
26
27 #include "util/u_memory.h"
28 #include "eg_sq.h"
29 #include <errno.h>
30
eg_bytecode_cf_build(struct r600_bytecode * bc,struct r600_bytecode_cf * cf)31 int eg_bytecode_cf_build(struct r600_bytecode *bc, struct r600_bytecode_cf *cf)
32 {
33 unsigned id = cf->id;
34
35 if (cf->op == CF_NATIVE) {
36 bc->bytecode[id++] = cf->isa[0];
37 bc->bytecode[id++] = cf->isa[1];
38 } else {
39 const struct cf_op_info *cfop = r600_isa_cf(cf->op);
40 unsigned opcode = r600_isa_cf_opcode(bc->isa->hw_class, cf->op);
41 if (cfop->flags & CF_ALU) { /* ALU clauses */
42
43 /* prepend ALU_EXTENDED if we need more than 2 kcache sets */
44 if (cf->eg_alu_extended) {
45 bc->bytecode[id++] =
46 S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE0(cf->kcache[0].index_mode) |
47 S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE1(cf->kcache[1].index_mode) |
48 S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE2(cf->kcache[2].index_mode) |
49 S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE3(cf->kcache[3].index_mode) |
50 S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK2(cf->kcache[2].bank) |
51 S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK3(cf->kcache[3].bank) |
52 S_SQ_CF_ALU_WORD0_EXT_KCACHE_MODE2(cf->kcache[2].mode);
53 bc->bytecode[id++] =
54 S_SQ_CF_ALU_WORD1_EXT_CF_INST(
55 r600_isa_cf_opcode(bc->isa->hw_class, CF_OP_ALU_EXT)) |
56 S_SQ_CF_ALU_WORD1_EXT_KCACHE_MODE3(cf->kcache[3].mode) |
57 S_SQ_CF_ALU_WORD1_EXT_KCACHE_ADDR2(cf->kcache[2].addr) |
58 S_SQ_CF_ALU_WORD1_EXT_KCACHE_ADDR3(cf->kcache[3].addr) |
59 S_SQ_CF_ALU_WORD1_EXT_BARRIER(1);
60 }
61 bc->bytecode[id++] = S_SQ_CF_ALU_WORD0_ADDR(cf->addr >> 1) |
62 S_SQ_CF_ALU_WORD0_KCACHE_MODE0(cf->kcache[0].mode) |
63 S_SQ_CF_ALU_WORD0_KCACHE_BANK0(cf->kcache[0].bank) |
64 S_SQ_CF_ALU_WORD0_KCACHE_BANK1(cf->kcache[1].bank);
65 bc->bytecode[id++] = S_SQ_CF_ALU_WORD1_CF_INST(opcode) |
66 S_SQ_CF_ALU_WORD1_KCACHE_MODE1(cf->kcache[1].mode) |
67 S_SQ_CF_ALU_WORD1_KCACHE_ADDR0(cf->kcache[0].addr) |
68 S_SQ_CF_ALU_WORD1_KCACHE_ADDR1(cf->kcache[1].addr) |
69 S_SQ_CF_ALU_WORD1_BARRIER(1) |
70 S_SQ_CF_ALU_WORD1_COUNT((cf->ndw / 2) - 1);
71 } else if (cfop->flags & CF_CLAUSE) {
72 /* CF_TEX/VTX (CF_ALU already handled above) */
73 bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->addr >> 1);
74 bc->bytecode[id] = S_SQ_CF_WORD1_CF_INST(opcode) |
75 S_SQ_CF_WORD1_BARRIER(1) |
76 S_SQ_CF_WORD1_VALID_PIXEL_MODE(cf->vpm) |
77 S_SQ_CF_WORD1_COUNT((cf->ndw / 4) - 1);
78 if (bc->gfx_level == EVERGREEN) /* no EOP on cayman */
79 bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
80 id++;
81 } else if (cfop->flags & CF_EXP) {
82 /* EXPORT instructions */
83 bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf->output.gpr) |
84 S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
85 S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
86 S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type) |
87 S_SQ_CF_ALLOC_EXPORT_WORD0_INDEX_GPR(cf->output.index_gpr);
88 bc->bytecode[id] =
89 S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
90 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf->output.swizzle_x) |
91 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(cf->output.swizzle_y) |
92 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(cf->output.swizzle_z) |
93 S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(cf->output.swizzle_w) |
94 S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->barrier) |
95 S_SQ_CF_ALLOC_EXPORT_WORD1_MARK(cf->mark) |
96 S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode);
97
98 if (bc->gfx_level == EVERGREEN) /* no EOP on cayman */
99 bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
100 id++;
101 } else if (cfop->flags & CF_RAT) {
102 bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RAT_RAT_ID(cf->rat.id) |
103 S_SQ_CF_ALLOC_EXPORT_WORD0_RAT_RAT_INST(cf->rat.inst) |
104 S_SQ_CF_ALLOC_EXPORT_WORD0_RAT_RAT_INDEX_MODE(cf->rat.index_mode) |
105 S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf->output.gpr) |
106 S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
107 S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type) |
108 S_SQ_CF_ALLOC_EXPORT_WORD0_INDEX_GPR(cf->output.index_gpr);
109 bc->bytecode[id] = S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
110 S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->barrier) |
111 S_SQ_CF_ALLOC_EXPORT_WORD1_MARK(cf->mark) |
