nir/vtn: add support for SpvBuiltInGlobalLinearId

v2: use formula with fewer operations

Signed-off-by: Karol Herbst <[email protected]>
Reviewed-by: Jason Ekstrand <[email protected]>
Reviewed-by: Bas Nieuwenhuizen <[email protected]>

3 files changed, 40 insertions, 12 deletions

diff --git a/src/compiler/nir/nir.c b/src/compiler/nir/nir.c
index d759dfdce21..4a7c757db3a 100644
--- a/src/compiler/nir/nir.c
+++ b/src/compiler/nir/nir.c
@@ -1860,6 +1860,8 @@ nir_intrinsic_from_system_value(gl_system_value val)
       return nir_intrinsic_load_local_group_size;
    case SYSTEM_VALUE_GLOBAL_INVOCATION_ID:
       return nir_intrinsic_load_global_invocation_id;
+   case SYSTEM_VALUE_GLOBAL_INVOCATION_INDEX:
+      return nir_intrinsic_load_global_invocation_index;
    case SYSTEM_VALUE_WORK_DIM:
       return nir_intrinsic_load_work_dim;
    default:
diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index d53b26c88d6..1ae453f757d 100644
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -533,6 +533,7 @@ system_value("num_subgroups", 1)
 system_value("subgroup_id", 1)
 system_value("local_group_size", 3)
 system_value("global_invocation_id", 3, bit_sizes=[32, 64])
+system_value("global_invocation_index", 1, bit_sizes=[32, 64])
 system_value("work_dim", 1)
 # Driver-specific viewport scale/offset parameters.
 #
diff --git a/src/compiler/nir/nir_lower_system_values.c b/src/compiler/nir/nir_lower_system_values.c
index de5ccab0f38..4436799fa52 100644
--- a/src/compiler/nir/nir_lower_system_values.c
+++ b/src/compiler/nir/nir_lower_system_values.c
@@ -99,6 +99,29 @@ build_local_invocation_id(nir_builder *b, unsigned bit_size)
    }
 }
 
+static nir_ssa_def*
+build_global_group_size(nir_builder *b, unsigned bit_size)
+{
+   nir_ssa_def *group_size = build_local_group_size(b, bit_size);
+   nir_ssa_def *num_work_groups = nir_u2u(b, nir_load_num_work_groups(b), bit_size);
+   return nir_imul(b, group_size, num_work_groups);
+}
+
+static nir_ssa_def*
+build_global_invocation_id(nir_builder *b, unsigned bit_size)
+{
+   /* From the GLSL man page for gl_GlobalInvocationID:
+    *
+    *    "The value of gl_GlobalInvocationID is equal to
+    *    gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID"
+    */
+   nir_ssa_def *group_size = build_local_group_size(b, bit_size);
+   nir_ssa_def *group_id = nir_u2u(b, nir_load_work_group_id(b), bit_size);
+   nir_ssa_def *local_id = build_local_invocation_id(b, bit_size);
+
+   return nir_iadd(b, nir_imul(b, group_id, group_size), local_id);
+}
+
 static bool
 convert_block(nir_block *block, nir_builder *b)
 {
@@ -133,16 +156,20 @@ convert_block(nir_block *block, nir_builder *b)
       nir_ssa_def *sysval = NULL;
       switch (var->data.location) {
       case SYSTEM_VALUE_GLOBAL_INVOCATION_ID: {
-         /* From the GLSL man page for gl_GlobalInvocationID:
-          *
-          *    "The value of gl_GlobalInvocationID is equal to
-          *    gl_WorkGroupID * gl_WorkGroupSize + gl_LocalInvocationID"
-          */
-         nir_ssa_def *group_size = build_local_group_size(b, bit_size);
-         nir_ssa_def *group_id = nir_u2u(b, nir_load_work_group_id(b), bit_size);
-         nir_ssa_def *local_id = build_local_invocation_id(b, bit_size);
+         sysval = build_global_invocation_id(b, bit_size);
+         break;
+      }
+
+      case SYSTEM_VALUE_GLOBAL_INVOCATION_INDEX: {
+         nir_ssa_def *global_id = build_global_invocation_id(b, bit_size);
+         nir_ssa_def *global_size = build_global_group_size(b, bit_size);
 
-         sysval = nir_iadd(b, nir_imul(b, group_id, group_size), local_id);
+         /* index = id.x + ((id.y + (id.z * size.y)) * size.x) */
+         sysval = nir_imul(b, nir_channel(b, global_id, 2),
+                              nir_channel(b, global_size, 1));
+         sysval = nir_iadd(b, nir_channel(b, global_id, 1), sysval);
+         sysval = nir_imul(b, nir_channel(b, global_size, 0), sysval);
+         sysval = nir_iadd(b, nir_channel(b, global_id, 0), sysval);
          break;
       }
 
@@ -259,9 +286,7 @@ convert_block(nir_block *block, nir_builder *b)
          break;
 
       case SYSTEM_VALUE_GLOBAL_GROUP_SIZE: {
-         nir_ssa_def *group_size = build_local_group_size(b, bit_size);
-         nir_ssa_def *num_work_groups = nir_u2u(b, nir_load_num_work_groups(b), bit_size);
-         sysval = nir_imul(b, group_size, num_work_groups);
+         sysval = build_global_group_size(b, bit_size);
          break;
       }