diff --git a/src/compiler/nir/nir_divergence_analysis.c b/src/compiler/nir/nir_divergence_analysis.c
index a5b770d4737..e31553f7ed0 100644
--- a/src/compiler/nir/nir_divergence_analysis.c
+++ b/src/compiler/nir/nir_divergence_analysis.c
@@ -706,6 +706,7 @@ visit_intrinsic(nir_intrinsic_instr *instr, struct divergence_state *state)
    case nir_intrinsic_mbcnt_amd:
    case nir_intrinsic_lane_permute_16_amd:
    case nir_intrinsic_elect:
+   case nir_intrinsic_elect_any_ir3:
    case nir_intrinsic_load_tlb_color_brcm:
    case nir_intrinsic_load_tess_rel_patch_id_amd:
    case nir_intrinsic_load_gs_vertex_offset_amd:
diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index 0db57c52f5d..d34bb159527 100644
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -1372,6 +1372,10 @@ intrinsic("preamble_start_ir3", [], dest_comp=1, flags=[CAN_ELIMINATE, CAN_REORD
 
 barrier("preamble_end_ir3")
 
+# IR3-specific intrinsic to choose any invocation. This is implemented the same
+# as elect, except that it doesn't require helper invocations. Used by preambles.
+intrinsic("elect_any_ir3", dest_comp=1, flags=[CAN_ELIMINATE])
+
 # IR3-specific intrinsic for stc. Should be used in the shader preamble.
 store("uniform_ir3", [], indices=[BASE])
 
diff --git a/src/freedreno/ir3/ir3.h b/src/freedreno/ir3/ir3.h
index 33eab610218..5b85cdec668 100644
--- a/src/freedreno/ir3/ir3.h
+++ b/src/freedreno/ir3/ir3.h
@@ -1207,15 +1207,18 @@ uses_helpers(struct ir3_instruction *instr)
    case OPC_BALLOT_MACRO:
    case OPC_ANY_MACRO:
    case OPC_ALL_MACRO:
-   case OPC_ELECT_MACRO:
    case OPC_READ_FIRST_MACRO:
    case OPC_READ_COND_MACRO:
    case OPC_MOVMSK:
    case OPC_BRCST_ACTIVE:
       return true;
 
-   /* Catch lowered READ_FIRST/READ_COND. */
+   /* Catch lowered READ_FIRST/READ_COND. For elect, don't include the getone
+    * in the preamble because it doesn't actually matter which fiber is
+    * selected.
+    */
    case OPC_MOV:
+   case OPC_ELECT_MACRO:
       return instr->flags & IR3_INSTR_NEEDS_HELPERS;
 
    default:
diff --git a/src/freedreno/ir3/ir3_compiler_nir.c b/src/freedreno/ir3/ir3_compiler_nir.c
index a432507be57..fc2eb40253f 100644
--- a/src/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/freedreno/ir3/ir3_compiler_nir.c
@@ -2713,6 +2713,10 @@ emit_intrinsic(struct ir3_context *ctx, nir_intrinsic_instr *intr)
       break;
    }
    case nir_intrinsic_elect:
+      dst[0] = ir3_ELECT_MACRO(ctx->block);
+      dst[0]->flags |= IR3_INSTR_NEEDS_HELPERS;
+      break;
+   case nir_intrinsic_elect_any_ir3:
       dst[0] = ir3_ELECT_MACRO(ctx->block);
       break;
    case nir_intrinsic_preamble_start_ir3:
@@ -3987,6 +3991,7 @@ instr_can_be_predicated(nir_instr *instr)
       case nir_intrinsic_brcst_active_ir3:
       case nir_intrinsic_ballot:
       case nir_intrinsic_elect:
+      case nir_intrinsic_elect_any_ir3:
       case nir_intrinsic_read_invocation_cond_ir3:
       case nir_intrinsic_demote:
       case nir_intrinsic_demote_if:
@@ -4126,7 +4131,8 @@ emit_if(struct ir3_context *ctx, nir_if *nif)
       ir3_BALL(ctx->block, pred, IR3_REG_PREDICATE);
    } else if (condition->opc == OPC_ELECT_MACRO &&
               condition->block == ctx->block) {
-      ir3_GETONE(ctx->block);
+      struct ir3_instruction *branch = ir3_GETONE(ctx->block);
+      branch->flags |= condition->flags & IR3_INSTR_NEEDS_HELPERS;
    } else if (condition->opc == OPC_SHPS_MACRO &&
               condition->block == ctx->block) {
       /* TODO: technically this only works if the block is the only user of the
diff --git a/src/freedreno/ir3/ir3_legalize.c b/src/freedreno/ir3/ir3_legalize.c
index 4754cd7794e..3bb4cfa1232 100644
--- a/src/freedreno/ir3/ir3_legalize.c
+++ b/src/freedreno/ir3/ir3_legalize.c
@@ -1399,7 +1399,8 @@ helper_sched(struct ir3_legalize_ctx *ctx, struct ir3 *ir,
       struct ir3_instruction *terminator = ir3_block_get_terminator(block);
       if (terminator) {
          if (terminator->opc == OPC_BALL || terminator->opc == OPC_BANY ||
-             terminator->opc == OPC_GETONE) {
+             (terminator->opc == OPC_GETONE &&
+              (terminator->flags & IR3_INSTR_NEEDS_HELPERS))) {
             bd->uses_helpers_beginning = true;
             bd->uses_helpers_end = true;
             non_prefetch_helpers = true;
diff --git a/src/freedreno/ir3/ir3_lower_subgroups.c b/src/freedreno/ir3/ir3_lower_subgroups.c
index 224c459cd11..62d43769016 100644
--- a/src/freedreno/ir3/ir3_lower_subgroups.c
+++ b/src/freedreno/ir3/ir3_lower_subgroups.c
@@ -223,11 +223,12 @@ link_blocks_jump(struct ir3_block *pred, struct ir3_block *succ)
 
