diff --git a/src/asahi/compiler/agx_compile.c b/src/asahi/compiler/agx_compile.c
index 0cf5bfde604..89c9c0a831a 100644
--- a/src/asahi/compiler/agx_compile.c
+++ b/src/asahi/compiler/agx_compile.c
@@ -2717,7 +2717,27 @@ agx_optimize_nir(nir_shader *nir, unsigned *preamble_size)
             });
    NIR_PASS(_, nir, nir_lower_pack);
 
+   nir_convert_to_lcssa(nir, true, true);
+   NIR_PASS_V(nir, nir_divergence_analysis);
    bool progress = false;
+
+   static const nir_lower_subgroups_options subgroups_options = {
+      .ballot_bit_size = 32,
+      .ballot_components = 1,
+      .lower_elect = true,
+      .lower_subgroup_masks = true,
+   };
+
+   NIR_PASS(progress, nir, nir_opt_uniform_atomics, true);
+   NIR_PASS(progress, nir, nir_opt_uniform_subgroup, &subgroups_options);
+
+   /* The above create operations that need lowering/optimizing */
+   if (progress) {
+      NIR_PASS(_, nir, agx_nir_lower_subgroups);
+      NIR_PASS(_, nir, nir_opt_algebraic);
+   }
+
+   progress = false;
    NIR_PASS(progress, nir, agx_nir_lower_address);
 
    /* If address lowering made progress, clean up before forming preambles.
diff --git a/src/asahi/compiler/agx_nir_lower_subgroups.c b/src/asahi/compiler/agx_nir_lower_subgroups.c
index 8ed85d90583..a59316cadc6 100644
--- a/src/asahi/compiler/agx_nir_lower_subgroups.c
+++ b/src/asahi/compiler/agx_nir_lower_subgroups.c
@@ -64,6 +64,12 @@ lower(nir_builder *b, nir_intrinsic_instr *intr, void *data)
       return true;
    }
 
+   case nir_intrinsic_last_invocation: {
+      nir_def *active_mask = nir_ballot(b, 1, 32, nir_imm_true(b));
+      nir_def_rewrite_uses(&intr->def, nir_ufind_msb(b, active_mask));
+      return true;
+   }
+
    case nir_intrinsic_vote_ieq:
    case nir_intrinsic_vote_feq: {
       /* The common lowering does: