r600g: use tables with ISA info v3

v3: added some flags including condition codes for ALU,
    fixed issue with CF reverse lookup (overlapping ranges of CF_ALU_xxx
    and other CF instructions)
    rebased on current master

Signed-off-by: Vadim Girlin <[email protected]>

1 files changed, 78 insertions, 104 deletions

diff --git a/src/gallium/drivers/r600/eg_asm.c b/src/gallium/drivers/r600/eg_asm.c
index 0dc3ffdaa38..fffc436e823 100644
--- a/src/gallium/drivers/r600/eg_asm.c
+++ b/src/gallium/drivers/r600/eg_asm.c
@@ -32,121 +32,94 @@ int eg_bytecode_cf_build(struct r600_bytecode *bc, struct r600_bytecode_cf *cf)
 {
 	unsigned id = cf->id;
 
-	switch (cf->inst) {
-	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);
-		}
+	if (cf->op == CF_NATIVE) {
+		bc->bytecode[id++] = cf->isa[0];
+		bc->bytecode[id++] = cf->isa[1];
+	} else {
+		const struct cf_op_info *cfop = r600_isa_cf(cf->op);
+		unsigned opcode = r600_isa_cf_opcode(bc->isa->hw_class, cf->op);
+		if (cfop->flags & CF_ALU) { /* ALU clauses */
 
-		bc->bytecode[id++] = S_SQ_CF_ALU_WORD0_ADDR(cf->addr >> 1) |
-			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) |
+			/* 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++] =
+						S_SQ_CF_ALU_WORD1_EXT_CF_INST(
+							r600_isa_cf_opcode(bc->isa->hw_class, CF_OP_ALU_EXT)) |
+						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->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++] = S_SQ_CF_ALU_WORD1_CF_INST(opcode) |
+					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++] = cf->inst |
+		} else if (cfop->flags & CF_CLAUSE) {
+			/* CF_TEX/VTX (CF_ALU already handled above) */
+			bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->addr >> 1);
+			bc->bytecode[id++] = S_SQ_CF_WORD1_CF_INST(opcode) |
 					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:
-		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_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->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_POP:
-	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL:
-	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_DX10:
-	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END:
-	case EG_V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE:
-	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++] = cf->inst |
+		} else if (cfop->flags & CF_EXP) {
+			/* EXPORT instructions */
+			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_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->output.barrier) |
+					S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode);
+
+			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++;
+		} else if (cfop->flags & CF_STRM) {
+			/* MEM_STREAM instructions */
+			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) |
+					S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(opcode) |
+					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++;
+		} else {
+			/* branch, loop, call, return instructions */
+			bc->bytecode[id++] = S_SQ_CF_WORD0_ADDR(cf->cf_addr >> 1);
+			bc->bytecode[id++] =  S_SQ_CF_WORD1_CF_INST(opcode)|
 					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 -EINVAL;
+		}
 	}
 	return 0;
 }
 
-void eg_bytecode_export_read(struct r600_bytecode_output *output, uint32_t word0, uint32_t word1)
+void eg_bytecode_export_read(struct r600_bytecode *bc,
+		struct r600_bytecode_output *output, uint32_t word0, uint32_t word1)
 {
 	output->array_base = G_SQ_CF_ALLOC_EXPORT_WORD0_ARRAY_BASE(word0);
 	output->type = G_SQ_CF_ALLOC_EXPORT_WORD0_TYPE(word0);
@@ -159,7 +132,8 @@ void eg_bytecode_export_read(struct r600_bytecode_output *output, uint32_t word0
 	output->swizzle_w = G_SQ_CF_ALLOC_EXPORT_WORD1_SWIZ_SEL_W(word1);
 	output->burst_count = G_SQ_CF_ALLOC_EXPORT_WORD1_BURST_COUNT(word1);
 	output->end_of_program = G_SQ_CF_ALLOC_EXPORT_WORD1_END_OF_PROGRAM(word1);
-	output->inst = EG_S_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(G_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(word1));
+	output->op = r600_isa_cf_by_opcode(bc->isa,
+			G_SQ_CF_ALLOC_EXPORT_WORD1_CF_INST(word1), /* is_cf_alu = */ 0 );
 	output->barrier = G_SQ_CF_ALLOC_EXPORT_WORD1_BARRIER(word1);
 	output->array_size = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_ARRAY_SIZE(word1);
 	output->comp_mask = G_SQ_CF_ALLOC_EXPORT_WORD1_BUF_COMP_MASK(word1);