diff --git a/src/amd/llvm/ac_nir_to_llvm.c b/src/amd/llvm/ac_nir_to_llvm.c
index 8ec5674d85f..6599b9ef2a6 100644
--- a/src/amd/llvm/ac_nir_to_llvm.c
+++ b/src/amd/llvm/ac_nir_to_llvm.c
@@ -3573,6 +3573,7 @@ static void visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
    case nir_intrinsic_load_tess_rel_patch_id_amd:
    case nir_intrinsic_load_patch_vertices_in:
    case nir_intrinsic_load_sample_mask_in:
+   case nir_intrinsic_load_viewport_xy_scale_and_offset:
    case nir_intrinsic_load_ring_tess_factors_amd:
    case nir_intrinsic_load_ring_tess_offchip_amd:
    case nir_intrinsic_load_ring_tess_offchip_offset_amd:
diff --git a/src/compiler/nir/nir_divergence_analysis.c b/src/compiler/nir/nir_divergence_analysis.c
index 084ceaa78de..0f6698eeca3 100644
--- a/src/compiler/nir/nir_divergence_analysis.c
+++ b/src/compiler/nir/nir_divergence_analysis.c
@@ -130,6 +130,7 @@ visit_intrinsic(nir_shader *shader, nir_intrinsic_instr *instr)
    case nir_intrinsic_load_viewport_x_offset:
    case nir_intrinsic_load_viewport_y_offset:
    case nir_intrinsic_load_viewport_z_offset:
+   case nir_intrinsic_load_viewport_xy_scale_and_offset:
    case nir_intrinsic_load_blend_const_color_a_float:
    case nir_intrinsic_load_blend_const_color_b_float:
    case nir_intrinsic_load_blend_const_color_g_float:
diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index aee5e5c9cfe..386773bcd01 100644
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -876,8 +876,6 @@ system_value("cull_mask", 1)
 #
 # Panfrost needs to implement all coordinate transformation in the
 # vertex shader; system values allow us to share this routine in NIR.
-#
-# RADV uses these for NGG primitive culling.
 system_value("viewport_x_scale", 1)
 system_value("viewport_y_scale", 1)
 system_value("viewport_z_scale", 1)
@@ -886,6 +884,8 @@ system_value("viewport_y_offset", 1)
 system_value("viewport_z_offset", 1)
 system_value("viewport_scale", 3)
 system_value("viewport_offset", 3)
+# Pack xy scale and offset into a vec4 load (used by AMD NGG primitive culling)
+system_value("viewport_xy_scale_and_offset", 4)
 
 # Blend constant color values.  Float values are clamped. Vectored versions are
 # provided as well for driver convenience