Home | History | Annotate | Download | only in r600
      1 /*
      2  * Copyright 2010 Jerome Glisse <glisse (at) 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_pipe.h"
     24 #include "r600_opcodes.h"
     25 #include "r600_shader.h"
     26 
     27 #include "util/u_memory.h"
     28 #include "eg_sq.h"
     29 #include <errno.h>
     30 
     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_COUNT((cf->ndw / 4) - 1);
     77 		} else if (cfop->flags & CF_EXP) {
     78 			/* EXPORT instructions */
     79 			bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf->output.gpr) |
     80 					S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
     81 					S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
     82 					S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type) |
     83 					S_SQ_CF_ALLOC_EXPORT_WORD0_INDEX_GPR(cf->output.index_gpr);
     84 			bc->bytecode[id] =
     85 					S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
     86 					S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf->output.swizzle_x) |
     87 					S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(cf->output.swizzle_y) |
     88 					S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(cf->output.swizzle_z) |
     89 					S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(cf->output.swizzle_w) |
     90 					S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->barrier) |
     91 					S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode);
     92 
     93 			if (bc->chip_class == EVERGREEN) /* no EOP on cayman */
     94 				bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
     95 			id++;
     96 		} else if (cfop->flags & CF_MEM) {
     97 			/* MEM_STREAM, MEM_RING instructions */
     98 			bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf->output.gpr) |
     99 					S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
    100 					S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
    101 					S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type) |
    102 					S_SQ_CF_ALLOC_EXPORT_WORD0_INDEX_GPR(cf->output.index_gpr);
    103 			bc->bytecode[id] = S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
    104 					S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->barrier) |
    105 					S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode) |
    106 					S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(cf->output.comp_mask) |
    107 					S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(cf->output.array_size);
    108 			if (bc->chip_class == EVERGREEN) /* no EOP on cayman */
    109 				bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->end_of_program);
    110 			id++;
    111 		} else {
    112 			/* other instructions */
    113 			bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->cf_addr >> 1);
    114 			bc->bytecode[id++] =  S_SQ_CF_WORD1_CF_INST(opcode)|
    115 					S_SQ_CF_WORD1_BARRIER(1) |
    116 					S_SQ_CF_WORD1_COND(cf->cond) |
    117 					S_SQ_CF_WORD1_POP_COUNT(cf->pop_count) |
    118 					S_SQ_CF_WORD1_COUNT(cf->count) |
    119 					S_SQ_CF_WORD1_END_OF_PROGRAM(cf->end_of_program);
    120 		}
    121 	}
    122 	return 0;
    123 }
    124 
    125 #if 0
    126 void eg_bytecode_export_read(struct r600_bytecode *bc,
    127 		struct r600_bytecode_output *output, uint32_t word0, uint32_t word1)
    128 {
    129 	output->array_base = G_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(word0);
    130 	output->type = G_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(word0);
    131 	output->gpr = G_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(word0);
    132 	output->elem_size = G_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(word0);
    133 
    134 	output->swizzle_x = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(word1);
    135 	output->swizzle_y = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(word1);
    136 	output->swizzle_z = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(word1);
    137 	output->swizzle_w = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(word1);
    138 	output->burst_count = G_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(word1);
    139 	output->end_of_program = G_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(word1);
    140 	output->op = r600_isa_cf_by_opcode(bc->isa,
    141 			G_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(word1), /* is_cf_alu = */ 0 );
    142 	output->barrier = G_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(word1);
    143 	output->array_size = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(word1);
    144 	output->comp_mask = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(word1);
    145 }
    146 #endif
    147 
    148 int egcm_load_index_reg(struct r600_bytecode *bc, unsigned id, bool inside_alu_clause)
    149 {
    150 	struct r600_bytecode_alu alu;
    151 	int r;
    152 	unsigned type;
    153 
    154 	assert(id < 2);
    155 	assert(bc->chip_class >= EVERGREEN);
    156 
    157 	if (bc->index_loaded[id])
    158 		return 0;
    159 
    160 	memset(&alu, 0, sizeof(alu));
    161 	alu.op = ALU_OP1_MOVA_INT;
    162 	alu.src[0].sel = bc->index_reg[id];
    163 	alu.src[0].chan = 0;
    164 	if (bc->chip_class == CAYMAN)
    165 		alu.dst.sel = id == 0 ? CM_V_SQ_MOVA_DST_CF_IDX0 : CM_V_SQ_MOVA_DST_CF_IDX1;
    166 
    167 	alu.last = 1;
    168 	r = r600_bytecode_add_alu(bc, &alu);
    169 	if (r)
    170 		return r;
    171 
    172 	bc->ar_loaded = 0; /* clobbered */
    173 
    174 	if (bc->chip_class == EVERGREEN) {
    175 		memset(&alu, 0, sizeof(alu));
    176 		alu.op = id == 0 ? ALU_OP0_SET_CF_IDX0 : ALU_OP0_SET_CF_IDX1;
    177 		alu.last = 1;
    178 		r = r600_bytecode_add_alu(bc, &alu);
    179 		if (r)
    180 			return r;
    181 	}
    182 
    183 	/* Must split ALU group as index only applies to following group */
    184 	if (inside_alu_clause) {
    185 		type = bc->cf_last->op;
    186 		if ((r = r600_bytecode_add_cf(bc))) {
    187 			return r;
    188 		}
    189 		bc->cf_last->op = type;
    190 	}
    191 
    192 	bc->index_loaded[id] = 1;
    193 
    194 	return 0;
    195 }
    196 
    197 int eg_bytecode_gds_build(struct r600_bytecode *bc, struct r600_bytecode_gds *gds, unsigned id)
    198 {
    199 	unsigned opcode = r600_isa_fetch_opcode(bc->isa->hw_class, gds->op) >> 8;
    200 	bc->bytecode[id++] = S_SQ_MEM_GDS_WORD0_MEM_INST(2) |
    201 		S_SQ_MEM_GDS_WORD0_MEM_OP(opcode) |
    202 		S_SQ_MEM_GDS_WORD0_SRC_GPR(gds->src_gpr) |
    203 		S_SQ_MEM_GDS_WORD0_SRC_REL(gds->src_rel) |
    204 		S_SQ_MEM_GDS_WORD0_SRC_SEL_X(gds->src_sel_x) |
    205 		S_SQ_MEM_GDS_WORD0_SRC_SEL_Y(gds->src_sel_y) |
    206 		S_SQ_MEM_GDS_WORD0_SRC_SEL_Z(gds->src_sel_z);
    207 
    208 	bc->bytecode[id++] = S_SQ_MEM_GDS_WORD1_DST_GPR(gds->dst_gpr) |
    209 		S_SQ_MEM_GDS_WORD1_DST_REL(gds->dst_rel) |
    210 		S_SQ_MEM_GDS_WORD1_GDS_OP(gds->gds_op) |
    211 		S_SQ_MEM_GDS_WORD1_SRC_GPR(gds->src_gpr2);
    212 
    213 	bc->bytecode[id++] = S_SQ_MEM_GDS_WORD2_DST_SEL_X(gds->dst_sel_x) |
    214 		S_SQ_MEM_GDS_WORD2_DST_SEL_Y(gds->dst_sel_y) |
    215 		S_SQ_MEM_GDS_WORD2_DST_SEL_Z(gds->dst_sel_z) |
    216 		S_SQ_MEM_GDS_WORD2_DST_SEL_W(gds->dst_sel_w);
    217 	return 0;
    218 }
    219 
    220 int eg_bytecode_alu_build(struct r600_bytecode *bc, struct r600_bytecode_alu *alu, unsigned id)
    221 {
    222 	if (alu->is_lds_idx_op) {
    223 		assert(!alu->src[0].abs && !alu->src[1].abs && !alu->src[2].abs);
    224 		assert(!alu->src[0].neg && !alu->src[1].neg && !alu->src[2].neg);
    225 		bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
    226 			S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) |
    227 			S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
    228 			S_SQ_ALU_WORD0_LDS_IDX_OP_IDX_OFFSET_4(alu->lds_idx >> 4) |
    229 			S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) |
    230 			S_SQ_ALU_WORD0_SRC1_REL(alu->src[1].rel) |
    231 			S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) |
    232 			S_SQ_ALU_WORD0_LDS_IDX_OP_IDX_OFFSET_5(alu->lds_idx >> 5) |
    233 			S_SQ_ALU_WORD0_INDEX_MODE(alu->index_mode) |
    234 			S_SQ_ALU_WORD0_PRED_SEL(alu->pred_sel) |
    235 			S_SQ_ALU_WORD0_LAST(alu->last);
    236 	} else {
    237 		bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
    238 			S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) |
    239 			S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
    240 			S_SQ_ALU_WORD0_SRC0_NEG(alu->src[0].neg) |
    241 			S_SQ_ALU_WORD0_SRC1_SEL(alu->src[1].sel) |
    242 			S_SQ_ALU_WORD0_SRC1_REL(alu->src[1].rel) |
    243 			S_SQ_ALU_WORD0_SRC1_CHAN(alu->src[1].chan) |
    244 			S_SQ_ALU_WORD0_SRC1_NEG(alu->src[1].neg) |
    245 			S_SQ_ALU_WORD0_PRED_SEL(alu->pred_sel) |
    246 			S_SQ_ALU_WORD0_LAST(alu->last);
    247 	}
    248 
    249 	/* don't replace gpr by pv or ps for destination register */
    250 	if (alu->is_lds_idx_op) {
    251 		unsigned lds_op = r600_isa_alu_opcode(bc->isa->hw_class, alu->op);
    252 		bc->bytecode[id++] =
    253 			S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu->src[2].sel) |
    254 			S_SQ_ALU_WORD1_OP3_SRC2_REL(alu->src[2].rel) |
    255 			S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu->src[2].chan) |
    256 			S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_1(alu->lds_idx >> 1) |
    257 
    258 			S_SQ_ALU_WORD1_OP3_ALU_INST(lds_op & 0xff) |
    259 			S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
    260 			S_SQ_ALU_WORD1_LDS_IDX_OP_LDS_OP((lds_op >> 8) & 0xff) |
    261 			S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_0(alu->lds_idx) |
    262 			S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_2(alu->lds_idx >> 2) |
    263 			S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
    264 			S_SQ_ALU_WORD1_LDS_IDX_OP_IDX_OFFSET_3(alu->lds_idx >> 3);
    265 
    266 	} else if (alu->is_op3) {
    267 		assert(!alu->src[0].abs && !alu->src[1].abs && !alu->src[2].abs);
    268 		bc->bytecode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) |
    269 					S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
    270 			                S_SQ_ALU_WORD1_DST_REL(alu->dst.rel) |
    271 			                S_SQ_ALU_WORD1_CLAMP(alu->dst.clamp) |
    272 					S_SQ_ALU_WORD1_OP3_SRC2_SEL(alu->src[2].sel) |
    273 					S_SQ_ALU_WORD1_OP3_SRC2_REL(alu->src[2].rel) |
    274 					S_SQ_ALU_WORD1_OP3_SRC2_CHAN(alu->src[2].chan) |
    275 					S_SQ_ALU_WORD1_OP3_SRC2_NEG(alu->src[2].neg) |
    276 					S_SQ_ALU_WORD1_OP3_ALU_INST(r600_isa_alu_opcode(bc->isa->hw_class, alu->op)) |
    277 					S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle);
    278 	} else {
    279 		bc->bytecode[id++] = S_SQ_ALU_WORD1_DST_GPR(alu->dst.sel) |
    280 					S_SQ_ALU_WORD1_DST_CHAN(alu->dst.chan) |
    281 			                S_SQ_ALU_WORD1_DST_REL(alu->dst.rel) |
    282 			                S_SQ_ALU_WORD1_CLAMP(alu->dst.clamp) |
    283 					S_SQ_ALU_WORD1_OP2_SRC0_ABS(alu->src[0].abs) |
    284 					S_SQ_ALU_WORD1_OP2_SRC1_ABS(alu->src[1].abs) |
    285 					S_SQ_ALU_WORD1_OP2_WRITE_MASK(alu->dst.write) |
    286 					S_SQ_ALU_WORD1_OP2_OMOD(alu->omod) |
    287 					S_SQ_ALU_WORD1_OP2_ALU_INST(r600_isa_alu_opcode(bc->isa->hw_class, alu->op)) |
    288 					S_SQ_ALU_WORD1_BANK_SWIZZLE(alu->bank_swizzle) |
    289 			                S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->execute_mask) |
    290 			                S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->update_pred);
    291 	}
    292 	return 0;
    293 }
    294