nir: Add quad vote intrinsics

Both Intel and AMD have special hardware support for these.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/218>

src/compiler/nir/nir_divergence_analysis.c

@@ -420,6 +420,8 @@ visit_intrinsic(nir_shader *shader, nir_intrinsic_instr *instr)
    case nir_intrinsic_quad_swap_horizontal:
    case nir_intrinsic_quad_swap_vertical:
    case nir_intrinsic_quad_swap_diagonal:
+   case nir_intrinsic_quad_vote_any:
+   case nir_intrinsic_quad_vote_all:
    case nir_intrinsic_load_deref:
    case nir_intrinsic_load_shared:
    case nir_intrinsic_load_shared2_amd:

src/compiler/nir/nir_gather_info.c

@@ -724,6 +724,8 @@ gather_intrinsic_info(nir_intrinsic_instr *instr, nir_shader *shader,
       }
       break;
 
+   case nir_intrinsic_quad_vote_any:
+   case nir_intrinsic_quad_vote_all:
    case nir_intrinsic_quad_broadcast:
    case nir_intrinsic_quad_swap_horizontal:
    case nir_intrinsic_quad_swap_vertical:

src/compiler/nir/nir_intrinsics.py

@@ -499,6 +499,12 @@ intrinsic("quad_swap_horizontal", src_comp=[0], dest_comp=0, bit_sizes=src0, fla
 intrinsic("quad_swap_vertical", src_comp=[0], dest_comp=0, bit_sizes=src0, flags=[CAN_ELIMINATE])
 intrinsic("quad_swap_diagonal", src_comp=[0], dest_comp=0, bit_sizes=src0, flags=[CAN_ELIMINATE])
 
+# Similar to vote_any and vote_all, but per-quad instead of per-wavefront.
+# Equivalent to subgroupOr(val, 4) and subgroupAnd(val, 4) assuming val is
+# boolean.
+intrinsic("quad_vote_any", src_comp=[1], dest_comp=1, flags=[CAN_ELIMINATE])
+intrinsic("quad_vote_all", src_comp=[1], dest_comp=1, flags=[CAN_ELIMINATE])
+
 # Rotate operation from SPIR-V: SpvOpGroupNonUniformRotateKHR.
 intrinsic("rotate", src_comp=[0, 1], dest_comp=0, bit_sizes=src0,
           indices=[EXECUTION_SCOPE, CLUSTER_SIZE], flags=[CAN_ELIMINATE]);