diff options
| author | Jason Ekstrand <jason.ekstrand@intel.com> | 2017-10-17 18:59:26 -0700 |
|---|---|---|
| committer | Jason Ekstrand <jason.ekstrand@intel.com> | 2017-11-07 10:37:52 -0800 |
| commit | 18fde36ced4279f2577097a1a7d31b55f2f5f141 (patch) | |
| tree | 5b53c9e5ad1e2b25b640e08b5785bb82274910d2 | |
| parent | d54f8ec744545673fd78f15ffce3cb4e47d4b5f1 (diff) | |
intel/fs: Use the original destination region for int MUL lowering
Some hardware (CHV, BXT) have special restrictions on register regions
when doing integer multiplication. We want to respect those when we
lower to DxW multiplication.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Cc: mesa-stable@lists.freedesktop.org
| -rw-r--r-- | src/intel/compiler/brw_fs.cpp | 16 |
1 files changed, 9 insertions, 7 deletions
diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp index fea5a145a8e..fdbc6dbd0ea 100644 --- a/src/intel/compiler/brw_fs.cpp +++ b/src/intel/compiler/brw_fs.cpp @@ -3489,18 +3489,20 @@ fs_visitor::lower_integer_multiplication() bool needs_mov = false; fs_reg orig_dst = inst->dst; + fs_reg low = inst->dst; if (orig_dst.is_null() || orig_dst.file == MRF || regions_overlap(inst->dst, inst->size_written, inst->src[0], inst->size_read(0)) || regions_overlap(inst->dst, inst->size_written, inst->src[1], inst->size_read(1))) { needs_mov = true; - inst->dst = fs_reg(VGRF, alloc.allocate(dispatch_width / 8), - inst->dst.type); + low.nr = alloc.allocate(regs_written(inst)); + low.offset = low.offset % REG_SIZE; } - fs_reg low = inst->dst; - fs_reg high(VGRF, alloc.allocate(dispatch_width / 8), - inst->dst.type); + + fs_reg high = inst->dst; + high.nr = alloc.allocate(regs_written(inst)); + high.offset = high.offset % REG_SIZE; if (devinfo->gen >= 7) { if (inst->src[1].file == IMM) { @@ -3521,13 +3523,13 @@ fs_visitor::lower_integer_multiplication() inst->src[1]); } - ibld.ADD(subscript(inst->dst, BRW_REGISTER_TYPE_UW, 1), + ibld.ADD(subscript(low, BRW_REGISTER_TYPE_UW, 1), subscript(low, BRW_REGISTER_TYPE_UW, 1), subscript(high, BRW_REGISTER_TYPE_UW, 0)); if (needs_mov || inst->conditional_mod) { set_condmod(inst->conditional_mod, - ibld.MOV(orig_dst, inst->dst)); + ibld.MOV(orig_dst, low)); } } |