nir: update opcode definitions for different bit sizes

Some opcodes need explicit bitsizes, and sometimes we need to use the
double version when constant folding.

v2: fix output type for u2f (Iago)

v3: do not change vecN opcodes to be float. The next commit will add
    infrastructure to enable 64-bit integer constant folding so this is isn't
    really necessary. Also, that created problems with source modifiers in
    some cases (Iago)

v4 (Jason):
  - do not change bcsel to work in terms of floats
  - leave ldexp generic

Squashed changes to handle different bit sizes when constant
folding since otherwise we would break the build.

v2:
- Use the bit-size information from the opcode information if defined (Iago)
- Use helpers to get type size and base type of nir_alu_type enum (Sam)
- Do not fallback to sized types to guess bit-size information. (Jason)

Squashed changes in i965 and gallium/nir drivers to support sized types.
These functions should only see sized types, but we can't make that change
until we make sure that nir uses the sized versions in all the relevant places.
A later commit will address this.

Signed-off-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
This commit is contained in:
Connor Abbott
2015-08-14 10:45:06 -07:00
committed by Samuel Iglesias Gonsálvez
parent 6700d7e423
commit 9076c4e289
7 changed files with 282 additions and 157 deletions

View File

@@ -46,10 +46,28 @@ constant_fold_alu_instr(nir_alu_instr *instr, void *mem_ctx)
if (!instr->dest.dest.is_ssa)
return false;
/* In the case that any outputs/inputs have unsized types, then we need to
* guess the bit-size. In this case, the validator ensures that all
* bit-sizes match so we can just take the bit-size from first
* output/input with an unsized type. If all the outputs/inputs are sized
* then we don't need to guess the bit-size at all because the code we
* generate for constant opcodes in this case already knows the sizes of
* the types involved and does not need the provided bit-size for anything
* (although it still requires to receive a valid bit-size).
*/
unsigned bit_size = 0;
if (!nir_alu_type_get_type_size(nir_op_infos[instr->op].output_type))
bit_size = instr->dest.dest.ssa.bit_size;
for (unsigned i = 0; i < nir_op_infos[instr->op].num_inputs; i++) {
if (!instr->src[i].src.is_ssa)
return false;
if (bit_size == 0 &&
!nir_alu_type_get_type_size(nir_op_infos[instr->op].input_sizes[i])) {
bit_size = instr->src[i].src.ssa->bit_size;
}
nir_instr *src_instr = instr->src[i].src.ssa->parent_instr;
if (src_instr->type != nir_instr_type_load_const)
@@ -58,24 +76,31 @@ constant_fold_alu_instr(nir_alu_instr *instr, void *mem_ctx)
for (unsigned j = 0; j < nir_ssa_alu_instr_src_components(instr, i);
j++) {
src[i].u[j] = load_const->value.u[instr->src[i].swizzle[j]];
if (load_const->def.bit_size == 64)
src[i].ul[j] = load_const->value.ul[instr->src[i].swizzle[j]];
else
src[i].u[j] = load_const->value.u[instr->src[i].swizzle[j]];
}
/* We shouldn't have any source modifiers in the optimization loop. */
assert(!instr->src[i].abs && !instr->src[i].negate);
}
if (bit_size == 0)
bit_size = 32;
/* We shouldn't have any saturate modifiers in the optimization loop. */
assert(!instr->dest.saturate);
nir_const_value dest =
nir_eval_const_opcode(instr->op, instr->dest.dest.ssa.num_components,
src);
bit_size, src);
nir_load_const_instr *new_instr =
nir_load_const_instr_create(mem_ctx,
instr->dest.dest.ssa.num_components);
new_instr->def.bit_size = instr->dest.dest.ssa.bit_size;
new_instr->value = dest;
nir_instr_insert_before(&instr->instr, &new_instr->instr);