nir: Add per-primitive I/O intrinsics

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/10600>

diff --git a/src/compiler/nir/nir_gather_info.c b/src/compiler/nir/nir_gather_info.c
index 9472f898c06801d150deb6e38600e8444fb523e3..b44a04bf2abeb7681557ed1a021d1cfaccae4629 100644 (file)
--- a/src/compiler/nir/nir_gather_info.c
+++ b/src/compiler/nir/nir_gather_info.c
@@ -533,6 +533,7 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
 
    case nir_intrinsic_load_output:
    case nir_intrinsic_load_per_vertex_output:
+   case nir_intrinsic_load_per_primitive_output:
       if (shader->info.stage == MESA_SHADER_TESS_CTRL &&
           instr->intrinsic == nir_intrinsic_load_output) {
          shader->info.patch_outputs_read |= slot_mask;
@@ -559,6 +560,7 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
 
    case nir_intrinsic_store_output:
    case nir_intrinsic_store_per_vertex_output:
+   case nir_intrinsic_store_per_primitive_output:
       if (shader->info.stage == MESA_SHADER_TESS_CTRL &&
           instr->intrinsic == nir_intrinsic_store_output) {
          shader->info.patch_outputs_written |= slot_mask;

diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index f3f62ad452b520a6afeb206f4b96cdfa993d6bf3..fdc5e7882eee3e1f60e24fed7c81571401d02990 100644 (file)
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -921,6 +921,8 @@ load("ssbo_address", [1], [], [CAN_ELIMINATE, CAN_REORDER])
 load("output", [1], [BASE, COMPONENT, DEST_TYPE, IO_SEMANTICS], flags=[CAN_ELIMINATE])
 # src[] = { vertex, offset }.
 load("per_vertex_output", [1, 1], [BASE, COMPONENT, DEST_TYPE, IO_SEMANTICS], [CAN_ELIMINATE])
+# src[] = { primitive, offset }.
+load("per_primitive_output", [1, 1], [BASE, COMPONENT, DEST_TYPE, IO_SEMANTICS], [CAN_ELIMINATE])
 # src[] = { offset }.
 load("shared", [1], [BASE, ALIGN_MUL, ALIGN_OFFSET], [CAN_ELIMINATE])
 # src[] = { offset }.
@@ -956,6 +958,8 @@ def store(name, srcs, indices=[], flags=[]):
 store("output", [1], [BASE, WRITE_MASK, COMPONENT, SRC_TYPE, IO_SEMANTICS])
 # src[] = { value, vertex, offset }.
 store("per_vertex_output", [1, 1], [BASE, WRITE_MASK, COMPONENT, SRC_TYPE, IO_SEMANTICS])
+# src[] = { value, primitive, offset }.
+store("per_primitive_output", [1, 1], [BASE, WRITE_MASK, COMPONENT, SRC_TYPE, IO_SEMANTICS])
 # src[] = { value, block_index, offset }
 store("ssbo", [-1, 1], [WRITE_MASK, ACCESS, ALIGN_MUL, ALIGN_OFFSET])
 # src[] = { value, offset }.

diff --git a/src/compiler/nir/nir_lower_io.c b/src/compiler/nir/nir_lower_io.c
index 71bd31c618b297a67a052124eaacac3c7d4e4385..c9aa0b26e9c0757842530acc54d9d402dc0e4aea 100644 (file)
--- a/src/compiler/nir/nir_lower_io.c
+++ b/src/compiler/nir/nir_lower_io.c
@@ -282,8 +282,9 @@ emit_load(struct lower_io_state *state,
       }
       break;
    case nir_var_shader_out:
-      op = array_index ? nir_intrinsic_load_per_vertex_output :
-                         nir_intrinsic_load_output;
+      op = !array_index            ? nir_intrinsic_load_output :
+           var->data.per_primitive ? nir_intrinsic_load_per_primitive_output :
+                                     nir_intrinsic_load_per_vertex_output;
       break;
    case nir_var_uniform:
       op = nir_intrinsic_load_uniform;
@@ -394,8 +395,9 @@ emit_store(struct lower_io_state *state, nir_ssa_def *data,
 
