pipeline: add register array to SWX
[dpdk.git] / lib / librte_pipeline / rte_swx_pipeline.c
index 419b676..3dd5784 100644 (file)
@@ -5,6 +5,7 @@
 #include <string.h>
 #include <stdio.h>
 #include <errno.h>
+#include <inttypes.h>
 #include <sys/queue.h>
 #include <arpa/inet.h>
 
@@ -22,7 +23,17 @@ do {                                                                           \
 } while (0)
 
 #define CHECK_NAME(name, err_code)                                             \
-       CHECK((name) && (name)[0], err_code)
+       CHECK((name) &&                                                        \
+             (name)[0] &&                                                     \
+             (strnlen((name), RTE_SWX_NAME_SIZE) < RTE_SWX_NAME_SIZE),        \
+             err_code)
+
+#define CHECK_INSTRUCTION(instr, err_code)                                     \
+       CHECK((instr) &&                                                       \
+             (instr)[0] &&                                                    \
+             (strnlen((instr), RTE_SWX_INSTRUCTION_SIZE) <                    \
+              RTE_SWX_INSTRUCTION_SIZE),                                      \
+             err_code)
 
 #ifndef TRACE_LEVEL
 #define TRACE_LEVEL 0
@@ -34,9 +45,59 @@ do {                                                                           \
 #define TRACE(...)
 #endif
 
+/*
+ * Environment.
+ */
 #define ntoh64(x) rte_be_to_cpu_64(x)
 #define hton64(x) rte_cpu_to_be_64(x)
 
+#ifndef RTE_SWX_PIPELINE_HUGE_PAGES_DISABLE
+
+#include <rte_malloc.h>
+
+static void *
+env_malloc(size_t size, size_t alignment, int numa_node)
+{
+       return rte_zmalloc_socket(NULL, size, alignment, numa_node);
+}
+
+static void
+env_free(void *start, size_t size __rte_unused)
+{
+       rte_free(start);
+}
+
+#else
+
+#include <numa.h>
+
+static void *
+env_malloc(size_t size, size_t alignment __rte_unused, int numa_node)
+{
+       void *start;
+
+       if (numa_available() == -1)
+               return NULL;
+
+       start = numa_alloc_onnode(size, numa_node);
+       if (!start)
+               return NULL;
+
+       memset(start, 0, size);
+       return start;
+}
+
+static void
+env_free(void *start, size_t size)
+{
+       if (numa_available() == -1)
+               return;
+
+       numa_free(start, size);
+}
+
+#endif
+
 /*
  * Struct.
  */
@@ -338,6 +399,151 @@ enum instruction_type {
        INSTR_ALU_SHL_HH, /* dst = H, src = H */
        INSTR_ALU_SHL_MI, /* dst = MEF, src = I */
        INSTR_ALU_SHL_HI, /* dst = H, src = I */
+
+       /* shr dst src
+        * dst >>= src
+        * dst = HMEF, src = HMEFTI
+        */
+       INSTR_ALU_SHR,    /* dst = MEF, src = MEF */
+       INSTR_ALU_SHR_MH, /* dst = MEF, src = H */
+       INSTR_ALU_SHR_HM, /* dst = H, src = MEF */
+       INSTR_ALU_SHR_HH, /* dst = H, src = H */
+       INSTR_ALU_SHR_MI, /* dst = MEF, src = I */
+       INSTR_ALU_SHR_HI, /* dst = H, src = I */
+
+       /* regprefetch REGARRAY index
+        * prefetch REGARRAY[index]
+        * index = HMEFTI
+        */
+       INSTR_REGPREFETCH_RH, /* index = H */
+       INSTR_REGPREFETCH_RM, /* index = MEFT */
+       INSTR_REGPREFETCH_RI, /* index = I */
+
+       /* regrd dst REGARRAY index
+        * dst = REGARRAY[index]
+        * dst = HMEF, index = HMEFTI
+        */
+       INSTR_REGRD_HRH, /* dst = H, index = H */
+       INSTR_REGRD_HRM, /* dst = H, index = MEFT */
+       INSTR_REGRD_HRI, /* dst = H, index = I */
+       INSTR_REGRD_MRH, /* dst = MEF, index = H */
+       INSTR_REGRD_MRM, /* dst = MEF, index = MEFT */
+       INSTR_REGRD_MRI, /* dst = MEF, index = I */
+
+       /* regwr REGARRAY index src
+        * REGARRAY[index] = src
+        * index = HMEFTI, src = HMEFTI
+        */
+       INSTR_REGWR_RHH, /* index = H, src = H */
+       INSTR_REGWR_RHM, /* index = H, src = MEFT */
+       INSTR_REGWR_RHI, /* index = H, src = I */
+       INSTR_REGWR_RMH, /* index = MEFT, src = H */
+       INSTR_REGWR_RMM, /* index = MEFT, src = MEFT */
+       INSTR_REGWR_RMI, /* index = MEFT, src = I */
+       INSTR_REGWR_RIH, /* index = I, src = H */
+       INSTR_REGWR_RIM, /* index = I, src = MEFT */
+       INSTR_REGWR_RII, /* index = I, src = I */
+
+       /* regadd REGARRAY index src
+        * REGARRAY[index] += src
+        * index = HMEFTI, src = HMEFTI
+        */
+       INSTR_REGADD_RHH, /* index = H, src = H */
+       INSTR_REGADD_RHM, /* index = H, src = MEFT */
+       INSTR_REGADD_RHI, /* index = H, src = I */
+       INSTR_REGADD_RMH, /* index = MEFT, src = H */
+       INSTR_REGADD_RMM, /* index = MEFT, src = MEFT */
+       INSTR_REGADD_RMI, /* index = MEFT, src = I */
+       INSTR_REGADD_RIH, /* index = I, src = H */
+       INSTR_REGADD_RIM, /* index = I, src = MEFT */
+       INSTR_REGADD_RII, /* index = I, src = I */
+
+       /* table TABLE */
+       INSTR_TABLE,
+
+       /* extern e.obj.func */
+       INSTR_EXTERN_OBJ,
+
+       /* extern f.func */
+       INSTR_EXTERN_FUNC,
+
+       /* jmp LABEL
+        * Unconditional jump
+        */
+       INSTR_JMP,
+
+       /* jmpv LABEL h.header
+        * Jump if header is valid
+        */
+       INSTR_JMP_VALID,
+
+       /* jmpnv LABEL h.header
+        * Jump if header is invalid
+        */
+       INSTR_JMP_INVALID,
+
+       /* jmph LABEL
+        * Jump if table lookup hit
+        */
+       INSTR_JMP_HIT,
+
+       /* jmpnh LABEL
+        * Jump if table lookup miss
+        */
+       INSTR_JMP_MISS,
+
+       /* jmpa LABEL ACTION
+        * Jump if action run
+        */
+       INSTR_JMP_ACTION_HIT,
+
+       /* jmpna LABEL ACTION
+        * Jump if action not run
+        */
+       INSTR_JMP_ACTION_MISS,
+
+       /* jmpeq LABEL a b
+        * Jump is a is equal to b
+        * a = HMEFT, b = HMEFTI
+        */
+       INSTR_JMP_EQ,   /* (a, b) = (MEFT, MEFT) or (a, b) = (H, H) */
+       INSTR_JMP_EQ_S, /* (a, b) = (MEFT, H) or (a, b) = (H, MEFT) */
+       INSTR_JMP_EQ_I, /* (a, b) = (MEFT, I) or (a, b) = (H, I) */
+
+       /* jmpneq LABEL a b
+        * Jump is a is not equal to b
+        * a = HMEFT, b = HMEFTI
+        */
+       INSTR_JMP_NEQ,   /* (a, b) = (MEFT, MEFT) or (a, b) = (H, H) */
+       INSTR_JMP_NEQ_S, /* (a, b) = (MEFT, H) or (a, b) = (H, MEFT) */
+       INSTR_JMP_NEQ_I, /* (a, b) = (MEFT, I) or (a, b) = (H, I) */
+
+       /* jmplt LABEL a b
+        * Jump if a is less than b
+        * a = HMEFT, b = HMEFTI
+        */
+       INSTR_JMP_LT,    /* a = MEF, b = MEF */
+       INSTR_JMP_LT_MH, /* a = MEF, b = H */
+       INSTR_JMP_LT_HM, /* a = H, b = MEF */
+       INSTR_JMP_LT_HH, /* a = H, b = H */
+       INSTR_JMP_LT_MI, /* a = MEF, b = I */
+       INSTR_JMP_LT_HI, /* a = H, b = I */
+
+       /* jmpgt LABEL a b
+        * Jump if a is greater than b
+        * a = HMEFT, b = HMEFTI
+        */
+       INSTR_JMP_GT,    /* a = MEF, b = MEF */
+       INSTR_JMP_GT_MH, /* a = MEF, b = H */
+       INSTR_JMP_GT_HM, /* a = H, b = MEF */
+       INSTR_JMP_GT_HH, /* a = H, b = H */
+       INSTR_JMP_GT_MI, /* a = MEF, b = I */
+       INSTR_JMP_GT_HI, /* a = H, b = I */
+
+       /* return
+        * Return from action
+        */
+       INSTR_RETURN,
 };
 
 struct instr_operand {
@@ -365,11 +571,39 @@ struct instr_hdr_validity {
        uint8_t header_id;
 };
 
+struct instr_table {
+       uint8_t table_id;
+};
+
+struct instr_extern_obj {
+       uint8_t ext_obj_id;
+       uint8_t func_id;
+};
+
+struct instr_extern_func {
+       uint8_t ext_func_id;
+};
+
 struct instr_dst_src {
        struct instr_operand dst;
        union {
                struct instr_operand src;
-               uint32_t src_val;
+               uint64_t src_val;
+       };
+};
+
+struct instr_regarray {
+       uint8_t regarray_id;
+       uint8_t pad[3];
+
+       union {
+               struct instr_operand idx;
+               uint32_t idx_val;
+       };
+
+       union {
+               struct instr_operand dstsrc;
+               uint64_t dstsrc_val;
        };
 };
 
@@ -386,14 +620,34 @@ struct instr_dma {
        uint16_t n_bytes[8];
 };
 
+struct instr_jmp {
+       struct instruction *ip;
+
+       union {
+               struct instr_operand a;
+               uint8_t header_id;
+               uint8_t action_id;
+       };
+
+       union {
+               struct instr_operand b;
+               uint64_t b_val;
+       };
+};
+
 struct instruction {
        enum instruction_type type;
        union {
                struct instr_io io;
                struct instr_hdr_validity valid;
                struct instr_dst_src mov;
+               struct instr_regarray regarray;
                struct instr_dma dma;
                struct instr_dst_src alu;
+               struct instr_table table;
+               struct instr_extern_obj ext_obj;
+               struct instr_extern_func ext_func;
+               struct instr_jmp jmp;
        };
 };
 
@@ -467,6 +721,24 @@ struct table_runtime {
        uint8_t **key;
 };
 
+/*
+ * Register array.
+ */
+struct regarray {
+       TAILQ_ENTRY(regarray) node;
+       char name[RTE_SWX_NAME_SIZE];
+       uint64_t init_val;
+       uint32_t size;
+       uint32_t id;
+};
+
+TAILQ_HEAD(regarray_tailq, regarray);
+
+struct regarray_runtime {
+       uint64_t *regarray;
+       uint32_t size_mask;
+};
+
 /*
  * Pipeline.
  */
@@ -508,6 +780,9 @@ struct thread {
 #define MASK64_BIT_SET(mask, pos) ((mask) | (1LLU << (pos)))
 #define MASK64_BIT_CLR(mask, pos) ((mask) & ~(1LLU << (pos)))
 
+#define HEADER_VALID(thread, header_id) \
+       MASK64_BIT_GET((thread)->valid_headers, header_id)
+
 #define ALU(thread, ip, operator)  \
 {                                                                              \
        uint8_t *dst_struct = (thread)->structs[(ip)->alu.dst.struct_id];      \
@@ -689,6 +964,118 @@ struct thread {
        *dst64_ptr = (dst64 & ~dst64_mask) | (src & dst64_mask);               \
 }
 