112 S_SQ_CF_ALLOC_EXPORT_WORD1_VALID_PIXEL_MODE(cf->vpm) |
113 S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode) |
114 S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(cf->output.comp_mask) |
115 S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(cf->output.array_size) |
116 S_SQ_CF_ALLOC_EXPORT_WORD1_MARK(cf->output.mark);
117 if (bc->gfx_level == EVERGREEN) /* no EOP on cayman */
118 bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
119 id++;
120
121 } else if (cfop->flags & CF_MEM) {
122 /* MEM_STREAM, MEM_RING instructions */
123 bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf->output.gpr) |
124 S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
125 S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
126 S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type) |
127 S_SQ_CF_ALLOC_EXPORT_WORD0_INDEX_GPR(cf->output.index_gpr);
128 bc->bytecode[id] = S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
129 S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->barrier) |
130 S_SQ_CF_ALLOC_EXPORT_WORD1_MARK(cf->mark) |
131 S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode) |
132 S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(cf->output.comp_mask) |
133 S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(cf->output.array_size);
134 if (bc->gfx_level == EVERGREEN) /* no EOP on cayman */
135 bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
136 id++;
137 } else {
138 /* other instructions */
139 bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->cf_addr >> 1);
140 bc->bytecode[id] = S_SQ_CF_WORD1_CF_INST(opcode) |
141 S_SQ_CF_WORD1_VALID_PIXEL_MODE(cf->vpm) |
142 S_SQ_CF_WORD1_BARRIER(1) |
143 S_SQ_CF_WORD1_COND(cf->cond) |
144 S_SQ_CF_WORD1_POP_COUNT(cf->pop_count) |
145 S_SQ_CF_WORD1_COUNT(cf->count);
146 if (bc->gfx_level == EVERGREEN) /* no EOP on cayman */
147 bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
148 id++;
149 }
150 }
151 return 0;
152 }
153
154 #if 0
155 void eg_bytecode_export_read(struct r600_bytecode *bc,
156 struct r600_bytecode_output *output, uint32_t word0, uint32_t word1)
157 {
158 output->array_base = G_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(word0);
159 output->type = G_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(word0);
160 output->gpr = G_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(word0);
161 output->elem_size = G_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(word0);
162
163 output->swizzle_x = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(word1);
164 output->swizzle_y = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(word1);
165 output->swizzle_z = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(word1);
166 output->swizzle_w = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(word1);
167 output->burst_count = G_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(word1);
168 output->end_of_program = G_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(word1);
169 output->op = r600_isa_cf_by_opcode(bc->isa,
170 G_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(word1), /* is_cf_alu = */ 0 );
171 output->barrier = G_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(word1);
172 output->array_size = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(word1);
173 output->comp_mask = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(word1);
174 }
175 #endif
176
eg_bytecode_gds_build(struct r600_bytecode * bc,struct r600_bytecode_gds * gds,unsigned id)177 int eg_bytecode_gds_build(struct r600_bytecode *bc, struct r600_bytecode_gds *gds, unsigned id)
178 {
179 unsigned gds_op = (r600_isa_fetch_opcode(bc->isa->hw_class, gds->op) >> 8) & 0x3f;
180 unsigned opcode;
181 if (gds->op == FETCH_OP_TF_WRITE) {
182 opcode = 5;
183 gds_op = 0;
184 } else
185 opcode = 4;
186 bc->bytecode[id++] = S_SQ_MEM_GDS_WORD0_MEM_INST(2) |
187 S_SQ_MEM_GDS_WORD0_MEM_OP(opcode) |
188 S_SQ_MEM_GDS_WORD0_SRC_GPR(gds->src_gpr) |
189 S_SQ_MEM_GDS_WORD0_SRC_REL(gds->src_rel) |
190 S_SQ_MEM_GDS_WORD0_SRC_SEL_X(gds->src_sel_x) |
191 S_SQ_MEM_GDS_WORD0_SRC_SEL_Y(gds->src_sel_y) |
192 S_SQ_MEM_GDS_WORD0_SRC_SEL_Z(gds->src_sel_z);
193
194 bc->bytecode[id++] = S_SQ_MEM_GDS_WORD1_DST_GPR(gds->dst_gpr) |
195 S_SQ_MEM_GDS_WORD1_DST_REL(gds->dst_rel) |
196 S_SQ_MEM_GDS_WORD1_GDS_OP(gds_op) |
197 S_SQ_MEM_GDS_WORD1_SRC_GPR(gds->src_gpr2) |
198 S_SQ_MEM_GDS_WORD1_UAV_INDEX_MODE(gds->uav_index_mode) |
