diff --git a/docs/features.txt b/docs/features.txt
index 356f22cfa2a..81e8fad8b0c 100644
--- a/docs/features.txt
+++ b/docs/features.txt
@@ -492,7 +492,7 @@ Vulkan 1.3 -- all DONE: anv, radv, tu, lvp, vn
   VK_EXT_pipeline_creation_feedback                     DONE (anv, hasvk, lvp, radv, tu, v3dv, vn)
   VK_EXT_private_data                                   DONE (anv, hasvk, lvp, nvk, pvr, radv, tu, v3dv, vn)
   VK_EXT_image_robustness                               DONE (anv, hasvk, lvp, nvk, radv, tu, v3dv, vn)
-  VK_EXT_shader_demote_to_helper_invocation             DONE (anv, hasvk, lvp, nvk, radv, tu, vn)
+  VK_EXT_shader_demote_to_helper_invocation             DONE (anv, hasvk, lvp, nvk, radv, tu, v3dv, vn)
   VK_EXT_subgroup_size_control                          DONE (anv, hasvk, lvp, nvk, radv, tu, vn)
   VK_EXT_texel_buffer_alignment                         DONE (anv, hasvk, lvp, nvk, pvr, radv, tu, v3dv, vn)
   VK_EXT_texture_compression_astc_hdr                   DONE (vn)
diff --git a/src/broadcom/compiler/nir_to_vir.c b/src/broadcom/compiler/nir_to_vir.c
index 782d015592b..a4b904f9d80 100644
--- a/src/broadcom/compiler/nir_to_vir.c
+++ b/src/broadcom/compiler/nir_to_vir.c
@@ -649,7 +649,10 @@ ntq_emit_tmu_general(struct v3d_compile *c, nir_intrinsic_instr *instr,
                                 V3D_TMU_OP_TYPE_ATOMIC;
 
                         /* Only load per-quad if we can be certain that all
-                         * lines in the quad are active.
+                         * lines in the quad are active. Notice that demoted
+                         * invocations, unlike terminated ones, are still
+                         * active: we want to skip memory writes for them but
+                         * loads should still work.
                          */
                         uint32_t perquad =
                                 is_load && !vir_in_nonuniform_control_flow(c) &&
@@ -1908,6 +1911,7 @@ emit_frag_end(struct v3d_compile *c)
         if (c->output_position_index == -1 &&
             !(c->s->info.num_images || c->s->info.num_ssbos) &&
             !c->s->info.fs.uses_discard &&
+            !c->s->info.fs.uses_demote &&
             !c->fs_key->sample_alpha_to_coverage &&
             c->output_sample_mask_index == -1 &&
             has_any_tlb_color_write) {
@@ -3426,8 +3430,19 @@ ntq_emit_intrinsic(struct v3d_compile *c, nir_intrinsic_instr *instr)
                 ntq_emit_image_size(c, instr);
                 break;
 
+        /* FIXME: the Vulkan and SPIR-V specs specify that OpTerminate (which
+         * is intended to match the semantics of GLSL's discard) should
+         * terminate the invocation immediately. Our implementation doesn't
+         * do that. What we do is actually a demote by removing the invocations
+         * from the sample mask. Maybe we could be more strict and force an
+         * early termination by emitting a (maybe conditional) jump to the
+         * end section of the fragment shader for affected invocations.
+         */
         case nir_intrinsic_discard:
         case nir_intrinsic_terminate:
+                c->emitted_discard = true;
+                FALLTHROUGH;
+        case nir_intrinsic_demote:
                 ntq_flush_tmu(c);
 
                 if (vir_in_nonuniform_control_flow(c)) {
@@ -3440,11 +3455,13 @@ ntq_emit_intrinsic(struct v3d_compile *c, nir_intrinsic_instr *instr)
                         vir_SETMSF_dest(c, vir_nop_reg(),
                                         vir_uniform_ui(c, 0));
                 }
-                c->emitted_discard = true;
                 break;
 
         case nir_intrinsic_discard_if:
-        case nir_intrinsic_terminate_if: {
+        case nir_intrinsic_terminate_if:
+                c->emitted_discard = true;
+                FALLTHROUGH;
+        case nir_intrinsic_demote_if: {
                 ntq_flush_tmu(c);
 
                 enum v3d_qpu_cond cond = ntq_emit_bool_to_cond(c, instr->src[0]);
@@ -3462,7 +3479,6 @@ ntq_emit_intrinsic(struct v3d_compile *c, nir_intrinsic_instr *instr)
 
                 vir_set_cond(vir_SETMSF_dest(c, vir_nop_reg(),
                                              vir_uniform_ui(c, 0)), cond);
-                c->emitted_discard = true;
                 break;
         }
 
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index 48d08a9ee0c..09190db9b9b 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -730,6 +730,7 @@ v3d_lower_nir(struct v3d_compile *c)
                  nir_var_function_temp,
                  0,
                  glsl_get_natural_size_align_bytes);
+        NIR_PASS(_, c->s, nir_lower_is_helper_invocation);
         NIR_PASS(_, c->s, v3d_nir_lower_scratch);
         NIR_PASS(_, c->s, v3d_nir_lower_null_pointers);
 }
diff --git a/src/broadcom/vulkan/v3dv_device.c b/src/broadcom/vulkan/v3dv_device.c
index 1b37529fc21..4497942ac4c 100644
--- a/src/broadcom/vulkan/v3dv_device.c
+++ b/src/broadcom/vulkan/v3dv_device.c
@@ -205,6 +205,7 @@ get_device_extensions(const struct v3dv_physical_device *device,
       .EXT_private_data                     = true,
       .EXT_provoking_vertex                 = true,
       .EXT_separate_stencil_usage           = true,
+      .EXT_shader_demote_to_helper_invocation = true,
       .EXT_shader_module_identifier         = true,
       .EXT_texel_buffer_alignment           = true,
       .EXT_tooling_info                     = true,
@@ -444,6 +445,9 @@ get_features(const struct v3dv_physical_device *physical_device,
 
       /* VK_KHR_shader_terminate_invocation */
       .shaderTerminateInvocation = true,
+
+      /* VK_EXT_shader_demote_to_helper_invocation */
+      .shaderDemoteToHelperInvocation = true,
    };
 }
 
diff --git a/src/broadcom/vulkan/v3dv_pipeline.c b/src/broadcom/vulkan/v3dv_pipeline.c
index fe343c7268c..756d1d52e55 100644
--- a/src/broadcom/vulkan/v3dv_pipeline.c
+++ b/src/broadcom/vulkan/v3dv_pipeline.c
@@ -174,6 +174,7 @@ static const struct spirv_to_nir_options default_spirv_options =  {
       .physical_storage_buffer_address = true,
       .workgroup_memory_explicit_layout = true,
       .image_read_without_format = true,
+      .demote_to_helper_invocation = true,
     },
    .ubo_addr_format = nir_address_format_32bit_index_offset,
    .ssbo_addr_format = nir_address_format_32bit_index_offset,