return rzalloc_size(shader, sz); /* TODO: don't use rzalloc */
}
-struct ir3 * ir3_create(struct ir3_compiler *compiler,
- unsigned nin, unsigned nout)
+struct ir3 * ir3_create(struct ir3_compiler *compiler, gl_shader_stage type)
{
- struct ir3 *shader = rzalloc(compiler, struct ir3);
+ struct ir3 *shader = rzalloc(NULL, struct ir3);
shader->compiler = compiler;
- shader->ninputs = nin;
- shader->inputs = ir3_alloc(shader, sizeof(shader->inputs[0]) * nin);
-
- shader->noutputs = nout;
- shader->outputs = ir3_alloc(shader, sizeof(shader->outputs[0]) * nout);
+ shader->type = type;
list_inithead(&shader->block_list);
list_inithead(&shader->array_list);
if (reg->flags & IR3_REG_RELATIV) {
components = reg->size;
val.idummy10 = reg->array.offset;
- max = (reg->array.offset + repeat + components - 1) >> 2;
+ max = (reg->array.offset + repeat + components - 1);
} else {
components = util_last_bit(reg->wrmask);
val.comp = reg->num & 0x3;
val.num = reg->num >> 2;
- max = (reg->num + repeat + components - 1) >> 2;
+ max = (reg->num + repeat + components - 1);
}
if (reg->flags & IR3_REG_CONST) {
- info->max_const = MAX2(info->max_const, max);
+ info->max_const = MAX2(info->max_const, max >> 2);
} else if (val.num == 63) {
/* ignore writes to dummy register r63.x */
- } else if (max < 48) {
+ } else if (max < regid(48, 0)) {
if (reg->flags & IR3_REG_HALF) {
if (info->gpu_id >= 600) {
/* starting w/ a6xx, half regs conflict with full regs: */
- info->max_reg = MAX2(info->max_reg, (max+1)/2);
+ info->max_reg = MAX2(info->max_reg, max >> 3);
} else {
- info->max_half_reg = MAX2(info->max_half_reg, max);
+ info->max_half_reg = MAX2(info->max_half_reg, max >> 2);
}
} else {
- info->max_reg = MAX2(info->max_reg, max);
+ info->max_reg = MAX2(info->max_reg, max >> 2);
}
}
}
cat0->sync = !!(instr->flags & IR3_INSTR_SY);
cat0->opc_cat = 0;
+ switch (instr->opc) {
+ case OPC_IF:
+ case OPC_ELSE:
+ case OPC_ENDIF:
+ cat0->dummy4 = 16;
+ break;
+ default:
+ break;
+ }
+
return 0;
}
} else if (src1->flags & IR3_REG_CONST) {
iassert(src1->num < (1 << 12));
cat2->c1.src1 = reg(src1, info, instr->repeat,
- IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF);
+ IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF |
+ absneg);
cat2->c1.src1_c = 1;
} else {
iassert(src1->num < (1 << 11));
} else if (src2->flags & IR3_REG_CONST) {
iassert(src2->num < (1 << 12));
cat2->c2.src2 = reg(src2, info, instr->repeat,
- IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF);
+ IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF |
+ absneg);
cat2->c2.src2_c = 1;
} else {
iassert(src2->num < (1 << 11));
} else if (src1->flags & IR3_REG_CONST) {
iassert(src1->num < (1 << 12));
cat3->c1.src1 = reg(src1, info, instr->repeat,
- IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF);
+ IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF | absneg);
cat3->c1.src1_c = 1;
} else {
iassert(src1->num < (1 << 11));
} else if (src3->flags & IR3_REG_CONST) {
iassert(src3->num < (1 << 12));
cat3->c2.src3 = reg(src3, info, instr->repeat,
- IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF);
+ IR3_REG_CONST | IR3_REG_R | IR3_REG_HALF | absneg);
cat3->c2.src3_c = 1;
} else {
iassert(src3->num < (1 << 11));
* than tex/sampler idx, we use the first src reg in the ir to hold
* samp_tex hvec2:
*/
- struct ir3_register *src1 = instr->regs[2];
- struct ir3_register *src2 = instr->regs[3];
+ struct ir3_register *src1;
+ struct ir3_register *src2;
instr_cat5_t *cat5 = ptr;
- iassert_type(dst, type_size(instr->cat5.type) == 32)
+ iassert((instr->regs_count == 2) ||
+ (instr->regs_count == 3) || (instr->regs_count == 4));
+
+ switch (instr->opc) {
+ case OPC_DSX:
+ case OPC_DSXPP_1:
+ case OPC_DSY:
+ case OPC_DSYPP_1:
+ case OPC_RGETPOS:
+ case OPC_RGETINFO:
+ iassert((instr->flags & IR3_INSTR_S2EN) == 0);
+ src1 = instr->regs[1];
+ src2 = instr->regs_count > 2 ? instr->regs[2] : NULL;
+ break;
+ default:
+ src1 = instr->regs[2];
+ src2 = instr->regs_count > 3 ? instr->regs[3] : NULL;
+ break;
+ }
assume(src1 || !src2);
return 0;
} else if (instr->cat6.src_offset || (instr->opc == OPC_LDG) ||
- (instr->opc == OPC_LDL)) {
+ (instr->opc == OPC_LDL) || (instr->opc == OPC_LDLW)) {
+ struct ir3_register *src3 = instr->regs[3];
instr_cat6a_t *cat6a = ptr;
cat6->src_off = true;
- cat6a->src1 = reg(src1, info, instr->repeat, IR3_REG_IMMED);
- cat6a->src1_im = !!(src1->flags & IR3_REG_IMMED);
- if (src2) {
- cat6a->src2 = reg(src2, info, instr->repeat, IR3_REG_IMMED);
- cat6a->src2_im = !!(src2->flags & IR3_REG_IMMED);
+ if (instr->opc == OPC_LDG) {
+ /* For LDG src1 can not be immediate, so src1_imm is redundant and
+ * instead used to signal whether (when true) 'off' is a 32 bit
+ * register or an immediate offset.
+ */
+ cat6a->src1 = reg(src1, info, instr->repeat, 0);
+ cat6a->src1_im = !(src3->flags & IR3_REG_IMMED);
+ cat6a->off = reg(src3, info, instr->repeat, IR3_REG_IMMED);
+ } else {
+ cat6a->src1 = reg(src1, info, instr->repeat, IR3_REG_IMMED);
+ cat6a->src1_im = !!(src1->flags & IR3_REG_IMMED);
+ cat6a->off = reg(src3, info, instr->repeat, IR3_REG_IMMED);
+ iassert(src3->flags & IR3_REG_IMMED);
}
- cat6a->off = instr->cat6.src_offset;
+
+ /* Num components */
+ cat6a->src2 = reg(src2, info, instr->repeat, IR3_REG_IMMED);
+ cat6a->src2_im = true;
} else {
instr_cat6b_t *cat6b = ptr;
}
if (instr->cat6.dst_offset || (instr->opc == OPC_STG) ||
- (instr->opc == OPC_STL)) {
+ (instr->opc == OPC_STL) || (instr->opc == OPC_STLW)) {
instr_cat6c_t *cat6c = ptr;
cat6->dst_off = true;
cat6c->dst = reg(dst, info, instr->repeat, IR3_REG_R | IR3_REG_HALF);
- cat6c->off = instr->cat6.dst_offset;
+
+ if (instr->flags & IR3_INSTR_G) {
+ struct ir3_register *src3 = instr->regs[4];
+ cat6c->off = reg(src3, info, instr->repeat, IR3_REG_R | IR3_REG_HALF);
+ if (src3->flags & IR3_REG_IMMED) {
+ /* Immediate offsets are in bytes... */
+ cat6->g = false;
+ cat6c->off *= 4;
+ }
+ } else {
+ cat6c->off = instr->cat6.dst_offset;
+ }
} else {
instr_cat6d_t *cat6d = ptr;
cat6->dst_off = false;
info->sizedwords = 0;
info->ss = info->sy = 0;
- list_for_each_entry (struct ir3_block, block, &shader->block_list, node) {
- list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
+ foreach_block (block, &shader->block_list) {
+ foreach_instr (instr, &block->instr_list) {
info->sizedwords += 2;
}
}
ptr = dwords = calloc(4, info->sizedwords);
- list_for_each_entry (struct ir3_block, block, &shader->block_list, node) {
- list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
+ foreach_block (block, &shader->block_list) {
+ foreach_instr (instr, &block->instr_list) {
int ret = emit[opc_cat(instr->opc)](instr, dwords, info);
if (ret)
goto fail;
+
+ if ((instr->opc == OPC_BARY_F) && (instr->regs[0]->flags & IR3_REG_EI))
+ info->last_baryf = info->instrs_count;
+
info->instrs_count += 1 + instr->repeat + instr->nop;
+ info->nops_count += instr->nop;
+ if (instr->opc == OPC_NOP)
+ info->nops_count += 1 + instr->repeat;
dwords += 2;
if (instr->flags & IR3_INSTR_SS)
{
if (instr->address != addr) {
struct ir3 *ir = instr->block->shader;
+
+ debug_assert(!instr->address);
+ debug_assert(instr->block == addr->block);
+
instr->address = addr;
array_insert(ir, ir->indirects, instr);
}
void
ir3_block_clear_mark(struct ir3_block *block)
{
- list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node)
+ foreach_instr (instr, &block->instr_list)
instr->flags &= ~IR3_INSTR_MARK;
}
void
ir3_clear_mark(struct ir3 *ir)
{
- list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
+ foreach_block (block, &ir->block_list) {
ir3_block_clear_mark(block);
}
}
ir3_count_instructions(struct ir3 *ir)
{
unsigned cnt = 0;
- list_for_each_entry (struct ir3_block, block, &ir->block_list, node) {
- list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
+ foreach_block (block, &ir->block_list) {
+ block->start_ip = cnt;
+ block->end_ip = cnt;
+ foreach_instr (instr, &block->instr_list) {
instr->ip = cnt++;
+ block->end_ip = instr->ip;
}
- block->start_ip = list_first_entry(&block->instr_list, struct ir3_instruction, node)->ip;
- block->end_ip = list_last_entry(&block->instr_list, struct ir3_instruction, node)->ip;
}
return cnt;
}
struct ir3_array *
ir3_lookup_array(struct ir3 *ir, unsigned id)
{
- list_for_each_entry (struct ir3_array, arr, &ir->array_list, node)
+ foreach_array (arr, &ir->array_list)
if (arr->id == id)
return arr;
return NULL;
projects / mesa.git / blobdiff