+#define JMP_CMP(thread, ip, operator)  \
+{                                                                              \
+       uint8_t *a_struct = (thread)->structs[(ip)->jmp.a.struct_id];          \
+       uint64_t *a64_ptr = (uint64_t *)&a_struct[(ip)->jmp.a.offset];         \
+       uint64_t a64 = *a64_ptr;                                               \
+       uint64_t a64_mask = UINT64_MAX >> (64 - (ip)->jmp.a.n_bits);           \
+       uint64_t a = a64 & a64_mask;                                           \
+                                                                              \
+       uint8_t *b_struct = (thread)->structs[(ip)->jmp.b.struct_id];          \
+       uint64_t *b64_ptr = (uint64_t *)&b_struct[(ip)->jmp.b.offset];         \
+       uint64_t b64 = *b64_ptr;                                               \
+       uint64_t b64_mask = UINT64_MAX >> (64 - (ip)->jmp.b.n_bits);           \
+       uint64_t b = b64 & b64_mask;                                           \
+                                                                              \
+       (thread)->ip = (a operator b) ? (ip)->jmp.ip : ((thread)->ip + 1);     \
+}
+
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+
+#define JMP_CMP_S(thread, ip, operator)  \
+{                                                                              \
+       uint8_t *a_struct = (thread)->structs[(ip)->jmp.a.struct_id];          \
+       uint64_t *a64_ptr = (uint64_t *)&a_struct[(ip)->jmp.a.offset];         \
+       uint64_t a64 = *a64_ptr;                                               \
+       uint64_t a64_mask = UINT64_MAX >> (64 - (ip)->jmp.a.n_bits);           \
+       uint64_t a = a64 & a64_mask;                                           \
+                                                                              \
+       uint8_t *b_struct = (thread)->structs[(ip)->jmp.b.struct_id];          \
+       uint64_t *b64_ptr = (uint64_t *)&b_struct[(ip)->jmp.b.offset];         \
+       uint64_t b64 = *b64_ptr;                                               \
+       uint64_t b = ntoh64(b64) >> (64 - (ip)->jmp.b.n_bits);                 \
+                                                                              \
+       (thread)->ip = (a operator b) ? (ip)->jmp.ip : ((thread)->ip + 1);     \
+}
+
+#define JMP_CMP_MH JMP_CMP_S
+
+#define JMP_CMP_HM(thread, ip, operator)  \
+{                                                                              \
+       uint8_t *a_struct = (thread)->structs[(ip)->jmp.a.struct_id];          \
+       uint64_t *a64_ptr = (uint64_t *)&a_struct[(ip)->jmp.a.offset];         \
+       uint64_t a64 = *a64_ptr;                                               \
+       uint64_t a = ntoh64(a64) >> (64 - (ip)->jmp.a.n_bits);                 \
+                                                                              \
+       uint8_t *b_struct = (thread)->structs[(ip)->jmp.b.struct_id];          \
+       uint64_t *b64_ptr = (uint64_t *)&b_struct[(ip)->jmp.b.offset];         \
+       uint64_t b64 = *b64_ptr;                                               \
+       uint64_t b64_mask = UINT64_MAX >> (64 - (ip)->jmp.b.n_bits);           \
+       uint64_t b = b64 & b64_mask;                                           \
+                                                                              \
+       (thread)->ip = (a operator b) ? (ip)->jmp.ip : ((thread)->ip + 1);     \
+}
+
+#define JMP_CMP_HH(thread, ip, operator)  \
+{                                                                              \
+       uint8_t *a_struct = (thread)->structs[(ip)->jmp.a.struct_id];          \
+       uint64_t *a64_ptr = (uint64_t *)&a_struct[(ip)->jmp.a.offset];         \
+       uint64_t a64 = *a64_ptr;                                               \
+       uint64_t a = ntoh64(a64) >> (64 - (ip)->jmp.a.n_bits);                 \
+                                                                              \
+       uint8_t *b_struct = (thread)->structs[(ip)->jmp.b.struct_id];          \
+       uint64_t *b64_ptr = (uint64_t *)&b_struct[(ip)->jmp.b.offset];         \
+       uint64_t b64 = *b64_ptr;                                               \
+       uint64_t b = ntoh64(b64) >> (64 - (ip)->jmp.b.n_bits);                 \
+                                                                              \
+       (thread)->ip = (a operator b) ? (ip)->jmp.ip : ((thread)->ip + 1);     \
+}
+
+#else
+
+#define JMP_CMP_S JMP_CMP
+#define JMP_CMP_MH JMP_CMP
+#define JMP_CMP_HM JMP_CMP
+#define JMP_CMP_HH JMP_CMP
+
+#endif
+
+#define JMP_CMP_I(thread, ip, operator)  \
+{                                                                              \
+       uint8_t *a_struct = (thread)->structs[(ip)->jmp.a.struct_id];          \
+       uint64_t *a64_ptr = (uint64_t *)&a_struct[(ip)->jmp.a.offset];         \
+       uint64_t a64 = *a64_ptr;                                               \
+       uint64_t a64_mask = UINT64_MAX >> (64 - (ip)->jmp.a.n_bits);           \
+       uint64_t a = a64 & a64_mask;                                           \
+                                                                              \
+       uint64_t b = (ip)->jmp.b_val;                                          \
+                                                                              \
+       (thread)->ip = (a operator b) ? (ip)->jmp.ip : ((thread)->ip + 1);     \
+}
+
+#define JMP_CMP_MI JMP_CMP_I
+
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+
+#define JMP_CMP_HI(thread, ip, operator)  \
+{                                                                              \
+       uint8_t *a_struct = (thread)->structs[(ip)->jmp.a.struct_id];          \
+       uint64_t *a64_ptr = (uint64_t *)&a_struct[(ip)->jmp.a.offset];         \
+       uint64_t a64 = *a64_ptr;                                               \
+       uint64_t a = ntoh64(a64) >> (64 - (ip)->jmp.a.n_bits);                 \
+                                                                              \
+       uint64_t b = (ip)->jmp.b_val;                                          \
+                                                                              \
+       (thread)->ip = (a operator b) ? (ip)->jmp.ip : ((thread)->ip + 1);     \
+}
+
+#else
+
+#define JMP_CMP_HI JMP_CMP_I
+
+#endif
+
 #define METADATA_READ(thread, offset, n_bits)                                  \
 ({                                                                             \
        uint64_t *m64_ptr = (uint64_t *)&(thread)->metadata[offset];           \
@@ -727,11 +1114,13 @@ struct rte_swx_pipeline {
        struct action_tailq actions;
        struct table_type_tailq table_types;
        struct table_tailq tables;
+       struct regarray_tailq regarrays;
 
        struct port_in_runtime *in;
        struct port_out_runtime *out;
        struct instruction **action_instructions;
        struct rte_swx_table_state *table_state;
+       struct regarray_runtime *regarray_runtime;
        struct instruction *instructions;
        struct thread threads[RTE_SWX_PIPELINE_THREADS_MAX];
 
@@ -742,6 +1131,7 @@ struct rte_swx_pipeline {
        uint32_t n_extern_funcs;
        uint32_t n_actions;
        uint32_t n_tables;
+       uint32_t n_regarrays;
        uint32_t n_headers;
        uint32_t thread_id;
        uint32_t port_id;
@@ -1252,6 +1642,50 @@ extern_obj_find(struct rte_swx_pipeline *p, const char *name)
        return NULL;
 }
 
+static struct extern_type_member_func *
+extern_obj_member_func_parse(struct rte_swx_pipeline *p,
+                            const char *name,
+                            struct extern_obj **obj)
+{
+       struct extern_obj *object;
+       struct extern_type_member_func *func;
+       char *object_name, *func_name;
+
+       if (name[0] != 'e' || name[1] != '.')
+               return NULL;
+
+       object_name = strdup(&name[2]);
+       if (!object_name)
+               return NULL;
+
+       func_name = strchr(object_name, '.');
+       if (!func_name) {
+               free(object_name);
+               return NULL;
+       }
+
+       *func_name = 0;
+       func_name++;
+
+       object = extern_obj_find(p, object_name);
+       if (!object) {
+               free(object_name);
+               return NULL;
+       }
+
+       func = extern_type_member_func_find(object->type, func_name);
+       if (!func) {
+               free(object_name);
+               return NULL;
+       }
+
+       if (obj)
+               *obj = object;
+
+       free(object_name);
+       return func;
+}
+
 static struct field *
 extern_obj_mailbox_field_parse(struct rte_swx_pipeline *p,
                               const char *name,
@@ -1346,12 +1780,12 @@ rte_swx_pipeline_extern_type_member_func_register(struct rte_swx_pipeline *p,
 
        CHECK(p, EINVAL);
 
-       CHECK(extern_type_name, EINVAL);
+       CHECK_NAME(extern_type_name, EINVAL);
        type = extern_type_find(p, extern_type_name);
        CHECK(type, EINVAL);
        CHECK(type->n_funcs < RTE_SWX_EXTERN_TYPE_MEMBER_FUNCS_MAX, ENOSPC);
 
-       CHECK(name, EINVAL);
+       CHECK_NAME(name, EINVAL);
        CHECK(!extern_type_member_func_find(type, name), EEXIST);
 
        CHECK(member_func, EINVAL);
@@ -1534,6 +1968,16 @@ extern_func_find(struct rte_swx_pipeline *p, const char *name)
        return NULL;
 }
 
+static struct extern_func *
+extern_func_parse(struct rte_swx_pipeline *p,
+                 const char *name)
+{
+       if (name[0] != 'f' || name[1] != '.')
+               return NULL;
+
+       return extern_func_find(p, &name[2]);
+}
+
 static struct field *
 extern_func_mailbox_field_parse(struct rte_swx_pipeline *p,
                                const char *name,
@@ -1958,6 +2402,42 @@ metadata_free(struct rte_swx_pipeline *p)
 /*
  * Instruction.
  */
+static int
+instruction_is_jmp(struct instruction *instr)
+{
+       switch (instr->type) {
+       case INSTR_JMP:
+       case INSTR_JMP_VALID:
+       case INSTR_JMP_INVALID:
+       case INSTR_JMP_HIT:
+       case INSTR_JMP_MISS:
+       case INSTR_JMP_ACTION_HIT:
+       case INSTR_JMP_ACTION_MISS:
+       case INSTR_JMP_EQ:
+       case INSTR_JMP_EQ_S:
+       case INSTR_JMP_EQ_I:
+       case INSTR_JMP_NEQ:
+       case INSTR_JMP_NEQ_S:
+       case INSTR_JMP_NEQ_I:
+       case INSTR_JMP_LT:
+       case INSTR_JMP_LT_MH:
+       case INSTR_JMP_LT_HM:
+       case INSTR_JMP_LT_HH:
+       case INSTR_JMP_LT_MI:
+       case INSTR_JMP_LT_HI:
+       case INSTR_JMP_GT:
+       case INSTR_JMP_GT_MH:
+       case INSTR_JMP_GT_HM:
+       case INSTR_JMP_GT_HH:
+       case INSTR_JMP_GT_MI:
+       case INSTR_JMP_GT_HI:
+               return 1;
+
+       default:
+               return 0;
+       }
+}
+
 static struct field *
 action_field_parse(struct action *action, const char *name);
 
@@ -2046,6 +2526,21 @@ thread_ip_reset(struct rte_swx_pipeline *p, struct thread *t)
        t->ip = p->instructions;
 }
 
+static inline void
+thread_ip_set(struct thread *t, struct instruction *ip)
+{
+       t->ip = ip;
+}
+
+static inline void
+thread_ip_action_call(struct rte_swx_pipeline *p,
+                     struct thread *t,
+                     uint32_t action_id)
+{
+       t->ret = t->ip + 1;
+       t->ip = p->action_instructions[action_id];
+}
+
 static inline void
 thread_ip_inc(struct rte_swx_pipeline *p);
 
@@ -2069,6 +2564,12 @@ thread_yield(struct rte_swx_pipeline *p)
        p->thread_id = (p->thread_id + 1) & (RTE_SWX_PIPELINE_THREADS_MAX - 1);
 }
 
+static inline void
+thread_yield_cond(struct rte_swx_pipeline *p, int cond)
+{
+       p->thread_id = (p->thread_id + cond) & (RTE_SWX_PIPELINE_THREADS_MAX - 1);
+}
+
 /*
  * rx.
  */
@@ -2660,80 +3161,242 @@ instr_hdr_invalidate_exec(struct rte_swx_pipeline *p)
 }
 
 /*
- * mov.
+ * table.
  */
+static struct table *
+table_find(struct rte_swx_pipeline *p, const char *name);
+
 static int
-instr_mov_translate(struct rte_swx_pipeline *p,
-                   struct action *action,
-                   char **tokens,
-                   int n_tokens,
-                   struct instruction *instr,
-                   struct instruction_data *data __rte_unused)
+instr_table_translate(struct rte_swx_pipeline *p,
+                     struct action *action,
+                     char **tokens,
+                     int n_tokens,
+                     struct instruction *instr,
+                     struct instruction_data *data __rte_unused)
 {
-       char *dst = tokens[1], *src = tokens[2];
-       struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
-
-       CHECK(n_tokens == 3, EINVAL);
-
-       fdst = struct_field_parse(p, NULL, dst, &dst_struct_id);
-       CHECK(fdst, EINVAL);
-
-       /* MOV or MOV_S. */
-       fsrc = struct_field_parse(p, action, src, &src_struct_id);
-       if (fsrc) {
-               instr->type = INSTR_MOV;
-               if ((dst[0] == 'h' && src[0] != 'h') ||
-                   (dst[0] != 'h' && src[0] == 'h'))
-                       instr->type = INSTR_MOV_S;
-
-               instr->mov.dst.struct_id = (uint8_t)dst_struct_id;
-               instr->mov.dst.n_bits = fdst->n_bits;
-               instr->mov.dst.offset = fdst->offset / 8;
-               instr->mov.src.struct_id = (uint8_t)src_struct_id;
-               instr->mov.src.n_bits = fsrc->n_bits;
-               instr->mov.src.offset = fsrc->offset / 8;
-               return 0;
-       }
+       struct table *t;
 
-       /* MOV_I. */
-       src_val = strtoul(src, &src, 0);
-       CHECK(!src[0], EINVAL);
+       CHECK(!action, EINVAL);
+       CHECK(n_tokens == 2, EINVAL);
 
-       if (dst[0] == 'h')
-               src_val = htonl(src_val);
+       t = table_find(p, tokens[1]);
+       CHECK(t, EINVAL);
 
-       instr->type = INSTR_MOV_I;
-       instr->mov.dst.struct_id = (uint8_t)dst_struct_id;
-       instr->mov.dst.n_bits = fdst->n_bits;
-       instr->mov.dst.offset = fdst->offset / 8;
-       instr->mov.src_val = (uint32_t)src_val;
+       instr->type = INSTR_TABLE;
+       instr->table.table_id = t->id;
        return 0;
 }
 
 static inline void
-instr_mov_exec(struct rte_swx_pipeline *p)
+instr_table_exec(struct rte_swx_pipeline *p)
 {
        struct thread *t = &p->threads[p->thread_id];
        struct instruction *ip = t->ip;
+       uint32_t table_id = ip->table.table_id;
+       struct rte_swx_table_state *ts = &t->table_state[table_id];
+       struct table_runtime *table = &t->tables[table_id];
+       uint64_t action_id;
+       uint8_t *action_data;
+       int done, hit;
+
+       /* Table. */
+       done = table->func(ts->obj,
+                          table->mailbox,
+                          table->key,
+                          &action_id,
+                          &action_data,
+                          &hit);
+       if (!done) {
+               /* Thread. */
+               TRACE("[Thread %2u] table %u (not finalized)\n",
+                     p->thread_id,
+                     table_id);
 
-       TRACE("[Thread %2u] mov\n",
-             p->thread_id);
+               thread_yield(p);
+               return;
+       }
 
-       MOV(t, ip);
+       action_id = hit ? action_id : ts->default_action_id;
+       action_data = hit ? action_data : ts->default_action_data;
+
+       TRACE("[Thread %2u] table %u (%s, action %u)\n",
+             p->thread_id,
+             table_id,
+             hit ? "hit" : "miss",
+             (uint32_t)action_id);
+
+       t->action_id = action_id;
+       t->structs[0] = action_data;
+       t->hit = hit;
 
        /* Thread. */
-       thread_ip_inc(p);
+       thread_ip_action_call(p, t, action_id);
 }
 
-static inline void
-instr_mov_s_exec(struct rte_swx_pipeline *p)
+/*
+ * extern.
+ */
+static int
+instr_extern_translate(struct rte_swx_pipeline *p,
+                      struct action *action __rte_unused,
+                      char **tokens,
+                      int n_tokens,
+                      struct instruction *instr,
+                      struct instruction_data *data __rte_unused)
 {
-       struct thread *t = &p->threads[p->thread_id];
-       struct instruction *ip = t->ip;
+       char *token = tokens[1];
 
-       TRACE("[Thread %2u] mov (s)\n",
-             p->thread_id);
+       CHECK(n_tokens == 2, EINVAL);
+
+       if (token[0] == 'e') {
+               struct extern_obj *obj;
+               struct extern_type_member_func *func;
+
+               func = extern_obj_member_func_parse(p, token, &obj);
+               CHECK(func, EINVAL);
+
+               instr->type = INSTR_EXTERN_OBJ;
+               instr->ext_obj.ext_obj_id = obj->id;
+               instr->ext_obj.func_id = func->id;
+
+               return 0;
+       }
+
+       if (token[0] == 'f') {
+               struct extern_func *func;
+
+               func = extern_func_parse(p, token);
+               CHECK(func, EINVAL);
+
+               instr->type = INSTR_EXTERN_FUNC;
+               instr->ext_func.ext_func_id = func->id;
+
+               return 0;
+       }
+
+       CHECK(0, EINVAL);
+}
+
+static inline void
+instr_extern_obj_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint32_t obj_id = ip->ext_obj.ext_obj_id;
+       uint32_t func_id = ip->ext_obj.func_id;
+       struct extern_obj_runtime *obj = &t->extern_objs[obj_id];
+       rte_swx_extern_type_member_func_t func = obj->funcs[func_id];
+
+       TRACE("[Thread %2u] extern obj %u member func %u\n",
+             p->thread_id,
+             obj_id,
+             func_id);
+
+       /* Extern object member function execute. */
+       uint32_t done = func(obj->obj, obj->mailbox);
+
+       /* Thread. */
+       thread_ip_inc_cond(t, done);
+       thread_yield_cond(p, done ^ 1);
+}
+
+static inline void
+instr_extern_func_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint32_t ext_func_id = ip->ext_func.ext_func_id;
+       struct extern_func_runtime *ext_func = &t->extern_funcs[ext_func_id];
+       rte_swx_extern_func_t func = ext_func->func;
+
+       TRACE("[Thread %2u] extern func %u\n",
+             p->thread_id,
+             ext_func_id);
+
+       /* Extern function execute. */
+       uint32_t done = func(ext_func->mailbox);
+
+       /* Thread. */
+       thread_ip_inc_cond(t, done);
+       thread_yield_cond(p, done ^ 1);
+}
+
+/*
+ * mov.
+ */
+static int
+instr_mov_translate(struct rte_swx_pipeline *p,
+                   struct action *action,
+                   char **tokens,
+                   int n_tokens,
+                   struct instruction *instr,
+                   struct instruction_data *data __rte_unused)
+{
+       char *dst = tokens[1], *src = tokens[2];
+       struct field *fdst, *fsrc;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
+
+       CHECK(n_tokens == 3, EINVAL);
+
+       fdst = struct_field_parse(p, NULL, dst, &dst_struct_id);
+       CHECK(fdst, EINVAL);
+
+       /* MOV or MOV_S. */
+       fsrc = struct_field_parse(p, action, src, &src_struct_id);
+       if (fsrc) {
+               instr->type = INSTR_MOV;
+               if ((dst[0] == 'h' && src[0] != 'h') ||
+                   (dst[0] != 'h' && src[0] == 'h'))
+                       instr->type = INSTR_MOV_S;
+
+               instr->mov.dst.struct_id = (uint8_t)dst_struct_id;
+               instr->mov.dst.n_bits = fdst->n_bits;
+               instr->mov.dst.offset = fdst->offset / 8;
+               instr->mov.src.struct_id = (uint8_t)src_struct_id;
+               instr->mov.src.n_bits = fsrc->n_bits;
+               instr->mov.src.offset = fsrc->offset / 8;
+               return 0;
+       }
+
+       /* MOV_I. */
+       src_val = strtoull(src, &src, 0);
+       CHECK(!src[0], EINVAL);
+
+       if (dst[0] == 'h')
+               src_val = hton64(src_val) >> (64 - fdst->n_bits);
+
+       instr->type = INSTR_MOV_I;
+       instr->mov.dst.struct_id = (uint8_t)dst_struct_id;
+       instr->mov.dst.n_bits = fdst->n_bits;
+       instr->mov.dst.offset = fdst->offset / 8;
+       instr->mov.src_val = src_val;
+       return 0;
+}
+
+static inline void
+instr_mov_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] mov\n",
+             p->thread_id);
+
+       MOV(t, ip);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_mov_s_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] mov (s)\n",
+             p->thread_id);
 
        MOV_S(t, ip);
 
@@ -2747,7 +3410,7 @@ instr_mov_i_exec(struct rte_swx_pipeline *p)
        struct thread *t = &p->threads[p->thread_id];
        struct instruction *ip = t->ip;
 
-       TRACE("[Thread %2u] mov m.f %x\n",
+       TRACE("[Thread %2u] mov m.f %" PRIx64 "\n",
              p->thread_id,
              ip->mov.src_val);
 
@@ -2934,7 +3597,8 @@ instr_alu_add_translate(struct rte_swx_pipeline *p,
 {
        char *dst = tokens[1], *src = tokens[2];
        struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
 
        CHECK(n_tokens == 3, EINVAL);
 
@@ -2945,9 +3609,9 @@ instr_alu_add_translate(struct rte_swx_pipeline *p,
        fsrc = struct_field_parse(p, action, src, &src_struct_id);
        if (fsrc) {
                instr->type = INSTR_ALU_ADD;
-               if (dst[0] == 'h' && src[0] == 'm')
+               if (dst[0] == 'h' && src[0] != 'h')
                        instr->type = INSTR_ALU_ADD_HM;
-               if (dst[0] == 'm' && src[0] == 'h')
+               if (dst[0] != 'h' && src[0] == 'h')
                        instr->type = INSTR_ALU_ADD_MH;
                if (dst[0] == 'h' && src[0] == 'h')
                        instr->type = INSTR_ALU_ADD_HH;
@@ -2962,7 +3626,7 @@ instr_alu_add_translate(struct rte_swx_pipeline *p,
        }
 
        /* ADD_MI, ADD_HI. */
-       src_val = strtoul(src, &src, 0);
+       src_val = strtoull(src, &src, 0);
        CHECK(!src[0], EINVAL);
 
        instr->type = INSTR_ALU_ADD_MI;
@@ -2972,7 +3636,7 @@ instr_alu_add_translate(struct rte_swx_pipeline *p,
        instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
        instr->alu.dst.n_bits = fdst->n_bits;
        instr->alu.dst.offset = fdst->offset / 8;
-       instr->alu.src_val = (uint32_t)src_val;
+       instr->alu.src_val = src_val;
        return 0;
 }
 
@@ -2986,7 +3650,8 @@ instr_alu_sub_translate(struct rte_swx_pipeline *p,
 {
        char *dst = tokens[1], *src = tokens[2];
        struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
 
        CHECK(n_tokens == 3, EINVAL);
 
@@ -2997,9 +3662,9 @@ instr_alu_sub_translate(struct rte_swx_pipeline *p,
        fsrc = struct_field_parse(p, action, src, &src_struct_id);
        if (fsrc) {
                instr->type = INSTR_ALU_SUB;
-               if (dst[0] == 'h' && src[0] == 'm')
+               if (dst[0] == 'h' && src[0] != 'h')
                        instr->type = INSTR_ALU_SUB_HM;
-               if (dst[0] == 'm' && src[0] == 'h')
+               if (dst[0] != 'h' && src[0] == 'h')
                        instr->type = INSTR_ALU_SUB_MH;
                if (dst[0] == 'h' && src[0] == 'h')
                        instr->type = INSTR_ALU_SUB_HH;
@@ -3014,7 +3679,7 @@ instr_alu_sub_translate(struct rte_swx_pipeline *p,
        }
 
