* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
-#include <stdio.h>
-#include <errno.h>
-#include "util/u_memory.h"
#include "r600_pipe.h"
-#include "r600_asm.h"
-#include "eg_sq.h"
#include "r600_opcodes.h"
-#include "evergreend.h"
-int eg_bc_cf_build(struct r600_bc *bc, struct r600_bc_cf *cf)
+#include "util/u_memory.h"
+#include "eg_sq.h"
+#include <errno.h>
+
+int eg_bytecode_cf_build(struct r600_bytecode *bc, struct r600_bytecode_cf *cf)
{
unsigned id = cf->id;
- unsigned end_of_program = bc->cf.prev == &cf->list;
switch (cf->inst) {
- case (EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU << 3):
- case (EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
- assert(!end_of_program);
+ case EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU:
+ case EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER:
+ case EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER:
+ case EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE:
+ /* prepend ALU_EXTENDED if we need more than 2 kcache sets */
+ if (cf->eg_alu_extended) {
+ bc->bytecode[id++] =
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE0(V_SQ_CF_INDEX_NONE) |
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE1(V_SQ_CF_INDEX_NONE) |
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE2(V_SQ_CF_INDEX_NONE) |
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK_INDEX_MODE3(V_SQ_CF_INDEX_NONE) |
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK2(cf->kcache[2].bank) |
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_BANK3(cf->kcache[3].bank) |
+ S_SQ_CF_ALU_WORD0_EXT_KCACHE_MODE2(cf->kcache[2].mode);
+ bc->bytecode[id++] = EG_V_SQ_CF_ALU_WORD1_SQ_CF_INST_EXTENDED |
+ S_SQ_CF_ALU_WORD1_EXT_KCACHE_MODE3(cf->kcache[3].mode) |
+ S_SQ_CF_ALU_WORD1_EXT_KCACHE_ADDR2(cf->kcache[2].addr) |
+ S_SQ_CF_ALU_WORD1_EXT_KCACHE_ADDR3(cf->kcache[3].addr) |
+ S_SQ_CF_ALU_WORD1_EXT_BARRIER(1);
+ }
+
bc->bytecode[id++] = S_SQ_CF_ALU_WORD0_ADDR(cf->addr >> 1) |
- S_SQ_CF_ALU_WORD0_KCACHE_MODE0(cf->kcache0_mode) |
- S_SQ_CF_ALU_WORD0_KCACHE_BANK0(cf->kcache0_bank) |
- S_SQ_CF_ALU_WORD0_KCACHE_BANK1(cf->kcache1_bank);
- bc->bytecode[id++] = S_SQ_CF_ALU_WORD1_CF_INST(cf->inst >> 3) |
- S_SQ_CF_ALU_WORD1_KCACHE_MODE1(cf->kcache1_mode) |
- S_SQ_CF_ALU_WORD1_KCACHE_ADDR0(cf->kcache0_addr) |
- S_SQ_CF_ALU_WORD1_KCACHE_ADDR1(cf->kcache1_addr) |
- S_SQ_CF_ALU_WORD1_BARRIER(cf->barrier) |
+ S_SQ_CF_ALU_WORD0_KCACHE_MODE0(cf->kcache[0].mode) |
+ S_SQ_CF_ALU_WORD0_KCACHE_BANK0(cf->kcache[0].bank) |
+ S_SQ_CF_ALU_WORD0_KCACHE_BANK1(cf->kcache[1].bank);
+ bc->bytecode[id++] = cf->inst |
+ S_SQ_CF_ALU_WORD1_KCACHE_MODE1(cf->kcache[1].mode) |
+ S_SQ_CF_ALU_WORD1_KCACHE_ADDR0(cf->kcache[0].addr) |
+ S_SQ_CF_ALU_WORD1_KCACHE_ADDR1(cf->kcache[1].addr) |
+ S_SQ_CF_ALU_WORD1_BARRIER(1) |
S_SQ_CF_ALU_WORD1_COUNT((cf->ndw / 2) - 1);
break;
case EG_V_SQ_CF_WORD1_SQ_CF_INST_TEX:
case EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX:
bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->addr >> 1);
- bc->bytecode[id++] = S_SQ_CF_WORD1_CF_INST(cf->inst) |
- S_SQ_CF_WORD1_BARRIER(cf->barrier) |
- S_SQ_CF_WORD1_COUNT((cf->ndw / 4) - 1) |
- S_SQ_CF_WORD1_END_OF_PROGRAM(end_of_program);
+ bc->bytecode[id++] = cf->inst |
+ S_SQ_CF_WORD1_BARRIER(1) |
+ S_SQ_CF_WORD1_COUNT((cf->ndw / 4) - 1);
break;
case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT:
case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE:
S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type);
- bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf->output.swizzle_x) |
+ bc->bytecode[id] = S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
+ S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_X(cf->output.swizzle_x) |
S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Y(cf->output.swizzle_y) |
S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_Z(cf->output.swizzle_z) |
S_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(cf->output.swizzle_w) |
- S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->barrier) |
- S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(cf->inst) |
- S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(end_of_program);
+ S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->output.barrier) |
+ cf->output.inst;
+ if (bc->chip_class == EVERGREEN) /* no EOP on cayman */
+ bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->output.end_of_program);
+ id++;
+ break;
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF0:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF1:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF2:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM0_BUF3:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM1_BUF0:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM1_BUF1:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM1_BUF2:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM1_BUF3:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM2_BUF0:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM2_BUF1:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM2_BUF2:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM2_BUF3:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM3_BUF0:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM3_BUF1:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM3_BUF2:
+ case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_MEM_STREAM3_BUF3:
+ bc->bytecode[id++] = S_SQ_CF_ALLOC_EXPORT_WORD0_RW_GPR(cf->output.gpr) |
+ S_SQ_CF_ALLOC_EXPORT_WORD0_ELEM_SIZE(cf->output.elem_size) |
+ S_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(cf->output.array_base) |
+ S_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(cf->output.type);
+ bc->bytecode[id] = S_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(cf->output.burst_count - 1) |
+ S_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(cf->output.barrier) |
+ cf->output.inst |
+ S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(cf->output.comp_mask) |
+ S_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(cf->output.array_size);
+ if (bc->chip_class == EVERGREEN) /* no EOP on cayman */
+ bc->bytecode[id] |= S_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(cf->output.end_of_program);
+ id++;
break;
case EG_V_SQ_CF_WORD1_SQ_CF_INST_JUMP:
case EG_V_SQ_CF_WORD1_SQ_CF_INST_ELSE:
case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK:
case EG_V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS:
case EG_V_SQ_CF_WORD1_SQ_CF_INST_RETURN:
+ case CM_V_SQ_CF_WORD1_SQ_CF_INST_END:
bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->cf_addr >> 1);
- bc->bytecode[id++] = S_SQ_CF_WORD1_CF_INST(cf->inst) |
- S_SQ_CF_WORD1_BARRIER(cf->barrier) |
- S_SQ_CF_WORD1_COND(cf->cond) |
- S_SQ_CF_WORD1_POP_COUNT(cf->pop_count) |
- S_SQ_CF_WORD1_END_OF_PROGRAM(end_of_program);
-
+ bc->bytecode[id++] = cf->inst |
+ S_SQ_CF_WORD1_BARRIER(1) |
+ S_SQ_CF_WORD1_COND(cf->cond) |
+ S_SQ_CF_WORD1_POP_COUNT(cf->pop_count);
+ break;
+ case CF_NATIVE:
+ bc->bytecode[id++] = cf->isa[0];
+ bc->bytecode[id++] = cf->isa[1];
break;
default:
R600_ERR("unsupported CF instruction (0x%X)\n", cf->inst);
}
return 0;
}
-
-void eg_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count)
-{
- struct r600_pipe_state *rstate;
- unsigned i = 0;
-
- if (count > 8) {
- bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
- bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
- S_SQ_CF_WORD1_BARRIER(1) |
- S_SQ_CF_WORD1_COUNT(8 - 1);
- bytecode[i++] = S_SQ_CF_WORD0_ADDR(40 >> 1);
- bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
- S_SQ_CF_WORD1_BARRIER(1) |
- S_SQ_CF_WORD1_COUNT(count - 8 - 1);
- } else {
- bytecode[i++] = S_SQ_CF_WORD0_ADDR(8 >> 1);
- bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_VTX) |
- S_SQ_CF_WORD1_BARRIER(1) |
- S_SQ_CF_WORD1_COUNT(count - 1);
- }
- bytecode[i++] = S_SQ_CF_WORD0_ADDR(0);
- bytecode[i++] = S_SQ_CF_WORD1_CF_INST(EG_V_SQ_CF_WORD1_SQ_CF_INST_RETURN) |
- S_SQ_CF_WORD1_BARRIER(1);
-
- rstate = &ve->rstate;
- rstate->id = R600_PIPE_STATE_FETCH_SHADER;
- rstate->nregs = 0;
- r600_pipe_state_add_reg(rstate, R_0288A8_SQ_PGM_RESOURCES_FS,
- 0x00000000, 0xFFFFFFFF, NULL);
- r600_pipe_state_add_reg(rstate, R_0288A4_SQ_PGM_START_FS,
- (r600_bo_offset(ve->fetch_shader)) >> 8,
- 0xFFFFFFFF, ve->fetch_shader);
-}