199 S_SQ_MEM_GDS_WORD1_UAV_ID(gds->uav_id) |
200 S_SQ_MEM_GDS_WORD1_ALLOC_CONSUME(gds->alloc_consume) |
201 S_SQ_MEM_GDS_WORD1_BCAST_FIRST_REQ(gds->bcast_first_req);
202
203 bc->bytecode[id++] = S_SQ_MEM_GDS_WORD2_DST_SEL_X(gds->dst_sel_x) |
204 S_SQ_MEM_GDS_WORD2_DST_SEL_Y(gds->dst_sel_y) |
205 S_SQ_MEM_GDS_WORD2_DST_SEL_Z(gds->dst_sel_z) |
206 S_SQ_MEM_GDS_WORD2_DST_SEL_W(gds->dst_sel_w);
207 return 0;
208 }
209
eg_bytecode_alu_build(struct r600_bytecode * bc,struct r600_bytecode_alu * alu,unsigned id)210 int eg_bytecode_alu_build(struct r600_bytecode *bc, struct r600_bytecode_alu *alu, unsigned id)
211 {
212 if (alu->is_lds_idx_op) {
213 assert(!alu->src[0].abs && !alu->src[1].abs && !alu->src[2].abs);
214 assert(!alu->src[0].neg && !alu->src[1].neg && !alu->src[2].neg);
215 bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
216 S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) |
217 S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
218 S_SQ_ALU_WORD0_LDS_IDX_OP_IDX_OFFSET_4(alu->lds_idx >> 4) |
219 S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) |
220 S_SQ_ALU_WORD0_SRC1_REL(alu->src[1].rel) |
221 S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) |
222 S_SQ_ALU_WORD0_LDS_IDX_OP_IDX_OFFSET_5(alu->lds_idx >> 5) |
223 S_SQ_ALU_WORD0_INDEX_MODE(alu->index_mode) |
224 S_SQ_ALU_WORD0_PRED_SEL(alu->pred_sel) |
225 S_SQ_ALU_WORD0_LAST(alu->last);
226 } else {
227 bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
228 S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) |
229 S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
230 S_SQ_ALU_WORD0_SRC0_NEG(alu->src[0].neg) |
231 S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) |
232 S_SQ_ALU_WORD0_SRC1_REL(alu->src[1].rel) |
233 S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) |
234 S_SQ_ALU_WORD0_SRC1_NEG(alu->src[1].neg) |
235 S_SQ_ALU_WORD0_PRED_SEL(alu->pred_sel) |
236 S_SQ_ALU_WORD0_LAST(alu->last);
237 }
238
239 /* don't replace gpr by pv or ps for destination register */
240 if (alu->is_lds_idx_op) {
241 unsigned lds_op = r600_isa_alu_opcode(bc->isa->hw_class, alu->op);
242 bc->bytecode[id++] =
243 S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu->src[2].sel) |
244 S_SQ_ALU_WORD1_OP3_SRC2_REL(alu->src[2].rel) |
245 S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu->src[2].chan) |
246 S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_1(alu->lds_idx >> 1) |
247
248 S_SQ_ALU_WORD1_OP3_ALU_INST(lds_op & 0xff) |
249 S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
250 S_SQ_ALU_WORD1_LDS_IDX_OP_LDS_OP((lds_op >> 8) & 0xff) |
251 S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_0(alu->lds_idx) |
252 S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_2(alu->lds_idx >> 2) |
253 S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
254 S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_3(alu->lds_idx >> 3);
255
256 } else if (alu->is_op3) {
257 assert(!alu->src[0].abs && !alu->src[1].abs && !alu->src[2].abs);
258 bc->bytecode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) |
259 S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
260 S_SQ_ALU_WORD1_DST_REL(alu->dst.rel) |
261 S_SQ_ALU_WORD1_CLAMP(alu->dst.clamp) |
262 S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu->src[2].sel) |
263 S_SQ_ALU_WORD1_OP3_SRC2_REL(alu->src[2].rel) |
264 S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu->src[2].chan) |
265 S_SQ_ALU_WORD1_OP3_SRC2_NEG(alu->src[2].neg) |
266 S_SQ_ALU_WORD1_OP3_ALU_INST(r600_isa_alu_opcode(bc->isa->hw_class, alu->op)) |
267 S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle);
268 } else {
269 bc->bytecode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) |
270 S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
271 S_SQ_ALU_WORD1_DST_REL(alu->dst.rel) |
272 S_SQ_ALU_WORD1_CLAMP(alu->dst.clamp) |
273 S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) |
274 S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) |
275 S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) |
276 S_SQ_ALU_WORD1_OP2_OMOD(alu->omod) |
277 S_SQ_ALU_WORD1_OP2_ALU_INST(r600_isa_alu_opcode(bc->isa->hw_class, alu->op)) |
278 S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
279 S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->execute_mask) |
280 S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->update_pred);
281 }
282 return 0;
283 }
284