diff --git a/src/panfrost/bifrost/bi_pack.c b/src/panfrost/bifrost/bi_pack.c
index 204affa06f5..977363e4894 100644
--- a/src/panfrost/bifrost/bi_pack.c
+++ b/src/panfrost/bifrost/bi_pack.c
@@ -701,6 +701,10 @@ bi_pack_add_special(bi_clause *clause, bi_instruction *ins, bi_registers *regs)
                 return pan_pack_add_cube_ssel(clause, ins, regs);
         case BI_SPECIAL_CUBE_TSEL:
                 return pan_pack_add_cube_tsel(clause, ins, regs);
+        case BI_SPECIAL_CLPER_V6:
+                return pan_pack_add_clper_v6_i32(clause, ins, regs);
+        case BI_SPECIAL_CLPER_V7:
+                return pan_pack_add_clper_v7_i32(clause, ins, regs);
         default:
                 unreachable("Unknown special op");
         }
diff --git a/src/panfrost/bifrost/bi_print.c b/src/panfrost/bifrost/bi_print.c
index a97366bbba6..c00f2180b27 100644
--- a/src/panfrost/bifrost/bi_print.c
+++ b/src/panfrost/bifrost/bi_print.c
@@ -211,6 +211,8 @@ bi_special_op_name(enum bi_special_op op)
         case BI_SPECIAL_CUBEFACE2: return "cubeface2";
         case BI_SPECIAL_CUBE_SSEL: return "cube_ssel";
         case BI_SPECIAL_CUBE_TSEL: return "cube_tsel";
+        case BI_SPECIAL_CLPER_V6: return "clper_v6";
+        case BI_SPECIAL_CLPER_V7: return "clper_v7";
         default: return "invalid";
         }
 }
diff --git a/src/panfrost/bifrost/compiler.h b/src/panfrost/bifrost/compiler.h
index d51294cf2b9..22cc615ba2b 100644
--- a/src/panfrost/bifrost/compiler.h
+++ b/src/panfrost/bifrost/compiler.h
@@ -241,6 +241,10 @@ enum bi_special_op {
         BI_SPECIAL_CUBEFACE2,
         BI_SPECIAL_CUBE_SSEL,
         BI_SPECIAL_CUBE_TSEL,
+
+        /* Cross-lane permute, used to implement dFd{x,y} */
+        BI_SPECIAL_CLPER_V6,
+        BI_SPECIAL_CLPER_V7,
 };
 
 struct bi_bitwise {
@@ -259,6 +263,46 @@ struct bi_texture {
         bool compute_lod;
 };
 
+enum bi_clper_lane_op_mod {
+        BI_CLPER_LANE_OP_MOD_NONE,
+        BI_CLPER_LANE_OP_MOD_XOR,
+        BI_CLPER_LANE_OP_MOD_ACCUMULATE,
+        BI_CLPER_LANE_OP_MOD_SHIFT,
+};
+
+enum bi_subgroup_sz {
+        BI_CLPER_SUBGROUP_SZ_2,
+        BI_CLPER_SUBGROUP_SZ_4,
+        BI_CLPER_SUBGROUP_SZ_8,
+};
+
+enum bi_clper_inactive_res {
+        BI_CLPER_INACTIVE_RES_ZERO,
+        BI_CLPER_INACTIVE_RES_UMAX,
+        BI_CLPER_INACTIVE_RES_I1,
+        BI_CLPER_INACTIVE_RES_V2I1,
+        BI_CLPER_INACTIVE_RES_SMIN,
+        BI_CLPER_INACTIVE_RES_SMAX,
+        BI_CLPER_INACTIVE_RES_V2SMIN,
+        BI_CLPER_INACTIVE_RES_V2SMAX,
+        BI_CLPER_INACTIVE_RES_V4SMIN,
+        BI_CLPER_INACTIVE_RES_V4SMAX,
+        BI_CLPER_INACTIVE_RES_F1,
+        BI_CLPER_INACTIVE_RES_V2F1,
+        BI_CLPER_INACTIVE_RES_INFN,
+        BI_CLPER_INACTIVE_RES_INF,
+        BI_CLPER_INACTIVE_RES_V2INFN,
+        BI_CLPER_INACTIVE_RES_V2INF,
+};
+
+struct bi_special {
+        struct {
+                enum bi_clper_lane_op_mod lane_op_mod;
+                enum bi_clper_inactive_res inactive_res;
+        } clper;
+        enum bi_subgroup_sz subgroup_sz;
+};
+
 typedef struct {
         struct list_head link; /* Must be first */
         enum bi_class type;
@@ -355,6 +399,7 @@ typedef struct {
 
                 struct bi_bitwise bitwise;
                 struct bi_texture texture;
+                struct bi_special special;
         };
 } bi_instruction;
 
diff --git a/src/panfrost/bifrost/gen_pack.py b/src/panfrost/bifrost/gen_pack.py
index d9d31fd57d5..03c2f0401f1 100644
--- a/src/panfrost/bifrost/gen_pack.py
+++ b/src/panfrost/bifrost/gen_pack.py
@@ -284,6 +284,13 @@ modifier_map = {
         # For +LD_VAR, infer sample from load_vary.interp_mode
         "sample": lambda a,b,c,d: 'ins->load_vary.interp_mode',
 
+        # +CLPER
+        "lane_op": lambda a,b,c,d: 'ins->special.clper.lane_op_mod',
+        "inactive_result": lambda a,b,c,d: 'ins->special.clper.inactive_res',
+
+        # +CLPER and +WMASK
+        "subgroup": lambda a,b,c,d: 'ins->special.subgroup_sz',
+
         # We don't support these in the IR yet (TODO)
         "saturate": lambda a,b,c,d: '0', # clamp to min/max int
         "mask": lambda a,b,c,d: '0', # clz(~0) = ~0
@@ -304,9 +311,6 @@ modifier_map = {
         "bytes2": lambda a,b,c,d: '0', # NIR shifts are in bits
         "result_word": lambda a,b,c,d: '0', # 32-bit only shifts for now (TODO)
         "source": lambda a,b,c,d: '7', # cycle_counter for LD_GCLK
-        "lane_op": lambda a,b,c,d: '0', # CLPER none
-        "subgroup": lambda a,b,c,d: '1', # CLPER subgroup4
-        "inactive_result": lambda a,b,c,d: '0', # CLPER zero
         "threads": lambda a,b,c,d: '0', # IMULD odd
         "combine": lambda a,b,c,d: '0', # BRANCHC any
         "format": lambda a,b,c,d: '1', # LEA_TEX_IMM u32