    assert(var->data.mode == nir_var_shader_out);
    nir_intrinsic_op op =
-      array_index ? nir_intrinsic_store_per_vertex_output :
-                    nir_intrinsic_store_output;
+      !array_index            ? nir_intrinsic_store_output :
+      var->data.per_primitive ? nir_intrinsic_store_per_primitive_output :
+                                nir_intrinsic_store_per_vertex_output;
 
    nir_intrinsic_instr *store =
       nir_intrinsic_instr_create(state->builder.shader, op);
@@ -2472,6 +2474,7 @@ nir_get_io_offset_src(nir_intrinsic_instr *instr)
    case nir_intrinsic_load_input_vertex:
    case nir_intrinsic_load_per_vertex_input:
    case nir_intrinsic_load_per_vertex_output:
+   case nir_intrinsic_load_per_primitive_output:
    case nir_intrinsic_load_interpolated_input:
    case nir_intrinsic_store_output:
    case nir_intrinsic_store_shared:
@@ -2494,6 +2497,7 @@ nir_get_io_offset_src(nir_intrinsic_instr *instr)
       return &instr->src[1];
    case nir_intrinsic_store_ssbo:
    case nir_intrinsic_store_per_vertex_output:
+   case nir_intrinsic_store_per_primitive_output:
       return &instr->src[2];
    default:
       return NULL;
@@ -2632,8 +2636,10 @@ is_output(nir_intrinsic_instr *intrin)
 {
    return intrin->intrinsic == nir_intrinsic_load_output ||
           intrin->intrinsic == nir_intrinsic_load_per_vertex_output ||
+          intrin->intrinsic == nir_intrinsic_load_per_primitive_output ||
           intrin->intrinsic == nir_intrinsic_store_output ||
-          intrin->intrinsic == nir_intrinsic_store_per_vertex_output;
+          intrin->intrinsic == nir_intrinsic_store_per_vertex_output ||
+          intrin->intrinsic == nir_intrinsic_store_per_primitive_output;
 }
 
 static bool is_dual_slot(nir_intrinsic_instr *intrin)

diff --git a/src/compiler/nir/nir_opt_undef.c b/src/compiler/nir/nir_opt_undef.c
index 20b31a4adad477691b946e970ee35eeedd85c5b0..fd8fb2c759016e01f04d79f8ec0e28fe125dd4ae 100644 (file)
--- a/src/compiler/nir/nir_opt_undef.c
+++ b/src/compiler/nir/nir_opt_undef.c
@@ -137,6 +137,7 @@ opt_undef_store(nir_intrinsic_instr *intrin)
       break;
    case nir_intrinsic_store_output:
    case nir_intrinsic_store_per_vertex_output:
+   case nir_intrinsic_store_per_primitive_output:
    case nir_intrinsic_store_ssbo:
    case nir_intrinsic_store_shared:
    case nir_intrinsic_store_global:

diff --git a/src/compiler/nir/nir_validate.c b/src/compiler/nir/nir_validate.c
index 648e5af7a1a19fe4173211b9338b79a047004572..b27cbc01d2642e243f745b481bf0300c8dc1f2ec 100644 (file)
--- a/src/compiler/nir/nir_validate.c
+++ b/src/compiler/nir/nir_validate.c
@@ -676,6 +676,7 @@ validate_intrinsic_instr(nir_intrinsic_instr *instr, validate_state *state)
    case nir_intrinsic_load_interpolated_input:
    case nir_intrinsic_load_output:
    case nir_intrinsic_load_per_vertex_output:
+   case nir_intrinsic_load_per_primitive_output:
    case nir_intrinsic_load_push_constant:
       /* All memory load operations must load at least a byte */
       validate_assert(state, nir_dest_bit_size(instr->dest) >= 8);