nir: Add floating point atomic add instrinsics

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This commit is contained in:
Ian Romanick 2018-04-09 18:36:22 -07:00
parent a390158d10
commit 69ce7baa9e
5 changed files with 22 additions and 5 deletions

View File

@ -240,6 +240,7 @@ lower_intrinsic(nir_intrinsic_instr *instr,
instr->intrinsic == nir_intrinsic_image_deref_atomic_xor ||
instr->intrinsic == nir_intrinsic_image_deref_atomic_exchange ||
instr->intrinsic == nir_intrinsic_image_deref_atomic_comp_swap ||
instr->intrinsic == nir_intrinsic_image_deref_atomic_fadd ||
instr->intrinsic == nir_intrinsic_image_deref_size) {
b->cursor = nir_before_instr(&instr->instr);

View File

@ -650,7 +650,9 @@ nir_visitor::visit(ir_call *ir)
op = nir_intrinsic_image_deref_store;
break;
case ir_intrinsic_image_atomic_add:
op = nir_intrinsic_image_deref_atomic_add;
op = ir->return_deref->type->is_integer_32_64()
? nir_intrinsic_image_deref_atomic_add
: nir_intrinsic_image_deref_atomic_fadd;
break;
case ir_intrinsic_image_atomic_min:
op = nir_intrinsic_image_deref_atomic_min;
@ -689,7 +691,8 @@ nir_visitor::visit(ir_call *ir)
op = nir_intrinsic_load_ssbo;
break;
case ir_intrinsic_ssbo_atomic_add:
op = nir_intrinsic_ssbo_atomic_add;
op = ir->return_deref->type->is_integer_32_64()
? nir_intrinsic_ssbo_atomic_add : nir_intrinsic_ssbo_atomic_fadd;
break;
case ir_intrinsic_ssbo_atomic_and:
op = nir_intrinsic_ssbo_atomic_and;
@ -755,7 +758,9 @@ nir_visitor::visit(ir_call *ir)
op = nir_intrinsic_store_shared;
break;
case ir_intrinsic_shared_atomic_add:
op = nir_intrinsic_shared_atomic_add;
op = ir->return_deref->type->is_integer_32_64()
? nir_intrinsic_shared_atomic_add
: nir_intrinsic_shared_atomic_fadd;
break;
case ir_intrinsic_shared_atomic_and:
op = nir_intrinsic_shared_atomic_and;
@ -865,6 +870,7 @@ nir_visitor::visit(ir_call *ir)
case nir_intrinsic_image_deref_atomic_xor:
case nir_intrinsic_image_deref_atomic_exchange:
case nir_intrinsic_image_deref_atomic_comp_swap:
case nir_intrinsic_image_deref_atomic_fadd:
case nir_intrinsic_image_deref_samples:
case nir_intrinsic_image_deref_size: {
nir_ssa_undef_instr *instr_undef =
@ -1035,7 +1041,8 @@ nir_visitor::visit(ir_call *ir)
case nir_intrinsic_ssbo_atomic_or:
case nir_intrinsic_ssbo_atomic_xor:
case nir_intrinsic_ssbo_atomic_exchange:
case nir_intrinsic_ssbo_atomic_comp_swap: {
case nir_intrinsic_ssbo_atomic_comp_swap:
case nir_intrinsic_ssbo_atomic_fadd: {
int param_count = ir->actual_parameters.length();
assert(param_count == 3 || param_count == 4);
@ -1118,7 +1125,8 @@ nir_visitor::visit(ir_call *ir)
case nir_intrinsic_shared_atomic_or:
case nir_intrinsic_shared_atomic_xor:
case nir_intrinsic_shared_atomic_exchange:
case nir_intrinsic_shared_atomic_comp_swap: {
case nir_intrinsic_shared_atomic_comp_swap:
case nir_intrinsic_shared_atomic_fadd: {
int param_count = ir->actual_parameters.length();
assert(param_count == 2 || param_count == 3);

View File

@ -307,6 +307,7 @@ intrinsic("image_deref_atomic_or", src_comp=[1, 4, 1, 1], dest_comp=1)
intrinsic("image_deref_atomic_xor", src_comp=[1, 4, 1, 1], dest_comp=1)
intrinsic("image_deref_atomic_exchange", src_comp=[1, 4, 1, 1], dest_comp=1)
intrinsic("image_deref_atomic_comp_swap", src_comp=[1, 4, 1, 1, 1], dest_comp=1)
intrinsic("image_deref_atomic_fadd", src_comp=[1, 4, 1, 1], dest_comp=1)
intrinsic("image_deref_size", src_comp=[1], dest_comp=0, flags=[CAN_ELIMINATE, CAN_REORDER])
intrinsic("image_deref_samples", src_comp=[1], dest_comp=1, flags=[CAN_ELIMINATE, CAN_REORDER])
@ -357,6 +358,7 @@ intrinsic("deref_atomic_or", src_comp=[1, 1], dest_comp=1)
intrinsic("deref_atomic_xor", src_comp=[1, 1], dest_comp=1)
intrinsic("deref_atomic_exchange", src_comp=[1, 1], dest_comp=1)
intrinsic("deref_atomic_comp_swap", src_comp=[1, 1, 1], dest_comp=1)
intrinsic("deref_atomic_fadd", src_comp=[1, 1], dest_comp=1)
# SSBO atomic intrinsics
#
@ -383,6 +385,7 @@ intrinsic("ssbo_atomic_or", src_comp=[1, 1, 1], dest_comp=1)
intrinsic("ssbo_atomic_xor", src_comp=[1, 1, 1], dest_comp=1)
intrinsic("ssbo_atomic_exchange", src_comp=[1, 1, 1], dest_comp=1)
intrinsic("ssbo_atomic_comp_swap", src_comp=[1, 1, 1, 1], dest_comp=1)
intrinsic("ssbo_atomic_fadd", src_comp=[1, 1, 1], dest_comp=1)
# CS shared variable atomic intrinsics
#
@ -408,6 +411,7 @@ intrinsic("shared_atomic_or", src_comp=[1, 1], dest_comp=1, indices=[BASE])
intrinsic("shared_atomic_xor", src_comp=[1, 1], dest_comp=1, indices=[BASE])
intrinsic("shared_atomic_exchange", src_comp=[1, 1], dest_comp=1, indices=[BASE])
intrinsic("shared_atomic_comp_swap", src_comp=[1, 1, 1], dest_comp=1, indices=[BASE])
intrinsic("shared_atomic_fadd", src_comp=[1, 1], dest_comp=1, indices=[BASE])
def system_value(name, dest_comp, indices=[]):
intrinsic("load_" + name, [], dest_comp, indices,

View File

@ -57,6 +57,7 @@ lower_instr(nir_intrinsic_instr *instr, unsigned ssbo_offset, nir_builder *b)
case nir_intrinsic_ssbo_atomic_xor:
case nir_intrinsic_ssbo_atomic_exchange:
case nir_intrinsic_ssbo_atomic_comp_swap:
case nir_intrinsic_ssbo_atomic_fadd:
case nir_intrinsic_store_ssbo:
case nir_intrinsic_load_ssbo:
case nir_intrinsic_get_buffer_size:

View File

@ -280,6 +280,7 @@ lower_atomic(nir_intrinsic_instr *intrin, struct lower_io_state *state,
OP(atomic_and)
OP(atomic_or)
OP(atomic_xor)
OP(atomic_fadd)
#undef OP
default:
unreachable("Invalid atomic");
@ -380,6 +381,7 @@ nir_lower_io_block(nir_block *block,
case nir_intrinsic_deref_atomic_xor:
case nir_intrinsic_deref_atomic_exchange:
case nir_intrinsic_deref_atomic_comp_swap:
case nir_intrinsic_deref_atomic_fadd:
/* We can lower the io for this nir instrinsic */
break;
case nir_intrinsic_interp_deref_at_centroid:
@ -441,6 +443,7 @@ nir_lower_io_block(nir_block *block,
case nir_intrinsic_deref_atomic_xor:
case nir_intrinsic_deref_atomic_exchange:
case nir_intrinsic_deref_atomic_comp_swap:
case nir_intrinsic_deref_atomic_fadd:
assert(vertex_index == NULL);
replacement = lower_atomic(intrin, state, var, offset);
break;