OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nir: add image fragment mask load intrinsic

Browse Source 
Like nir_texop_fragment_mask_fetch_amd, this is used to load multi
sample image fmask data for AMD GPU.

We will lower multi sample image load and samples_identical intrinsics
to use it latter for radeonsi. RADV does not need this because it
always expand fmask images before dispatch compute shader.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18666>
This commit is contained in:
Qiang Yu
2022-09-07 17:53:33 +08:00
parent 88f6ca39a7
commit 1461b5f61b
7 changed files with 17 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
src/compiler/nir/nir.c
View File

@@ -2725,6 +2725,7 @@ nir_rewrite_image_intrinsic(nir_intrinsic_instr *intrin, nir_ssa_def *src,
CASE(samples)
CASE(load_raw_intel)
CASE(store_raw_intel)
CASE(fragment_mask_load_amd)
#undef CASE
default:
unreachable("Unhanded image intrinsic");

1
src/compiler/nir/nir.h
View File

@@ -1995,6 +1995,7 @@ nir_intrinsic_can_reorder(nir_intrinsic_instr *instr)
if (nir_intrinsic_has_access(instr) &&
nir_intrinsic_access(instr) & ACCESS_VOLATILE)
return false;
if (instr->intrinsic == nir_intrinsic_load_deref) {
nir_deref_instr *deref = nir_src_as_deref(instr->src[0]);
return nir_deref_mode_is_in_set(deref, nir_var_read_only_modes) ||

3
src/compiler/nir/nir_divergence_analysis.c
View File

@@ -356,6 +356,9 @@ visit_intrinsic(nir_shader *shader, nir_intrinsic_instr *instr)
case nir_intrinsic_image_samples_identical:
case nir_intrinsic_image_deref_samples_identical:
case nir_intrinsic_bindless_image_samples_identical:
case nir_intrinsic_image_fragment_mask_load_amd:
case nir_intrinsic_image_deref_fragment_mask_load_amd:
case nir_intrinsic_bindless_image_fragment_mask_load_amd:
is_divergent = (instr->src[0].ssa->divergent && (nir_intrinsic_access(instr) & ACCESS_NON_UNIFORM)) ||
instr->src[1].ssa->divergent;
break;

3
src/compiler/nir/nir_group_loads.c
View File

@@ -91,6 +91,9 @@ get_intrinsic_resource(nir_intrinsic_instr *intr)
case nir_intrinsic_bindless_image_load:
case nir_intrinsic_bindless_image_sparse_load:
case nir_intrinsic_load_ssbo:
case nir_intrinsic_image_fragment_mask_load_amd:
case nir_intrinsic_image_deref_fragment_mask_load_amd:
case nir_intrinsic_bindless_image_fragment_mask_load_amd:
return intr->src[0].ssa->parent_instr;
default:
return NULL;

3
src/compiler/nir/nir_intrinsics.py
View File

@@ -665,6 +665,9 @@ image("descriptor_amd", dest_comp=0, src_comp=[], flags=[CAN_ELIMINATE, CAN_REOR
# CL-specific format queries
image("format", dest_comp=1, flags=[CAN_ELIMINATE, CAN_REORDER])
image("order", dest_comp=1, flags=[CAN_ELIMINATE, CAN_REORDER])
# Multisample fragment mask load
# src_comp[0] is same as image load src_comp[0]
image("fragment_mask_load_amd", src_comp=[4], dest_comp=1, bit_sizes=[32], flags=[CAN_ELIMINATE, CAN_REORDER])
# Vulkan descriptor set intrinsics
#

3
src/compiler/nir/nir_lower_non_uniform_access.c
View File

@@ -273,6 +273,7 @@ nir_lower_non_uniform_access_impl(nir_function_impl *impl,
case nir_intrinsic_image_size:
case nir_intrinsic_image_samples:
case nir_intrinsic_image_samples_identical:
case nir_intrinsic_image_fragment_mask_load_amd:
case nir_intrinsic_bindless_image_load:
case nir_intrinsic_bindless_image_sparse_load:
case nir_intrinsic_bindless_image_store:
@@ -292,6 +293,7 @@ nir_lower_non_uniform_access_impl(nir_function_impl *impl,
case nir_intrinsic_bindless_image_size:
case nir_intrinsic_bindless_image_samples:
case nir_intrinsic_bindless_image_samples_identical:
case nir_intrinsic_bindless_image_fragment_mask_load_amd:
case nir_intrinsic_image_deref_load:
case nir_intrinsic_image_deref_sparse_load:
case nir_intrinsic_image_deref_store:
@@ -311,6 +313,7 @@ nir_lower_non_uniform_access_impl(nir_function_impl *impl,
case nir_intrinsic_image_deref_size:
case nir_intrinsic_image_deref_samples:
case nir_intrinsic_image_deref_samples_identical:
case nir_intrinsic_image_deref_fragment_mask_load_amd:
if ((options->types & nir_lower_non_uniform_image_access) &&
lower_non_uniform_access_intrin(options, &b, intrin, 0))
progress = true;

3
src/compiler/nir/nir_opt_non_uniform_access.c
View File

@@ -79,6 +79,7 @@ is_image_intrinsic(nir_intrinsic_instr *intrin)
case nir_intrinsic_image_atomic_fmax:
case nir_intrinsic_image_size:
case nir_intrinsic_image_samples:
case nir_intrinsic_image_fragment_mask_load_amd:
case nir_intrinsic_bindless_image_load:
case nir_intrinsic_bindless_image_sparse_load:
case nir_intrinsic_bindless_image_store:
@@ -97,6 +98,7 @@ is_image_intrinsic(nir_intrinsic_instr *intrin)
case nir_intrinsic_bindless_image_atomic_fmax:
case nir_intrinsic_bindless_image_size:
case nir_intrinsic_bindless_image_samples:
case nir_intrinsic_bindless_image_fragment_mask_load_amd:
case nir_intrinsic_image_deref_load:
case nir_intrinsic_image_deref_sparse_load:
case nir_intrinsic_image_deref_store:
@@ -115,6 +117,7 @@ is_image_intrinsic(nir_intrinsic_instr *intrin)
case nir_intrinsic_image_deref_atomic_fmax:
case nir_intrinsic_image_deref_size:
case nir_intrinsic_image_deref_samples:
case nir_intrinsic_image_deref_fragment_mask_load_amd:
return true;
default: