if (inst->src[i].file == BAD_FILE) {
/* Do nothing but otherwise increment as normal */
+ } else if (dst.file == MRF &&
+ dst.width == 8 &&
+ brw->has_compr4 &&
+ i + 4 < inst->sources &&
+ inst->src[i + 4].equals(horiz_offset(inst->src[i], 8))) {
+ fs_reg compr4_dst = dst;
+ compr4_dst.reg += BRW_MRF_COMPR4;
+ compr4_dst.width = 16;
+ fs_reg compr4_src = inst->src[i];
+ compr4_src.width = 16;
+ fs_inst *mov = MOV(compr4_dst, compr4_src);
+ mov->force_writemask_all = true;
+ inst->insert_before(block, mov);
+ /* Mark i+4 as BAD_FILE so we don't emit a MOV for it */
+ inst->src[i + 4].file = BAD_FILE;
} else {
fs_inst *mov = MOV(dst, inst->src[i]);
if (inst->src[i].file == GRF) {
return reg;
}
+static inline fs_reg
+horiz_offset(fs_reg reg, unsigned delta)
+{
+ switch (reg.file) {
+ case BAD_FILE:
+ case UNIFORM:
+ case IMM:
+ /* These only have a single component that is implicitly splatted. A
+ * horizontal offset should be a harmless no-op.
+ */
+ break;
+ case GRF:
+ case MRF:
+ return byte_offset(reg, delta * reg.stride * type_sz(reg.type));
+ default:
+ assert(delta == 0);
+ }
+ return reg;
+}
+
static inline fs_reg
offset(fs_reg reg, unsigned delta)
{
assert(idx == 0 || (reg.file != HW_REG && reg.file != IMM));
assert(reg.width == 16);
reg.width = 8;
- return byte_offset(reg, 8 * idx * reg.stride * type_sz(reg.type));
+ return horiz_offset(reg, 8 * idx);
}
static const fs_reg reg_undef;