 static void
 link_blocks_branch(struct ir3_block *pred, struct ir3_block *target,
-                   struct ir3_block *fallthrough, unsigned opc,
+                   struct ir3_block *fallthrough, unsigned opc, unsigned flags,
                    struct ir3_instruction *condition)
 {
    unsigned nsrc = condition ? 1 : 0;
    struct ir3_instruction *branch = ir3_instr_create(pred, opc, 0, nsrc);
+   branch->flags |= flags;
 
    if (condition) {
       struct ir3_register *cond_dst = condition->dsts[0];
@@ -242,13 +243,14 @@ link_blocks_branch(struct ir3_block *pred, struct ir3_block *target,
 
 static struct ir3_block *
 create_if(struct ir3 *ir, struct ir3_block *before_block,
-          struct ir3_block *after_block, unsigned opc,
+          struct ir3_block *after_block, unsigned opc, unsigned flags,
           struct ir3_instruction *condition)
 {
    struct ir3_block *then_block = ir3_block_create(ir);
    list_add(&then_block->node, &before_block->node);
 
-   link_blocks_branch(before_block, then_block, after_block, opc, condition);
+   link_blocks_branch(before_block, then_block, after_block, opc, flags,
+                      condition);
    link_blocks_jump(then_block, after_block);
 
    return then_block;
@@ -320,7 +322,8 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
 
       link_blocks_jump(before_block, header);
 
-      link_blocks_branch(header, exit, footer, OPC_GETONE, NULL);
+      link_blocks_branch(header, exit, footer, OPC_GETONE,
+                         IR3_INSTR_NEEDS_HELPERS, NULL);
 
       link_blocks_jump(exit, after_block);
       ir3_block_link_physical(exit, footer);
@@ -369,9 +372,10 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
 
       link_blocks_jump(before_block, body);
 
-      link_blocks_branch(body, store, after_block, OPC_GETLAST, NULL);
+      link_blocks_branch(body, store, after_block, OPC_GETLAST, 0, NULL);
 
-      link_blocks_branch(store, after_block, body, OPC_GETONE, NULL);
+      link_blocks_branch(store, after_block, body, OPC_GETONE,
+                         IR3_INSTR_NEEDS_HELPERS, NULL);
 
       struct ir3_register *reduce = instr->dsts[0];
       struct ir3_register *inclusive = instr->dsts[1];
@@ -418,6 +422,7 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
 
       struct ir3_instruction *condition = NULL;
       unsigned branch_opc = 0;
+      unsigned branch_flags = 0;
 
       switch (instr->opc) {
       case OPC_BALLOT_MACRO:
@@ -445,13 +450,15 @@ lower_instr(struct ir3 *ir, struct ir3_block **block, struct ir3_instruction *in
       case OPC_ELECT_MACRO:
          after_block->reconvergence_point = true;
          branch_opc = OPC_GETONE;
+         branch_flags = instr->flags & IR3_INSTR_NEEDS_HELPERS;
          break;
       default:
          unreachable("bad opcode");
       }
 
       struct ir3_block *then_block =
-         create_if(ir, before_block, after_block, branch_opc, condition);
+         create_if(ir, before_block, after_block, branch_opc, branch_flags,
+                   condition);
 
       switch (instr->opc) {
       case OPC_ALL_MACRO: