[Mesa-dev] [PATCH v3 40/42] intel/compiler: support half-float in the combine constants pass
Iago Toral
itoral at igalia.com
Tue Jan 29 13:27:46 UTC 2019
On Thu, 2019-01-17 at 18:18 -0600, Jason Ekstrand wrote:
> On Tue, Jan 15, 2019 at 7:55 AM Iago Toral Quiroga <itoral at igalia.com
> > wrote:
> > Reviewed-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
> >
> > ---
> >
> > .../compiler/brw_fs_combine_constants.cpp | 60
> > +++++++++++++++----
> >
> > 1 file changed, 49 insertions(+), 11 deletions(-)
> >
> >
> >
> > diff --git a/src/intel/compiler/brw_fs_combine_constants.cpp
> > b/src/intel/compiler/brw_fs_combine_constants.cpp
> >
> > index 7343f77bb45..54017e5668b 100644
> >
> > --- a/src/intel/compiler/brw_fs_combine_constants.cpp
> >
> > +++ b/src/intel/compiler/brw_fs_combine_constants.cpp
> >
> > @@ -36,6 +36,7 @@
> >
> >
> >
> > #include "brw_fs.h"
> >
> > #include "brw_cfg.h"
> >
> > +#include "util/half_float.h"
> >
> >
> >
> > using namespace brw;
> >
> >
> >
> > @@ -114,8 +115,9 @@ struct imm {
> >
> > */
> >
> > exec_list *uses;
> >
> >
> >
> > - /** The immediate value. We currently only handle floats. */
> >
> > + /** The immediate value. We currently only handle float and
> > half-float. */
> >
> > float val;
> >
> > + brw_reg_type type;
>
> I had a brief chat with Matt today and I think that this may be going
> in the wrong direction. In particular, I'd like us to eventually
> (maybe we can do it now?) generalize the combine_constants pass to
> more data types; in particular, integers. I recently came across a
> shader where the fact that we couldn't do combine_constants on
> integers was causing significant register pressure problems and
> spilling. (The test was doing a bunch of BFI2/BFE with constant
> sources.) It could also be a huge win for 8-bit and 64-bit where we
> can't put immediates in regular 2-src instructions.
> What does this mean for the pass? I suspect we want a bit size
> instead of a type and a simple char[8] for the data and just make it
> a blob of bits. We may also want some sort of heuristic so we don't
> burn constant table space for things that are only used once or maybe
> even twice.
I started working on it yesterday and have a skeleton that is currently
capable of handling F, HF, D, UD, W, UW immediates. Seems to be working
going by a bunch of selected tests from CTS I have around but I'd like
to discuss a few things before I move forward:
The combine constants pass is very specific about the instructions it
acts on. Currently it only acts in two scenarios:
1. When we want to take advantage of co-issuing (this is gen7 only
apparently).
2. When we need to fix instructions that can't take an immediate
argument (such as MAD or LRP)
In order to fix the issue you mention above with BFI2/BFE, we only need
to let constant propagation happen on these instructions and the add
them to the list of opcodes under must_promote() like we do for LRP and
MAD (plus add the code to support combinationof D/UD immediates of
course). That's what I am doing right now. The moment we decide to
let constant propagation happen for these instructions then we must
promote the immediates to GRF by the time we run the combine pass, so
heuristics won't play any role in this, as they don't for MAD or LRP.
However, you mentioned heuristics specifically, so that makes me wonder
if your idea was to have the pass promote immediates for any
instruction in the program (that support immediates, since that would
be the only case in which we're free ot make a choice), but I don't
think there is any benefit in doing that, right?
Also, I think we still need a type instead of a size, I see a few
reasons for this:
- The pass stores the absolute value of the constants it finds and
then relies on the signbit function (which only works for 32-bit and
64-bit floats) to decide whether it should negate the source whe it
rewrites it. This allows us to promote an immediate and its negated
version at the same time. As we start to support more types, including
unsigned integer types, I think we need to know the ty¡pe to do the
right thing. I think that we should at least know if the type is
unsigned.
- There is an assertion in the pass that is floating-point specific:
before it rewrites an immediate source to read from the promoted
constant, it checks that the absolute value of the original immediate
source matches what we stored in the table, or that both are NaN. This
assertion doesn't make sense as is for integers, so we need to either
know the type to have a type-specific assertion or drop the assertion.
- Some restrictions take place only on specific types. For example,
there is a restriction on BDW that affects ALIGN16 with half-float, and
we need to account for that in the pass when we promote the constant,
but there is no such restriction for W or UW.
Thoughts?
Iago
> Normally, I would say "do it as a fixup" but if we go the direction
> of having a float and using _mesa_half_to_float and
> _mesa_float_to_half, I suspect it'll be harder to go for the bag-of-
> bits approach.
>
> Thoughts?
> --Jason
>
> >
> > /**
> >
> > * The GRF register and subregister number where we've decided
> > to store the
> >
> > @@ -145,10 +147,10 @@ struct table {
> >
> > };
> >
> >
> >
> > static struct imm *
> >
> > -find_imm(struct table *table, float val)
> >
> > +find_imm(struct table *table, float val, brw_reg_type type)
> >
> > {
> >
> > for (int i = 0; i < table->len; i++) {
> >
> > - if (table->imm[i].val == val) {
> >
> > + if (table->imm[i].val == val && table->imm[i].type == type)
> > {
> >
> > return &table->imm[i];
> >
> > }
> >
> > }
> >
> > @@ -190,6 +192,20 @@ compare(const void *_a, const void *_b)
> >
> > return a->first_use_ip - b->first_use_ip;
> >
> > }
> >
> >
> >
> > +static bool
> >
> > +needs_negate(float reg_val, float imm_val, brw_reg_type type)
> >
> > +{
> >
> > + /* reg_val represents the immediate value in the register in
> > its original
> >
> > + * bit-size, while imm_val is always a valid 32-bit float
> > value.
> >
> > + */
> >
> > + if (type == BRW_REGISTER_TYPE_HF) {
> >
> > + uint32_t reg_val_ud = *((uint32_t *) ®_val);
> >
> > + reg_val = _mesa_half_to_float(reg_val_ud & 0xffff);
> >
> > + }
> >
> > +
> >
> > + return signbit(imm_val) != signbit(reg_val);
> >
> > +}
> >
> > +
> >
> > bool
> >
> > fs_visitor::opt_combine_constants()
> >
> > {
> >
> > @@ -215,12 +231,20 @@ fs_visitor::opt_combine_constants()
> >
> >
> >
> > for (int i = 0; i < inst->sources; i++) {
> >
> > if (inst->src[i].file != IMM ||
> >
> > - inst->src[i].type != BRW_REGISTER_TYPE_F)
> >
> > + (inst->src[i].type != BRW_REGISTER_TYPE_F &&
> >
> > + inst->src[i].type != BRW_REGISTER_TYPE_HF))
> >
> > continue;
> >
> >
> >
> > - float val = !inst->can_do_source_mods(devinfo) ? inst-
> > >src[i].f :
> >
> > - fabs(inst->src[i].f);
> >
> > - struct imm *imm = find_imm(&table, val);
> >
> > + float val;
> >
> > + if (inst->src[i].type == BRW_REGISTER_TYPE_F) {
> >
> > + val = !inst->can_do_source_mods(devinfo) ? inst-
> > >src[i].f :
> >
> > + fabs(inst->src[i].f);
> >
> > + } else {
> >
> > + val = !inst->can_do_source_mods(devinfo) ?
> >
> > + _mesa_half_to_float(inst->src[i].d & 0xffff) :
> >
> > + fabs(_mesa_half_to_float(inst->src[i].d & 0xffff));
> >
> > + }
> >
> > + struct imm *imm = find_imm(&table, val, inst-
> > >src[i].type);
> >
> >
> >
> > if (imm) {
> >
> > bblock_t *intersection = cfg_t::intersect(block, imm-
> > >block);
> >
> > @@ -238,6 +262,7 @@ fs_visitor::opt_combine_constants()
> >
> > imm->uses = new(const_ctx) exec_list();
> >
> > imm->uses->push_tail(link(const_ctx, &inst->src[i]));
> >
> > imm->val = val;
> >
> > + imm->type = inst->src[i].type;
> >
> > imm->uses_by_coissue = could_coissue(devinfo, inst);
> >
> > imm->must_promote = must_promote_imm(devinfo, inst);
> >
> > imm->first_use_ip = ip;
> >
> > @@ -278,12 +303,23 @@ fs_visitor::opt_combine_constants()
> >
> > imm->block->last_non_control_flow_inst()-
> > >next);
> >
> > const fs_builder ibld = bld.at(imm->block,
> > n).exec_all().group(1, 0);
> >
> >
> >
> > - ibld.MOV(reg, brw_imm_f(imm->val));
> >
> > + reg = retype(reg, imm->type);
> >
> > + if (imm->type == BRW_REGISTER_TYPE_F) {
> >
> > + ibld.MOV(reg, brw_imm_f(imm->val));
> >
> > + } else {
> >
> > + const uint16_t val_hf = _mesa_float_to_half(imm->val);
> >
> > + ibld.MOV(reg, retype(brw_imm_uw(val_hf),
> > BRW_REGISTER_TYPE_HF));
> >
> > + }
> >
> > imm->nr = reg.nr;
> >
> > imm->subreg_offset = reg.offset;
> >
> >
> >
> > + /* Keep offsets 32-bit aligned since we are mixing 32-bit
> > and 16-bit
> >
> > + * constants into the same register
> >
> > + *
> >
> > + * TODO: try to pack pairs of HF constants into each 32-bit
> > slot
> >
> > + */
> >
> > reg.offset += sizeof(float);
> >
> > - if (reg.offset == 8 * sizeof(float)) {
> >
> > + if (reg.offset == REG_SIZE) {
> >
> > reg.nr = alloc.allocate(1);
> >
> > reg.offset = 0;
> >
> > }
> >
> > @@ -295,12 +331,14 @@ fs_visitor::opt_combine_constants()
> >
> > foreach_list_typed(reg_link, link, link, table.imm[i].uses)
> > {
> >
> > fs_reg *reg = link->reg;
> >
> > assert((isnan(reg->f) && isnan(table.imm[i].val)) ||
> >
> > - fabsf(reg->f) == fabs(table.imm[i].val));
> >
> > + fabsf(reg->f) == fabs(table.imm[i].val) ||
> >
> > + table.imm[i].type == BRW_REGISTER_TYPE_HF);
> >
> >
> >
> > reg->file = VGRF;
> >
> > + reg->type = table.imm[i].type;
> >
> > reg->offset = table.imm[i].subreg_offset;
> >
> > reg->stride = 0;
> >
> > - reg->negate = signbit(reg->f) !=
> > signbit(table.imm[i].val);
> >
> > + reg->negate = needs_negate(reg->f, table.imm[i].val,
> > table.imm[i].type);
> >
> > reg->nr = table.imm[i].nr;
> >
> > }
> >
> > }
> >
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20190129/e8b52d07/attachment.html>
More information about the mesa-dev
mailing list