Samuel Iglesias Gonsálvez <[email protected]> writes: > From: "Juan A. Suarez Romero" <[email protected]> > > Take in account the offset value when getting the var from register. > > This is required when dealing with an operation that writes half of the > register (like one d2x in IVB/BYT, which uses exec_size == 4). > > Note that for live analysis variables we need to stick to per-GRF > analysis. In this case, we use var_from_reg() with GRF precision.
This looks bogus to me. If the specified register had a non-zero
sub-GRF offset and we weren't taking it into account we have a bug. The
vec4 liveness analysis pass keeps track of dataflow with 32-bit
granularity these days so we want the variable index computation to be
done with the same precision.
> ---
> src/mesa/drivers/dri/i965/brw_vec4_live_variables.cpp | 12 ++++++++----
> src/mesa/drivers/dri/i965/brw_vec4_live_variables.h | 10 ++++++----
> 2 files changed, 14 insertions(+), 8 deletions(-)
>
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_live_variables.cpp
> b/src/mesa/drivers/dri/i965/brw_vec4_live_variables.cpp
> index 73f658cd8fa..54ebd0994ee 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4_live_variables.cpp
> +++ b/src/mesa/drivers/dri/i965/brw_vec4_live_variables.cpp
> @@ -78,7 +78,8 @@ vec4_live_variables::setup_def_use()
> if (inst->src[i].file == VGRF) {
> for (unsigned j = 0; j < DIV_ROUND_UP(inst->size_read(i),
> 16); j++) {
> for (int c = 0; c < 4; c++) {
> - const unsigned v = var_from_reg(alloc, inst->src[i], c,
> j);
> + const unsigned v =
> + var_from_reg(alloc, inst->src[i], c, j, false);
> if (!BITSET_TEST(bd->def, v))
> BITSET_SET(bd->use, v);
> }
> @@ -101,7 +102,8 @@ vec4_live_variables::setup_def_use()
> for (unsigned i = 0; i < DIV_ROUND_UP(inst->size_written, 16);
> i++) {
> for (int c = 0; c < 4; c++) {
> if (inst->dst.writemask & (1 << c)) {
> - const unsigned v = var_from_reg(alloc, inst->dst, c, i);
> + const unsigned v =
> + var_from_reg(alloc, inst->dst, c, i, false);
> if (!BITSET_TEST(bd->use, v))
> BITSET_SET(bd->def, v);
> }
> @@ -257,7 +259,8 @@ vec4_visitor::calculate_live_intervals()
> if (inst->src[i].file == VGRF) {
> for (unsigned j = 0; j < DIV_ROUND_UP(inst->size_read(i), 16);
> j++) {
> for (int c = 0; c < 4; c++) {
> - const unsigned v = var_from_reg(alloc, inst->src[i], c, j);
> + const unsigned v =
> + var_from_reg(alloc, inst->src[i], c, j, false);
> start[v] = MIN2(start[v], ip);
> end[v] = ip;
> }
> @@ -269,7 +272,8 @@ vec4_visitor::calculate_live_intervals()
> for (unsigned i = 0; i < DIV_ROUND_UP(inst->size_written, 16); i++)
> {
> for (int c = 0; c < 4; c++) {
> if (inst->dst.writemask & (1 << c)) {
> - const unsigned v = var_from_reg(alloc, inst->dst, c, i);
> + const unsigned v =
> + var_from_reg(alloc, inst->dst, c, i, false);
> start[v] = MIN2(start[v], ip);
> end[v] = ip;
> }
> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_live_variables.h
> b/src/mesa/drivers/dri/i965/brw_vec4_live_variables.h
> index 8807c453743..9c505d15f1f 100644
> --- a/src/mesa/drivers/dri/i965/brw_vec4_live_variables.h
> +++ b/src/mesa/drivers/dri/i965/brw_vec4_live_variables.h
> @@ -83,13 +83,14 @@ protected:
> */
> inline unsigned
> var_from_reg(const simple_allocator &alloc, const src_reg ®,
> - unsigned c = 0, unsigned k = 0)
> + unsigned c = 0, unsigned k = 0, bool subreg_precision = true)
> {
> assert(reg.file == VGRF && reg.nr < alloc.count && c < 4);
> const unsigned csize = DIV_ROUND_UP(type_sz(reg.type), 4);
> unsigned result =
> 8 * (alloc.offsets[reg.nr] + reg.offset / REG_SIZE) +
> - (BRW_GET_SWZ(reg.swizzle, c) + k / csize * 4) * csize + k % csize;
> + (BRW_GET_SWZ(reg.swizzle, c) + k / csize * 4) * csize + k % csize +
> + (subreg_precision ? (reg.offset % REG_SIZE) / type_sz(reg.type) : 0);
> /* Do not exceed the limit for this register */
> assert(result < 8 * (alloc.offsets[reg.nr] + alloc.sizes[reg.nr]));
> return result;
> @@ -97,13 +98,14 @@ var_from_reg(const simple_allocator &alloc, const src_reg
> ®,
>
> inline unsigned
> var_from_reg(const simple_allocator &alloc, const dst_reg ®,
> - unsigned c = 0, unsigned k = 0)
> + unsigned c = 0, unsigned k = 0, bool subreg_precision = true)
> {
> assert(reg.file == VGRF && reg.nr < alloc.count && c < 4);
> const unsigned csize = DIV_ROUND_UP(type_sz(reg.type), 4);
> unsigned result =
> 8 * (alloc.offsets[reg.nr] + reg.offset / REG_SIZE) +
> - (c + k / csize * 4) * csize + k % csize;
> + (c + k / csize * 4) * csize + k % csize +
> + (subreg_precision ? (reg.offset % REG_SIZE) / type_sz(reg.type) : 0);
> /* Do not exceed the limit for this register */
> assert(result < 8 * (alloc.offsets[reg.nr] + alloc.sizes[reg.nr]));
> return result;
> --
> 2.11.0
>
> _______________________________________________
> mesa-dev mailing list
> [email protected]
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev
signature.asc
Description: PGP signature
_______________________________________________ mesa-dev mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-dev