        /* SUB_MI, SUB_HI. */
-       src_val = strtoul(src, &src, 0);
+       src_val = strtoull(src, &src, 0);
        CHECK(!src[0], EINVAL);
 
        instr->type = INSTR_ALU_SUB_MI;
@@ -3024,7 +3689,7 @@ instr_alu_sub_translate(struct rte_swx_pipeline *p,
        instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
        instr->alu.dst.n_bits = fdst->n_bits;
        instr->alu.dst.offset = fdst->offset / 8;
-       instr->alu.src_val = (uint32_t)src_val;
+       instr->alu.src_val = src_val;
        return 0;
 }
 
@@ -3115,7 +3780,8 @@ instr_alu_shl_translate(struct rte_swx_pipeline *p,
 {
        char *dst = tokens[1], *src = tokens[2];
        struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
 
        CHECK(n_tokens == 3, EINVAL);
 
@@ -3126,9 +3792,9 @@ instr_alu_shl_translate(struct rte_swx_pipeline *p,
        fsrc = struct_field_parse(p, action, src, &src_struct_id);
        if (fsrc) {
                instr->type = INSTR_ALU_SHL;
-               if (dst[0] == 'h' && src[0] == 'm')
+               if (dst[0] == 'h' && src[0] != 'h')
                        instr->type = INSTR_ALU_SHL_HM;
-               if (dst[0] == 'm' && src[0] == 'h')
+               if (dst[0] != 'h' && src[0] == 'h')
                        instr->type = INSTR_ALU_SHL_MH;
                if (dst[0] == 'h' && src[0] == 'h')
                        instr->type = INSTR_ALU_SHL_HH;
@@ -3143,7 +3809,7 @@ instr_alu_shl_translate(struct rte_swx_pipeline *p,
        }
 
        /* SHL_MI, SHL_HI. */
-       src_val = strtoul(src, &src, 0);
+       src_val = strtoull(src, &src, 0);
        CHECK(!src[0], EINVAL);
 
        instr->type = INSTR_ALU_SHL_MI;
@@ -3153,7 +3819,60 @@ instr_alu_shl_translate(struct rte_swx_pipeline *p,
        instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
        instr->alu.dst.n_bits = fdst->n_bits;
        instr->alu.dst.offset = fdst->offset / 8;
-       instr->alu.src_val = (uint32_t)src_val;
+       instr->alu.src_val = src_val;
+       return 0;
+}
+
+static int
+instr_alu_shr_translate(struct rte_swx_pipeline *p,
+                       struct action *action,
+                       char **tokens,
+                       int n_tokens,
+                       struct instruction *instr,
+                       struct instruction_data *data __rte_unused)
+{
+       char *dst = tokens[1], *src = tokens[2];
+       struct field *fdst, *fsrc;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
+
+       CHECK(n_tokens == 3, EINVAL);
+
+       fdst = struct_field_parse(p, NULL, dst, &dst_struct_id);
+       CHECK(fdst, EINVAL);
+
+       /* SHR, SHR_HM, SHR_MH, SHR_HH. */
+       fsrc = struct_field_parse(p, action, src, &src_struct_id);
+       if (fsrc) {
+               instr->type = INSTR_ALU_SHR;
+               if (dst[0] == 'h' && src[0] != 'h')
+                       instr->type = INSTR_ALU_SHR_HM;
+               if (dst[0] != 'h' && src[0] == 'h')
+                       instr->type = INSTR_ALU_SHR_MH;
+               if (dst[0] == 'h' && src[0] == 'h')
+                       instr->type = INSTR_ALU_SHR_HH;
+
+               instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
+               instr->alu.dst.n_bits = fdst->n_bits;
+               instr->alu.dst.offset = fdst->offset / 8;
+               instr->alu.src.struct_id = (uint8_t)src_struct_id;
+               instr->alu.src.n_bits = fsrc->n_bits;
+               instr->alu.src.offset = fsrc->offset / 8;
+               return 0;
+       }
+
+       /* SHR_MI, SHR_HI. */
+       src_val = strtoull(src, &src, 0);
+       CHECK(!src[0], EINVAL);
+
+       instr->type = INSTR_ALU_SHR_MI;
+       if (dst[0] == 'h')
+               instr->type = INSTR_ALU_SHR_HI;
+
+       instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
+       instr->alu.dst.n_bits = fdst->n_bits;
+       instr->alu.dst.offset = fdst->offset / 8;
+       instr->alu.src_val = src_val;
        return 0;
 }
 
@@ -3167,7 +3886,8 @@ instr_alu_and_translate(struct rte_swx_pipeline *p,
 {
        char *dst = tokens[1], *src = tokens[2];
        struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
 
        CHECK(n_tokens == 3, EINVAL);
 
@@ -3192,17 +3912,17 @@ instr_alu_and_translate(struct rte_swx_pipeline *p,
        }
 
        /* AND_I. */
-       src_val = strtoul(src, &src, 0);
+       src_val = strtoull(src, &src, 0);
        CHECK(!src[0], EINVAL);
 
        if (dst[0] == 'h')
-               src_val = htonl(src_val);
+               src_val = hton64(src_val) >> (64 - fdst->n_bits);
 
        instr->type = INSTR_ALU_AND_I;
        instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
        instr->alu.dst.n_bits = fdst->n_bits;
        instr->alu.dst.offset = fdst->offset / 8;
-       instr->alu.src_val = (uint32_t)src_val;
+       instr->alu.src_val = src_val;
        return 0;
 }
 
@@ -3216,7 +3936,8 @@ instr_alu_or_translate(struct rte_swx_pipeline *p,
 {
        char *dst = tokens[1], *src = tokens[2];
        struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
 
        CHECK(n_tokens == 3, EINVAL);
 
@@ -3241,17 +3962,17 @@ instr_alu_or_translate(struct rte_swx_pipeline *p,
        }
 
        /* OR_I. */
-       src_val = strtoul(src, &src, 0);
+       src_val = strtoull(src, &src, 0);
        CHECK(!src[0], EINVAL);
 
        if (dst[0] == 'h')
-               src_val = htonl(src_val);
+               src_val = hton64(src_val) >> (64 - fdst->n_bits);
 
        instr->type = INSTR_ALU_OR_I;
        instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
        instr->alu.dst.n_bits = fdst->n_bits;
        instr->alu.dst.offset = fdst->offset / 8;
-       instr->alu.src_val = (uint32_t)src_val;
+       instr->alu.src_val = src_val;
        return 0;
 }
 
@@ -3265,7 +3986,8 @@ instr_alu_xor_translate(struct rte_swx_pipeline *p,
 {
        char *dst = tokens[1], *src = tokens[2];
        struct field *fdst, *fsrc;
-       uint32_t dst_struct_id, src_struct_id, src_val;
+       uint64_t src_val;
+       uint32_t dst_struct_id, src_struct_id;
 
        CHECK(n_tokens == 3, EINVAL);
 
@@ -3290,17 +4012,17 @@ instr_alu_xor_translate(struct rte_swx_pipeline *p,
        }
 
        /* XOR_I. */
-       src_val = strtoul(src, &src, 0);
+       src_val = strtoull(src, &src, 0);
        CHECK(!src[0], EINVAL);
 
        if (dst[0] == 'h')
-               src_val = htonl(src_val);
+               src_val = hton64(src_val) >> (64 - fdst->n_bits);
 
        instr->type = INSTR_ALU_XOR_I;
        instr->alu.dst.struct_id = (uint8_t)dst_struct_id;
        instr->alu.dst.n_bits = fdst->n_bits;
        instr->alu.dst.offset = fdst->offset / 8;
-       instr->alu.src_val = (uint32_t)src_val;
+       instr->alu.src_val = src_val;
        return 0;
 }
 
@@ -3574,6 +4296,96 @@ instr_alu_shl_hi_exec(struct rte_swx_pipeline *p)
        thread_ip_inc(p);
 }
 
+static inline void
+instr_alu_shr_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] shr\n", p->thread_id);
+
+       /* Structs. */
+       ALU(t, ip, >>);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_alu_shr_mh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] shr (mh)\n", p->thread_id);
+
+       /* Structs. */
+       ALU_MH(t, ip, >>);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_alu_shr_hm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] shr (hm)\n", p->thread_id);
+
+       /* Structs. */
+       ALU_HM(t, ip, >>);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_alu_shr_hh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] shr (hh)\n", p->thread_id);
+
+       /* Structs. */
+       ALU_HH(t, ip, >>);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_alu_shr_mi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] shr (mi)\n", p->thread_id);
+
+       /* Structs. */
+       ALU_MI(t, ip, >>);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_alu_shr_hi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] shr (hi)\n", p->thread_id);
+
+       /* Structs. */
+       ALU_HI(t, ip, >>);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
 static inline void
 instr_alu_and_exec(struct rte_swx_pipeline *p)
 {
@@ -3943,43 +4755,1626 @@ instr_alu_ckadd_struct_exec(struct rte_swx_pipeline *p)
        thread_ip_inc(p);
 }
 
-#define RTE_SWX_INSTRUCTION_TOKENS_MAX 16
+/*
+ * Register array.
+ */
+static struct regarray *
+regarray_find(struct rte_swx_pipeline *p, const char *name);
 
 static int
-instr_translate(struct rte_swx_pipeline *p,
-               struct action *action,
-               char *string,
-               struct instruction *instr,
-               struct instruction_data *data)
-{
-       char *tokens[RTE_SWX_INSTRUCTION_TOKENS_MAX];
-       int n_tokens = 0, tpos = 0;
+instr_regprefetch_translate(struct rte_swx_pipeline *p,
+                     struct action *action,
+                     char **tokens,
+                     int n_tokens,
+                     struct instruction *instr,
+                     struct instruction_data *data __rte_unused)
+{
+       char *regarray = tokens[1], *idx = tokens[2];
+       struct regarray *r;
+       struct field *fidx;
+       uint32_t idx_struct_id, idx_val;
 
-       /* Parse the instruction string into tokens. */
-       for ( ; ; ) {
-               char *token;
+       CHECK(n_tokens == 3, EINVAL);
 
-               token = strtok_r(string, " \t\v", &string);
-               if (!token)
-                       break;
+       r = regarray_find(p, regarray);
+       CHECK(r, EINVAL);
+
+       /* REGPREFETCH_RH, REGPREFETCH_RM. */
+       fidx = struct_field_parse(p, action, idx, &idx_struct_id);
+       if (fidx) {
+               instr->type = INSTR_REGPREFETCH_RM;
+               if (idx[0] == 'h')
+                       instr->type = INSTR_REGPREFETCH_RH;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx.struct_id = (uint8_t)idx_struct_id;
+               instr->regarray.idx.n_bits = fidx->n_bits;
+               instr->regarray.idx.offset = fidx->offset / 8;
+               instr->regarray.dstsrc_val = 0; /* Unused. */
+               return 0;
+       }
 
-               CHECK(n_tokens < RTE_SWX_INSTRUCTION_TOKENS_MAX, EINVAL);
+       /* REGPREFETCH_RI. */
+       idx_val = strtoul(idx, &idx, 0);
+       CHECK(!idx[0], EINVAL);
 
-               tokens[n_tokens] = token;
-               n_tokens++;
-       }
+       instr->type = INSTR_REGPREFETCH_RI;
+       instr->regarray.regarray_id = r->id;
+       instr->regarray.idx_val = idx_val;
+       instr->regarray.dstsrc_val = 0; /* Unused. */
+       return 0;
+}
 
-       CHECK(n_tokens, EINVAL);
+static int
+instr_regrd_translate(struct rte_swx_pipeline *p,
+                     struct action *action,
+                     char **tokens,
+                     int n_tokens,
+                     struct instruction *instr,
+                     struct instruction_data *data __rte_unused)
+{
+       char *dst = tokens[1], *regarray = tokens[2], *idx = tokens[3];
+       struct regarray *r;
+       struct field *fdst, *fidx;
+       uint32_t dst_struct_id, idx_struct_id, idx_val;
 
-       /* Handle the optional instruction label. */
-       if ((n_tokens >= 2) && !strcmp(tokens[1], ":")) {
-               strcpy(data->label, tokens[0]);
+       CHECK(n_tokens == 4, EINVAL);
 
-               tpos += 2;
-               CHECK(n_tokens - tpos, EINVAL);
-       }
+       r = regarray_find(p, regarray);
+       CHECK(r, EINVAL);
 
-       /* Identify the instruction type. */
+       fdst = struct_field_parse(p, NULL, dst, &dst_struct_id);
+       CHECK(fdst, EINVAL);
+
+       /* REGRD_HRH, REGRD_HRM, REGRD_MRH, REGRD_MRM. */
+       fidx = struct_field_parse(p, action, idx, &idx_struct_id);
+       if (fidx) {
+               instr->type = INSTR_REGRD_MRM;
+               if (dst[0] == 'h' && idx[0] != 'h')
+                       instr->type = INSTR_REGRD_HRM;
+               if (dst[0] != 'h' && idx[0] == 'h')
+                       instr->type = INSTR_REGRD_MRH;
+               if (dst[0] == 'h' && idx[0] == 'h')
+                       instr->type = INSTR_REGRD_HRH;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx.struct_id = (uint8_t)idx_struct_id;
+               instr->regarray.idx.n_bits = fidx->n_bits;
+               instr->regarray.idx.offset = fidx->offset / 8;
+               instr->regarray.dstsrc.struct_id = (uint8_t)dst_struct_id;
+               instr->regarray.dstsrc.n_bits = fdst->n_bits;
+               instr->regarray.dstsrc.offset = fdst->offset / 8;
+               return 0;
+       }
+
+       /* REGRD_MRI, REGRD_HRI. */
+       idx_val = strtoul(idx, &idx, 0);
+       CHECK(!idx[0], EINVAL);
+
+       instr->type = INSTR_REGRD_MRI;
+       if (dst[0] == 'h')
+               instr->type = INSTR_REGRD_HRI;
+
+       instr->regarray.regarray_id = r->id;
+       instr->regarray.idx_val = idx_val;
+       instr->regarray.dstsrc.struct_id = (uint8_t)dst_struct_id;
+       instr->regarray.dstsrc.n_bits = fdst->n_bits;
+       instr->regarray.dstsrc.offset = fdst->offset / 8;
+       return 0;
+}
+
+static int
+instr_regwr_translate(struct rte_swx_pipeline *p,
+                     struct action *action,
+                     char **tokens,
+                     int n_tokens,
+                     struct instruction *instr,
+                     struct instruction_data *data __rte_unused)
+{
+       char *regarray = tokens[1], *idx = tokens[2], *src = tokens[3];
+       struct regarray *r;
+       struct field *fidx, *fsrc;
+       uint64_t src_val;
+       uint32_t idx_struct_id, idx_val, src_struct_id;
+
+       CHECK(n_tokens == 4, EINVAL);
+
+       r = regarray_find(p, regarray);
+       CHECK(r, EINVAL);
+
+       /* REGWR_RHH, REGWR_RHM, REGWR_RMH, REGWR_RMM. */
+       fidx = struct_field_parse(p, action, idx, &idx_struct_id);
+       fsrc = struct_field_parse(p, action, src, &src_struct_id);
+       if (fidx && fsrc) {
+               instr->type = INSTR_REGWR_RMM;
+               if (idx[0] == 'h' && src[0] != 'h')
+                       instr->type = INSTR_REGWR_RHM;
+               if (idx[0] != 'h' && src[0] == 'h')
+                       instr->type = INSTR_REGWR_RMH;
+               if (idx[0] == 'h' && src[0] == 'h')
+                       instr->type = INSTR_REGWR_RHH;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx.struct_id = (uint8_t)idx_struct_id;
+               instr->regarray.idx.n_bits = fidx->n_bits;
+               instr->regarray.idx.offset = fidx->offset / 8;
+               instr->regarray.dstsrc.struct_id = (uint8_t)src_struct_id;
+               instr->regarray.dstsrc.n_bits = fsrc->n_bits;
+               instr->regarray.dstsrc.offset = fsrc->offset / 8;
+               return 0;
+       }
+
+       /* REGWR_RHI, REGWR_RMI. */
+       if (fidx && !fsrc) {
+               src_val = strtoull(src, &src, 0);
+               CHECK(!src[0], EINVAL);
+
+               instr->type = INSTR_REGWR_RMI;
+               if (idx[0] == 'h')
+                       instr->type = INSTR_REGWR_RHI;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx.struct_id = (uint8_t)idx_struct_id;
+               instr->regarray.idx.n_bits = fidx->n_bits;
+               instr->regarray.idx.offset = fidx->offset / 8;
+               instr->regarray.dstsrc_val = src_val;
+               return 0;
+       }
+
+       /* REGWR_RIH, REGWR_RIM. */
+       if (!fidx && fsrc) {
+               idx_val = strtoul(idx, &idx, 0);
+               CHECK(!idx[0], EINVAL);
+
+               instr->type = INSTR_REGWR_RIM;
+               if (src[0] == 'h')
+                       instr->type = INSTR_REGWR_RIH;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx_val = idx_val;
+               instr->regarray.dstsrc.struct_id = (uint8_t)src_struct_id;
+               instr->regarray.dstsrc.n_bits = fsrc->n_bits;
+               instr->regarray.dstsrc.offset = fsrc->offset / 8;
+               return 0;
+       }
+
+       /* REGWR_RII. */
+       src_val = strtoull(src, &src, 0);
+       CHECK(!src[0], EINVAL);
+
+       idx_val = strtoul(idx, &idx, 0);
+       CHECK(!idx[0], EINVAL);
+
+       instr->type = INSTR_REGWR_RII;
+       instr->regarray.idx_val = idx_val;
+       instr->regarray.dstsrc_val = src_val;
+
+       return 0;
+}
+
+static int
+instr_regadd_translate(struct rte_swx_pipeline *p,
+                      struct action *action,
+                      char **tokens,
+                      int n_tokens,
+                      struct instruction *instr,
+                      struct instruction_data *data __rte_unused)
+{
+       char *regarray = tokens[1], *idx = tokens[2], *src = tokens[3];
+       struct regarray *r;
+       struct field *fidx, *fsrc;
+       uint64_t src_val;
+       uint32_t idx_struct_id, idx_val, src_struct_id;
+
+       CHECK(n_tokens == 4, EINVAL);
+
+       r = regarray_find(p, regarray);
+       CHECK(r, EINVAL);
+
+       /* REGADD_RHH, REGADD_RHM, REGADD_RMH, REGADD_RMM. */
+       fidx = struct_field_parse(p, action, idx, &idx_struct_id);
+       fsrc = struct_field_parse(p, action, src, &src_struct_id);
+       if (fidx && fsrc) {
+               instr->type = INSTR_REGADD_RMM;
+               if (idx[0] == 'h' && src[0] != 'h')
+                       instr->type = INSTR_REGADD_RHM;
+               if (idx[0] != 'h' && src[0] == 'h')
+                       instr->type = INSTR_REGADD_RMH;
+               if (idx[0] == 'h' && src[0] == 'h')
+                       instr->type = INSTR_REGADD_RHH;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx.struct_id = (uint8_t)idx_struct_id;
+               instr->regarray.idx.n_bits = fidx->n_bits;
+               instr->regarray.idx.offset = fidx->offset / 8;
+               instr->regarray.dstsrc.struct_id = (uint8_t)src_struct_id;
+               instr->regarray.dstsrc.n_bits = fsrc->n_bits;
+               instr->regarray.dstsrc.offset = fsrc->offset / 8;
+               return 0;
+       }
+
+       /* REGADD_RHI, REGADD_RMI. */
+       if (fidx && !fsrc) {
+               src_val = strtoull(src, &src, 0);
+               CHECK(!src[0], EINVAL);
+
+               instr->type = INSTR_REGADD_RMI;
+               if (idx[0] == 'h')
+                       instr->type = INSTR_REGADD_RHI;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx.struct_id = (uint8_t)idx_struct_id;
+               instr->regarray.idx.n_bits = fidx->n_bits;
+               instr->regarray.idx.offset = fidx->offset / 8;
+               instr->regarray.dstsrc_val = src_val;
+               return 0;
+       }
+
+       /* REGADD_RIH, REGADD_RIM. */
+       if (!fidx && fsrc) {
+               idx_val = strtoul(idx, &idx, 0);
+               CHECK(!idx[0], EINVAL);
+
+               instr->type = INSTR_REGADD_RIM;
+               if (src[0] == 'h')
+                       instr->type = INSTR_REGADD_RIH;
+
+               instr->regarray.regarray_id = r->id;
+               instr->regarray.idx_val = idx_val;
+               instr->regarray.dstsrc.struct_id = (uint8_t)src_struct_id;
+               instr->regarray.dstsrc.n_bits = fsrc->n_bits;
+               instr->regarray.dstsrc.offset = fsrc->offset / 8;
+               return 0;
+       }
+
+       /* REGADD_RII. */
+       src_val = strtoull(src, &src, 0);
+       CHECK(!src[0], EINVAL);
+
+       idx_val = strtoul(idx, &idx, 0);
+       CHECK(!idx[0], EINVAL);
+
+       instr->type = INSTR_REGADD_RII;
+       instr->regarray.idx_val = idx_val;
+       instr->regarray.dstsrc_val = src_val;
+       return 0;
+}
+
+static inline uint64_t *
+instr_regarray_regarray(struct rte_swx_pipeline *p, struct instruction *ip)
+{
+       struct regarray_runtime *r = &p->regarray_runtime[ip->regarray.regarray_id];
+       return r->regarray;
+}
+
+static inline uint64_t
+instr_regarray_idx_hbo(struct rte_swx_pipeline *p, struct thread *t, struct instruction *ip)
+{
+       struct regarray_runtime *r = &p->regarray_runtime[ip->regarray.regarray_id];
+
+       uint8_t *idx_struct = t->structs[ip->regarray.idx.struct_id];
+       uint64_t *idx64_ptr = (uint64_t *)&idx_struct[ip->regarray.idx.offset];
+       uint64_t idx64 = *idx64_ptr;
+       uint64_t idx64_mask = UINT64_MAX >> (64 - ip->regarray.idx.n_bits);
+       uint64_t idx = idx64 & idx64_mask & r->size_mask;
+
+       return idx;
+}
+
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+
+static inline uint64_t
+instr_regarray_idx_nbo(struct rte_swx_pipeline *p, struct thread *t, struct instruction *ip)
+{
+       struct regarray_runtime *r = &p->regarray_runtime[ip->regarray.regarray_id];
+
+       uint8_t *idx_struct = t->structs[ip->regarray.idx.struct_id];
+       uint64_t *idx64_ptr = (uint64_t *)&idx_struct[ip->regarray.idx.offset];
+       uint64_t idx64 = *idx64_ptr;
+       uint64_t idx = (ntoh64(idx64) >> (64 - ip->regarray.idx.n_bits)) & r->size_mask;
+
+       return idx;
+}
+
+#else
+
+#define instr_regarray_idx_nbo instr_regarray_idx_hbo
+
+#endif
+
+static inline uint64_t
+instr_regarray_idx_imm(struct rte_swx_pipeline *p, struct instruction *ip)
+{
+       struct regarray_runtime *r = &p->regarray_runtime[ip->regarray.regarray_id];
+
+       uint64_t idx = ip->regarray.idx_val & r->size_mask;
+
+       return idx;
+}
+
+static inline uint64_t
+instr_regarray_src_hbo(struct thread *t, struct instruction *ip)
+{
+       uint8_t *src_struct = t->structs[ip->regarray.dstsrc.struct_id];
+       uint64_t *src64_ptr = (uint64_t *)&src_struct[ip->regarray.dstsrc.offset];
+       uint64_t src64 = *src64_ptr;
+       uint64_t src64_mask = UINT64_MAX >> (64 - ip->regarray.dstsrc.n_bits);
+       uint64_t src = src64 & src64_mask;
+
+       return src;
+}
+
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+
+static inline uint64_t
+instr_regarray_src_nbo(struct thread *t, struct instruction *ip)
+{
+       uint8_t *src_struct = t->structs[ip->regarray.dstsrc.struct_id];
+       uint64_t *src64_ptr = (uint64_t *)&src_struct[ip->regarray.dstsrc.offset];
+       uint64_t src64 = *src64_ptr;
+       uint64_t src = ntoh64(src64) >> (64 - ip->regarray.dstsrc.n_bits);
+
+       return src;
+}
+
+#else
+
+#define instr_regarray_src_nbo instr_regarray_src_hbo
+
+#endif
+
+static inline void
+instr_regarray_dst_hbo_src_hbo_set(struct thread *t, struct instruction *ip, uint64_t src)
+{
+       uint8_t *dst_struct = t->structs[ip->regarray.dstsrc.struct_id];
+       uint64_t *dst64_ptr = (uint64_t *)&dst_struct[ip->regarray.dstsrc.offset];
+       uint64_t dst64 = *dst64_ptr;
+       uint64_t dst64_mask = UINT64_MAX >> (64 - ip->regarray.dstsrc.n_bits);
+
+       *dst64_ptr = (dst64 & ~dst64_mask) | (src & dst64_mask);
+
+}
+
+#if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
+
+static inline void
+instr_regarray_dst_nbo_src_hbo_set(struct thread *t, struct instruction *ip, uint64_t src)
+{
+       uint8_t *dst_struct = t->structs[ip->regarray.dstsrc.struct_id];
+       uint64_t *dst64_ptr = (uint64_t *)&dst_struct[ip->regarray.dstsrc.offset];
+       uint64_t dst64 = *dst64_ptr;
+       uint64_t dst64_mask = UINT64_MAX >> (64 - ip->regarray.dstsrc.n_bits);
+
+       src = hton64(src) >> (64 - ip->regarray.dstsrc.n_bits);
+       *dst64_ptr = (dst64 & ~dst64_mask) | (src & dst64_mask);
+}
+
+#else
+
+#define instr_regarray_dst_nbo_src_hbo_set instr_regarray_dst_hbo_src_hbo_set
+
+#endif
+
+static inline void
+instr_regprefetch_rh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regprefetch (r[h])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       rte_prefetch0(&regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regprefetch_rm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regprefetch (r[m])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       rte_prefetch0(&regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regprefetch_ri_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regprefetch (r[i])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       rte_prefetch0(&regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regrd_hrh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regrd (h = r[h])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       instr_regarray_dst_nbo_src_hbo_set(t, ip, regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regrd_hrm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regrd (h = r[m])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       instr_regarray_dst_nbo_src_hbo_set(t, ip, regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regrd_mrh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regrd (m = r[h])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       instr_regarray_dst_hbo_src_hbo_set(t, ip, regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regrd_mrm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       instr_regarray_dst_hbo_src_hbo_set(t, ip, regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regrd_hri_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regrd (h = r[i])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       instr_regarray_dst_nbo_src_hbo_set(t, ip, regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regrd_mri_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx;
+
+       TRACE("[Thread %2u] regrd (m = r[i])\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       instr_regarray_dst_hbo_src_hbo_set(t, ip, regarray[idx]);
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rhh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[h] = h)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       src = instr_regarray_src_nbo(t, ip);
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rhm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[h] = m)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       src = instr_regarray_src_hbo(t, ip);
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rmh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[m] = h)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       src = instr_regarray_src_nbo(t, ip);
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rmm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[m] = m)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       src = instr_regarray_src_hbo(t, ip);
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rhi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[h] = i)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       src = ip->regarray.dstsrc_val;
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rmi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[m] = i)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       src = ip->regarray.dstsrc_val;
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rih_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[i] = h)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       src = instr_regarray_src_nbo(t, ip);
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rim_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[i] = m)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       src = instr_regarray_src_hbo(t, ip);
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regwr_rii_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regwr (r[i] = i)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       src = ip->regarray.dstsrc_val;
+       regarray[idx] = src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rhh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[h] += h)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       src = instr_regarray_src_nbo(t, ip);
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rhm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[h] += m)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       src = instr_regarray_src_hbo(t, ip);
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rmh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[m] += h)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       src = instr_regarray_src_nbo(t, ip);
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rmm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[m] += m)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       src = instr_regarray_src_hbo(t, ip);
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rhi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[h] += i)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_nbo(p, t, ip);
+       src = ip->regarray.dstsrc_val;
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rmi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[m] += i)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_hbo(p, t, ip);
+       src = ip->regarray.dstsrc_val;
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rih_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[i] += h)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       src = instr_regarray_src_nbo(t, ip);
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rim_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[i] += m)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       src = instr_regarray_src_hbo(t, ip);
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+static inline void
+instr_regadd_rii_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint64_t *regarray, idx, src;
+
+       TRACE("[Thread %2u] regadd (r[i] += i)\n", p->thread_id);
+
+       /* Structs. */
+       regarray = instr_regarray_regarray(p, ip);
+       idx = instr_regarray_idx_imm(p, ip);
+       src = ip->regarray.dstsrc_val;
+       regarray[idx] += src;
+
+       /* Thread. */
+       thread_ip_inc(p);
+}
+
+/*
+ * jmp.
+ */
+static struct action *
+action_find(struct rte_swx_pipeline *p, const char *name);
+
+static int
+instr_jmp_translate(struct rte_swx_pipeline *p __rte_unused,
+                   struct action *action __rte_unused,
+                   char **tokens,
+                   int n_tokens,
+                   struct instruction *instr,
+                   struct instruction_data *data)
+{
+       CHECK(n_tokens == 2, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       instr->type = INSTR_JMP;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       return 0;
+}
+
+static int
+instr_jmp_valid_translate(struct rte_swx_pipeline *p,
+                         struct action *action __rte_unused,
+                         char **tokens,
+                         int n_tokens,
+                         struct instruction *instr,
+                         struct instruction_data *data)
+{
+       struct header *h;
+
+       CHECK(n_tokens == 3, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       h = header_parse(p, tokens[2]);
+       CHECK(h, EINVAL);
+
+       instr->type = INSTR_JMP_VALID;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       instr->jmp.header_id = h->id;
+       return 0;
+}
+
+static int
+instr_jmp_invalid_translate(struct rte_swx_pipeline *p,
+                           struct action *action __rte_unused,
+                           char **tokens,
+                           int n_tokens,
+                           struct instruction *instr,
+                           struct instruction_data *data)
+{
+       struct header *h;
+
+       CHECK(n_tokens == 3, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       h = header_parse(p, tokens[2]);
+       CHECK(h, EINVAL);
+
+       instr->type = INSTR_JMP_INVALID;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       instr->jmp.header_id = h->id;
+       return 0;
+}
+
+static int
+instr_jmp_hit_translate(struct rte_swx_pipeline *p __rte_unused,
+                       struct action *action,
+                       char **tokens,
+                       int n_tokens,
+                       struct instruction *instr,
+                       struct instruction_data *data)
+{
+       CHECK(!action, EINVAL);
+       CHECK(n_tokens == 2, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       instr->type = INSTR_JMP_HIT;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       return 0;
+}
+
+static int
+instr_jmp_miss_translate(struct rte_swx_pipeline *p __rte_unused,
+                        struct action *action,
+                        char **tokens,
+                        int n_tokens,
+                        struct instruction *instr,
+                        struct instruction_data *data)
+{
+       CHECK(!action, EINVAL);
+       CHECK(n_tokens == 2, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       instr->type = INSTR_JMP_MISS;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       return 0;
+}
+
+static int
+instr_jmp_action_hit_translate(struct rte_swx_pipeline *p,
+                              struct action *action,
+                              char **tokens,
+                              int n_tokens,
+                              struct instruction *instr,
+                              struct instruction_data *data)
+{
+       struct action *a;
+
+       CHECK(!action, EINVAL);
+       CHECK(n_tokens == 3, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       a = action_find(p, tokens[2]);
+       CHECK(a, EINVAL);
+
+       instr->type = INSTR_JMP_ACTION_HIT;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       instr->jmp.action_id = a->id;
+       return 0;
+}
+
+static int
+instr_jmp_action_miss_translate(struct rte_swx_pipeline *p,
+                               struct action *action,
+                               char **tokens,
+                               int n_tokens,
+                               struct instruction *instr,
+                               struct instruction_data *data)
+{
+       struct action *a;
+
+       CHECK(!action, EINVAL);
+       CHECK(n_tokens == 3, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       a = action_find(p, tokens[2]);
+       CHECK(a, EINVAL);
+
+       instr->type = INSTR_JMP_ACTION_MISS;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       instr->jmp.action_id = a->id;
+       return 0;
+}
+
+static int
+instr_jmp_eq_translate(struct rte_swx_pipeline *p,
+                      struct action *action,
+                      char **tokens,
+                      int n_tokens,
+                      struct instruction *instr,
+                      struct instruction_data *data)
+{
+       char *a = tokens[2], *b = tokens[3];
+       struct field *fa, *fb;
+       uint64_t b_val;
+       uint32_t a_struct_id, b_struct_id;
+
+       CHECK(n_tokens == 4, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       fa = struct_field_parse(p, action, a, &a_struct_id);
+       CHECK(fa, EINVAL);
+
+       /* JMP_EQ or JMP_EQ_S. */
+       fb = struct_field_parse(p, action, b, &b_struct_id);
+       if (fb) {
+               instr->type = INSTR_JMP_EQ;
+               if ((a[0] == 'h' && b[0] != 'h') ||
+                   (a[0] != 'h' && b[0] == 'h'))
+                       instr->type = INSTR_JMP_EQ_S;
+               instr->jmp.ip = NULL; /* Resolved later. */
+
+               instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+               instr->jmp.a.n_bits = fa->n_bits;
+               instr->jmp.a.offset = fa->offset / 8;
+               instr->jmp.b.struct_id = (uint8_t)b_struct_id;
+               instr->jmp.b.n_bits = fb->n_bits;
+               instr->jmp.b.offset = fb->offset / 8;
+               return 0;
+       }
+
+       /* JMP_EQ_I. */
+       b_val = strtoull(b, &b, 0);
+       CHECK(!b[0], EINVAL);
+
+       if (a[0] == 'h')
+               b_val = hton64(b_val) >> (64 - fa->n_bits);
+
+       instr->type = INSTR_JMP_EQ_I;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+       instr->jmp.a.n_bits = fa->n_bits;
+       instr->jmp.a.offset = fa->offset / 8;
+       instr->jmp.b_val = b_val;
+       return 0;
+}
+
+static int
+instr_jmp_neq_translate(struct rte_swx_pipeline *p,
+                       struct action *action,
+                       char **tokens,
+                       int n_tokens,
+                       struct instruction *instr,
+                       struct instruction_data *data)
+{
+       char *a = tokens[2], *b = tokens[3];
+       struct field *fa, *fb;
+       uint64_t b_val;
+       uint32_t a_struct_id, b_struct_id;
+
+       CHECK(n_tokens == 4, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       fa = struct_field_parse(p, action, a, &a_struct_id);
+       CHECK(fa, EINVAL);
+
+       /* JMP_NEQ or JMP_NEQ_S. */
+       fb = struct_field_parse(p, action, b, &b_struct_id);
+       if (fb) {
+               instr->type = INSTR_JMP_NEQ;
+               if ((a[0] == 'h' && b[0] != 'h') ||
+                   (a[0] != 'h' && b[0] == 'h'))
+                       instr->type = INSTR_JMP_NEQ_S;
+               instr->jmp.ip = NULL; /* Resolved later. */
+
+               instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+               instr->jmp.a.n_bits = fa->n_bits;
+               instr->jmp.a.offset = fa->offset / 8;
+               instr->jmp.b.struct_id = (uint8_t)b_struct_id;
+               instr->jmp.b.n_bits = fb->n_bits;
+               instr->jmp.b.offset = fb->offset / 8;
+               return 0;
+       }
+
+       /* JMP_NEQ_I. */
+       b_val = strtoull(b, &b, 0);
+       CHECK(!b[0], EINVAL);
+
+       if (a[0] == 'h')
+               b_val = hton64(b_val) >> (64 - fa->n_bits);
+
+       instr->type = INSTR_JMP_NEQ_I;
+       instr->jmp.ip = NULL; /* Resolved later. */
+       instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+       instr->jmp.a.n_bits = fa->n_bits;
+       instr->jmp.a.offset = fa->offset / 8;
+       instr->jmp.b_val = b_val;
+       return 0;
+}
+
+static int
+instr_jmp_lt_translate(struct rte_swx_pipeline *p,
+                      struct action *action,
+                      char **tokens,
+                      int n_tokens,
+                      struct instruction *instr,
+                      struct instruction_data *data)
+{
+       char *a = tokens[2], *b = tokens[3];
+       struct field *fa, *fb;
+       uint64_t b_val;
+       uint32_t a_struct_id, b_struct_id;
+
+       CHECK(n_tokens == 4, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       fa = struct_field_parse(p, action, a, &a_struct_id);
+       CHECK(fa, EINVAL);
+
+       /* JMP_LT, JMP_LT_MH, JMP_LT_HM, JMP_LT_HH. */
+       fb = struct_field_parse(p, action, b, &b_struct_id);
+       if (fb) {
+               instr->type = INSTR_JMP_LT;
+               if (a[0] == 'h' && b[0] != 'h')
+                       instr->type = INSTR_JMP_LT_HM;
+               if (a[0] != 'h' && b[0] == 'h')
+                       instr->type = INSTR_JMP_LT_MH;
+               if (a[0] == 'h' && b[0] == 'h')
+                       instr->type = INSTR_JMP_LT_HH;
+               instr->jmp.ip = NULL; /* Resolved later. */
+
+               instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+               instr->jmp.a.n_bits = fa->n_bits;
+               instr->jmp.a.offset = fa->offset / 8;
+               instr->jmp.b.struct_id = (uint8_t)b_struct_id;
+               instr->jmp.b.n_bits = fb->n_bits;
+               instr->jmp.b.offset = fb->offset / 8;
+               return 0;
+       }
+
+       /* JMP_LT_MI, JMP_LT_HI. */
+       b_val = strtoull(b, &b, 0);
+       CHECK(!b[0], EINVAL);
+
+       instr->type = INSTR_JMP_LT_MI;
+       if (a[0] == 'h')
+               instr->type = INSTR_JMP_LT_HI;
+       instr->jmp.ip = NULL; /* Resolved later. */
+
+       instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+       instr->jmp.a.n_bits = fa->n_bits;
+       instr->jmp.a.offset = fa->offset / 8;
+       instr->jmp.b_val = b_val;
+       return 0;
+}
+
+static int
+instr_jmp_gt_translate(struct rte_swx_pipeline *p,
+                      struct action *action,
+                      char **tokens,
+                      int n_tokens,
+                      struct instruction *instr,
+                      struct instruction_data *data)
+{
+       char *a = tokens[2], *b = tokens[3];
+       struct field *fa, *fb;
+       uint64_t b_val;
+       uint32_t a_struct_id, b_struct_id;
+
+       CHECK(n_tokens == 4, EINVAL);
+
+       strcpy(data->jmp_label, tokens[1]);
+
+       fa = struct_field_parse(p, action, a, &a_struct_id);
+       CHECK(fa, EINVAL);
+
+       /* JMP_GT, JMP_GT_MH, JMP_GT_HM, JMP_GT_HH. */
+       fb = struct_field_parse(p, action, b, &b_struct_id);
+       if (fb) {
+               instr->type = INSTR_JMP_GT;
+               if (a[0] == 'h' && b[0] != 'h')
+                       instr->type = INSTR_JMP_GT_HM;
+               if (a[0] != 'h' && b[0] == 'h')
+                       instr->type = INSTR_JMP_GT_MH;
+               if (a[0] == 'h' && b[0] == 'h')
+                       instr->type = INSTR_JMP_GT_HH;
+               instr->jmp.ip = NULL; /* Resolved later. */
+
+               instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+               instr->jmp.a.n_bits = fa->n_bits;
+               instr->jmp.a.offset = fa->offset / 8;
+               instr->jmp.b.struct_id = (uint8_t)b_struct_id;
+               instr->jmp.b.n_bits = fb->n_bits;
+               instr->jmp.b.offset = fb->offset / 8;
+               return 0;
+       }
+
+       /* JMP_GT_MI, JMP_GT_HI. */
+       b_val = strtoull(b, &b, 0);
+       CHECK(!b[0], EINVAL);
+
+       instr->type = INSTR_JMP_GT_MI;
+       if (a[0] == 'h')
+               instr->type = INSTR_JMP_GT_HI;
+       instr->jmp.ip = NULL; /* Resolved later. */
+
+       instr->jmp.a.struct_id = (uint8_t)a_struct_id;
+       instr->jmp.a.n_bits = fa->n_bits;
+       instr->jmp.a.offset = fa->offset / 8;
+       instr->jmp.b_val = b_val;
+       return 0;
+}
+
+static inline void
+instr_jmp_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmp\n", p->thread_id);
+
+       thread_ip_set(t, ip->jmp.ip);
+}
+
+static inline void
+instr_jmp_valid_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint32_t header_id = ip->jmp.header_id;
+
+       TRACE("[Thread %2u] jmpv\n", p->thread_id);
+
+       t->ip = HEADER_VALID(t, header_id) ? ip->jmp.ip : (t->ip + 1);
+}
+
+static inline void
+instr_jmp_invalid_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       uint32_t header_id = ip->jmp.header_id;
+
+       TRACE("[Thread %2u] jmpnv\n", p->thread_id);
+
+       t->ip = HEADER_VALID(t, header_id) ? (t->ip + 1) : ip->jmp.ip;
+}
+
+static inline void
+instr_jmp_hit_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       struct instruction *ip_next[] = {t->ip + 1, ip->jmp.ip};
+
+       TRACE("[Thread %2u] jmph\n", p->thread_id);
+
+       t->ip = ip_next[t->hit];
+}
+
+static inline void
+instr_jmp_miss_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+       struct instruction *ip_next[] = {ip->jmp.ip, t->ip + 1};
+
+       TRACE("[Thread %2u] jmpnh\n", p->thread_id);
+
+       t->ip = ip_next[t->hit];
+}
+
+static inline void
+instr_jmp_action_hit_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpa\n", p->thread_id);
+
+       t->ip = (ip->jmp.action_id == t->action_id) ? ip->jmp.ip : (t->ip + 1);
+}
+
+static inline void
+instr_jmp_action_miss_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpna\n", p->thread_id);
+
+       t->ip = (ip->jmp.action_id == t->action_id) ? (t->ip + 1) : ip->jmp.ip;
+}
+
+static inline void
+instr_jmp_eq_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpeq\n", p->thread_id);
+
+       JMP_CMP(t, ip, ==);
+}
+
+static inline void
+instr_jmp_eq_s_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpeq (s)\n", p->thread_id);
+
+       JMP_CMP_S(t, ip, ==);
+}
+
+static inline void
+instr_jmp_eq_i_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpeq (i)\n", p->thread_id);
+
+       JMP_CMP_I(t, ip, ==);
+}
+
+static inline void
+instr_jmp_neq_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpneq\n", p->thread_id);
+
+       JMP_CMP(t, ip, !=);
+}
+
+static inline void
+instr_jmp_neq_s_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpneq (s)\n", p->thread_id);
+
+       JMP_CMP_S(t, ip, !=);
+}
+
+static inline void
+instr_jmp_neq_i_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpneq (i)\n", p->thread_id);
+
+       JMP_CMP_I(t, ip, !=);
+}
+
+static inline void
+instr_jmp_lt_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmplt\n", p->thread_id);
+
+       JMP_CMP(t, ip, <);
+}
+
+static inline void
+instr_jmp_lt_mh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmplt (mh)\n", p->thread_id);
+
+       JMP_CMP_MH(t, ip, <);
+}
+
+static inline void
+instr_jmp_lt_hm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmplt (hm)\n", p->thread_id);
+
+       JMP_CMP_HM(t, ip, <);
+}
+
+static inline void
+instr_jmp_lt_hh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmplt (hh)\n", p->thread_id);
+
+       JMP_CMP_HH(t, ip, <);
+}
+
+static inline void
+instr_jmp_lt_mi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmplt (mi)\n", p->thread_id);
+
+       JMP_CMP_MI(t, ip, <);
+}
+
+static inline void
+instr_jmp_lt_hi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmplt (hi)\n", p->thread_id);
+
+       JMP_CMP_HI(t, ip, <);
+}
+
+static inline void
+instr_jmp_gt_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpgt\n", p->thread_id);
+
+       JMP_CMP(t, ip, >);
+}
+
+static inline void
+instr_jmp_gt_mh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpgt (mh)\n", p->thread_id);
+
+       JMP_CMP_MH(t, ip, >);
+}
+
+static inline void
+instr_jmp_gt_hm_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpgt (hm)\n", p->thread_id);
+
+       JMP_CMP_HM(t, ip, >);
+}
+
+static inline void
+instr_jmp_gt_hh_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpgt (hh)\n", p->thread_id);
+
+       JMP_CMP_HH(t, ip, >);
+}
+
+static inline void
+instr_jmp_gt_mi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpgt (mi)\n", p->thread_id);
+
+       JMP_CMP_MI(t, ip, >);
+}
+
+static inline void
+instr_jmp_gt_hi_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+       struct instruction *ip = t->ip;
+
+       TRACE("[Thread %2u] jmpgt (hi)\n", p->thread_id);
+
+       JMP_CMP_HI(t, ip, >);
+}
+
+/*
+ * return.
+ */
+static int
+instr_return_translate(struct rte_swx_pipeline *p __rte_unused,
+                      struct action *action,
+                      char **tokens __rte_unused,
+                      int n_tokens,
+                      struct instruction *instr,
+                      struct instruction_data *data __rte_unused)
+{
+       CHECK(action, EINVAL);
+       CHECK(n_tokens == 1, EINVAL);
+
+       instr->type = INSTR_RETURN;
+       return 0;
+}
+
+static inline void
+instr_return_exec(struct rte_swx_pipeline *p)
+{
+       struct thread *t = &p->threads[p->thread_id];
+
+       TRACE("[Thread %2u] return\n", p->thread_id);
+
+       t->ip = t->ret;
+}
+
+static int
+instr_translate(struct rte_swx_pipeline *p,
+               struct action *action,
+               char *string,
+               struct instruction *instr,
+               struct instruction_data *data)
+{
+       char *tokens[RTE_SWX_INSTRUCTION_TOKENS_MAX];
+       int n_tokens = 0, tpos = 0;
+
+       /* Parse the instruction string into tokens. */
+       for ( ; ; ) {
+               char *token;
+
+               token = strtok_r(string, " \t\v", &string);
+               if (!token)
+                       break;
+
+               CHECK(n_tokens < RTE_SWX_INSTRUCTION_TOKENS_MAX, EINVAL);
+               CHECK_NAME(token, EINVAL);
+
+               tokens[n_tokens] = token;
+               n_tokens++;
+       }
+
+       CHECK(n_tokens, EINVAL);
+
+       /* Handle the optional instruction label. */
+       if ((n_tokens >= 2) && !strcmp(tokens[1], ":")) {
+               strcpy(data->label, tokens[0]);
+
+               tpos += 2;
+               CHECK(n_tokens - tpos, EINVAL);
+       }
+
+       /* Identify the instruction type. */
        if (!strcmp(tokens[tpos], "rx"))
                return instr_rx_translate(p,
                                          action,
@@ -4108,9 +6503,173 @@ instr_translate(struct rte_swx_pipeline *p,
                                               instr,
                                               data);
 
+       if (!strcmp(tokens[tpos], "shr"))
+               return instr_alu_shr_translate(p,
+                                              action,
+                                              &tokens[tpos],
+                                              n_tokens - tpos,
+                                              instr,
+                                              data);
+
+       if (!strcmp(tokens[tpos], "regprefetch"))
+               return instr_regprefetch_translate(p,
+                                                  action,
+                                                  &tokens[tpos],
+                                                  n_tokens - tpos,
+                                                  instr,
+                                                  data);
+
+       if (!strcmp(tokens[tpos], "regrd"))
+               return instr_regrd_translate(p,
+                                            action,
+                                            &tokens[tpos],
+                                            n_tokens - tpos,
+                                            instr,
+                                            data);
+
+       if (!strcmp(tokens[tpos], "regwr"))
+               return instr_regwr_translate(p,
+                                            action,
+                                            &tokens[tpos],
+                                            n_tokens - tpos,
+                                            instr,
+                                            data);
+
+       if (!strcmp(tokens[tpos], "regadd"))
+               return instr_regadd_translate(p,
+                                             action,
+                                             &tokens[tpos],
+                                             n_tokens - tpos,
+                                             instr,
+                                             data);
+
+       if (!strcmp(tokens[tpos], "table"))
+               return instr_table_translate(p,
+                                            action,
+                                            &tokens[tpos],
+                                            n_tokens - tpos,
+                                            instr,
+                                            data);
+
+       if (!strcmp(tokens[tpos], "extern"))
+               return instr_extern_translate(p,
+                                             action,
+                                             &tokens[tpos],
+                                             n_tokens - tpos,
+                                             instr,
+                                             data);
+
+       if (!strcmp(tokens[tpos], "jmp"))
+               return instr_jmp_translate(p,
+                                          action,
+                                          &tokens[tpos],
+                                          n_tokens - tpos,
+                                          instr,
+                                          data);
+
+       if (!strcmp(tokens[tpos], "jmpv"))
+               return instr_jmp_valid_translate(p,
+                                                action,
+                                                &tokens[tpos],
+                                                n_tokens - tpos,
+                                                instr,
+                                                data);
+
+       if (!strcmp(tokens[tpos], "jmpnv"))
+               return instr_jmp_invalid_translate(p,
+                                                  action,
+                                                  &tokens[tpos],
+                                                  n_tokens - tpos,
+                                                  instr,
+                                                  data);
+
+       if (!strcmp(tokens[tpos], "jmph"))
+               return instr_jmp_hit_translate(p,
+                                              action,
+                                              &tokens[tpos],
+                                              n_tokens - tpos,
+                                              instr,
+                                              data);
+
+       if (!strcmp(tokens[tpos], "jmpnh"))
+               return instr_jmp_miss_translate(p,
+                                               action,
+                                               &tokens[tpos],
+                                               n_tokens - tpos,
+                                               instr,
+                                               data);
+
+       if (!strcmp(tokens[tpos], "jmpa"))
+               return instr_jmp_action_hit_translate(p,
+                                                     action,
+                                                     &tokens[tpos],
+                                                     n_tokens - tpos,
+                                                     instr,
+                                                     data);
+
+       if (!strcmp(tokens[tpos], "jmpna"))
+               return instr_jmp_action_miss_translate(p,
+                                                      action,
+                                                      &tokens[tpos],
+                                                      n_tokens - tpos,
+                                                      instr,
+                                                      data);
+
+       if (!strcmp(tokens[tpos], "jmpeq"))
+               return instr_jmp_eq_translate(p,
+                                             action,
+                                             &tokens[tpos],
+                                             n_tokens - tpos,
+                                             instr,
+                                             data);
+
+       if (!strcmp(tokens[tpos], "jmpneq"))
+               return instr_jmp_neq_translate(p,
+                                              action,
+                                              &tokens[tpos],
+                                              n_tokens - tpos,
+                                              instr,
+                                              data);
+
+       if (!strcmp(tokens[tpos], "jmplt"))
+               return instr_jmp_lt_translate(p,
+                                             action,
+                                             &tokens[tpos],
+                                             n_tokens - tpos,
+                                             instr,
+                                             data);
+
+       if (!strcmp(tokens[tpos], "jmpgt"))
+               return instr_jmp_gt_translate(p,
+                                             action,
+                                             &tokens[tpos],
+                                             n_tokens - tpos,
+                                             instr,
+                                             data);
+
+       if (!strcmp(tokens[tpos], "return"))
+               return instr_return_translate(p,
+                                             action,
+                                             &tokens[tpos],
+                                             n_tokens - tpos,
+                                             instr,
+                                             data);
+
        CHECK(0, EINVAL);
 }
 
+static struct instruction_data *
+label_find(struct instruction_data *data, uint32_t n, const char *label)
+{
+       uint32_t i;
+
+       for (i = 0; i < n; i++)
+               if (!strcmp(label, data[i].label))
+                       return &data[i];
+
+       return NULL;
+}
+
 static uint32_t
 label_is_used(struct instruction_data *data, uint32_t n, const char *label)
 {
@@ -4145,17 +6704,317 @@ instr_label_check(struct instruction_data *instruction_data,
                        CHECK(strcmp(label, data[j].label), EINVAL);
        }
 
-       /* Get users for each instruction label. */
+       /* Get users for each instruction label. */
+       for (i = 0; i < n_instructions; i++) {
+               struct instruction_data *data = &instruction_data[i];
+               char *label = data->label;
+
+               data->n_users = label_is_used(instruction_data,
+                                             n_instructions,
+                                             label);
+       }
+
+       return 0;
+}
+
+static int
+instr_jmp_resolve(struct instruction *instructions,
+                 struct instruction_data *instruction_data,
+                 uint32_t n_instructions)
+{
+       uint32_t i;
+
+       for (i = 0; i < n_instructions; i++) {
+               struct instruction *instr = &instructions[i];
+               struct instruction_data *data = &instruction_data[i];
+               struct instruction_data *found;
+
+               if (!instruction_is_jmp(instr))
+                       continue;
+
+               found = label_find(instruction_data,
+                                  n_instructions,
+                                  data->jmp_label);
+               CHECK(found, EINVAL);
+
+               instr->jmp.ip = &instructions[found - instruction_data];
+       }
+
+       return 0;
+}
+
+static int
+instr_verify(struct rte_swx_pipeline *p __rte_unused,
+            struct action *a,
+            struct instruction *instr,
+            struct instruction_data *data __rte_unused,
+            uint32_t n_instructions)
+{
+       if (!a) {
+               enum instruction_type type;
+               uint32_t i;
+
+               /* Check that the first instruction is rx. */
+               CHECK(instr[0].type == INSTR_RX, EINVAL);
+
+               /* Check that there is at least one tx instruction. */
+               for (i = 0; i < n_instructions; i++) {
+                       type = instr[i].type;
+
+                       if (type == INSTR_TX)
+                               break;
+               }
+               CHECK(i < n_instructions, EINVAL);
+
+               /* Check that the last instruction is either tx or unconditional
+                * jump.
+                */
+               type = instr[n_instructions - 1].type;
+               CHECK((type == INSTR_TX) || (type == INSTR_JMP), EINVAL);
+       }
+
+       if (a) {
+               enum instruction_type type;
+               uint32_t i;
+
+               /* Check that there is at least one return or tx instruction. */
+               for (i = 0; i < n_instructions; i++) {
+                       type = instr[i].type;
+
+                       if ((type == INSTR_RETURN) || (type == INSTR_TX))
+                               break;
+               }
+               CHECK(i < n_instructions, EINVAL);
+       }
+
+       return 0;
+}
+
+static int
+instr_pattern_extract_many_detect(struct instruction *instr,
+                                 struct instruction_data *data,
+                                 uint32_t n_instr,
+                                 uint32_t *n_pattern_instr)
+{
+       uint32_t i;
+
+       for (i = 0; i < n_instr; i++) {
+               if (data[i].invalid)
+                       break;
+
+               if (instr[i].type != INSTR_HDR_EXTRACT)
+                       break;
+
+               if (i == RTE_DIM(instr->io.hdr.header_id))
+                       break;
+
+               if (i && data[i].n_users)
+                       break;
+       }
+
+       if (i < 2)
+               return 0;
+
+       *n_pattern_instr = i;
+       return 1;
+}
+
+static void
+instr_pattern_extract_many_optimize(struct instruction *instr,
+                                   struct instruction_data *data,
+                                   uint32_t n_instr)
+{
+       uint32_t i;
+
+       for (i = 1; i < n_instr; i++) {
+               instr[0].type++;
+               instr[0].io.hdr.header_id[i] = instr[i].io.hdr.header_id[0];
+               instr[0].io.hdr.struct_id[i] = instr[i].io.hdr.struct_id[0];
+               instr[0].io.hdr.n_bytes[i] = instr[i].io.hdr.n_bytes[0];
+
+               data[i].invalid = 1;
+       }
+}
+
+static int
+instr_pattern_emit_many_tx_detect(struct instruction *instr,
+                                 struct instruction_data *data,
+                                 uint32_t n_instr,
+                                 uint32_t *n_pattern_instr)
+{
+       uint32_t i;
+
+       for (i = 0; i < n_instr; i++) {
+               if (data[i].invalid)
+                       break;
+
+               if (instr[i].type != INSTR_HDR_EMIT)
+                       break;
+
+               if (i == RTE_DIM(instr->io.hdr.header_id))
+                       break;
+
+               if (i && data[i].n_users)
+                       break;
+       }
+
+       if (!i)
+               return 0;
+
+       if (instr[i].type != INSTR_TX)
+               return 0;
+
+       if (data[i].n_users)
+               return 0;
+
+       i++;
+
+       *n_pattern_instr = i;
+       return 1;
+}
+
+static void
+instr_pattern_emit_many_tx_optimize(struct instruction *instr,
+                                   struct instruction_data *data,
+                                   uint32_t n_instr)
+{
+       uint32_t i;
+
+       /* Any emit instruction in addition to the first one. */
+       for (i = 1; i < n_instr - 1; i++) {
+               instr[0].type++;
+               instr[0].io.hdr.header_id[i] = instr[i].io.hdr.header_id[0];
+               instr[0].io.hdr.struct_id[i] = instr[i].io.hdr.struct_id[0];
+               instr[0].io.hdr.n_bytes[i] = instr[i].io.hdr.n_bytes[0];
+
+               data[i].invalid = 1;
+       }
+
+       /* The TX instruction is the last one in the pattern. */
+       instr[0].type++;
+       instr[0].io.io.offset = instr[i].io.io.offset;
+       instr[0].io.io.n_bits = instr[i].io.io.n_bits;
+       data[i].invalid = 1;
+}
+
+static int
+instr_pattern_dma_many_detect(struct instruction *instr,
+                             struct instruction_data *data,
+                             uint32_t n_instr,
+                             uint32_t *n_pattern_instr)
+{
+       uint32_t i;
+
+       for (i = 0; i < n_instr; i++) {
+               if (data[i].invalid)
+                       break;
+
+               if (instr[i].type != INSTR_DMA_HT)
+                       break;
+
+               if (i == RTE_DIM(instr->dma.dst.header_id))
+                       break;
+
+               if (i && data[i].n_users)
+                       break;
+       }
+
+       if (i < 2)
+               return 0;
+
+       *n_pattern_instr = i;
+       return 1;
+}
+
+static void
+instr_pattern_dma_many_optimize(struct instruction *instr,
+                               struct instruction_data *data,
+                               uint32_t n_instr)
+{
+       uint32_t i;
+
+       for (i = 1; i < n_instr; i++) {
+               instr[0].type++;
+               instr[0].dma.dst.header_id[i] = instr[i].dma.dst.header_id[0];
+               instr[0].dma.dst.struct_id[i] = instr[i].dma.dst.struct_id[0];
+               instr[0].dma.src.offset[i] = instr[i].dma.src.offset[0];
+               instr[0].dma.n_bytes[i] = instr[i].dma.n_bytes[0];
+
+               data[i].invalid = 1;
+       }
+}
+
+static uint32_t
+instr_optimize(struct instruction *instructions,
+              struct instruction_data *instruction_data,
+              uint32_t n_instructions)
+{
+       uint32_t i, pos = 0;
+
+       for (i = 0; i < n_instructions; ) {
+               struct instruction *instr = &instructions[i];
+               struct instruction_data *data = &instruction_data[i];
+               uint32_t n_instr = 0;
+               int detected;
+
+               /* Extract many. */
+               detected = instr_pattern_extract_many_detect(instr,
+                                                            data,
+                                                            n_instructions - i,
+                                                            &n_instr);
+               if (detected) {
+                       instr_pattern_extract_many_optimize(instr,
+                                                           data,
+                                                           n_instr);
+                       i += n_instr;
+                       continue;
+               }
+
+               /* Emit many + TX. */
+               detected = instr_pattern_emit_many_tx_detect(instr,
+                                                            data,
+                                                            n_instructions - i,
+                                                            &n_instr);
+               if (detected) {
+                       instr_pattern_emit_many_tx_optimize(instr,
+                                                           data,
+                                                           n_instr);
+                       i += n_instr;
+                       continue;
+               }
+
+               /* DMA many. */
+               detected = instr_pattern_dma_many_detect(instr,
+                                                        data,
+                                                        n_instructions - i,
+                                                        &n_instr);
+               if (detected) {
+                       instr_pattern_dma_many_optimize(instr, data, n_instr);
+                       i += n_instr;
+                       continue;
+               }
+
+               /* No pattern starting at the current instruction. */
+               i++;
+       }
+
+       /* Eliminate the invalid instructions that have been optimized out. */
        for (i = 0; i < n_instructions; i++) {
+               struct instruction *instr = &instructions[i];
                struct instruction_data *data = &instruction_data[i];
-               char *label = data->label;
 
-               data->n_users = label_is_used(instruction_data,
-                                             n_instructions,
-                                             label);
+               if (data->invalid)
+                       continue;
+
+               if (i != pos) {
+                       memcpy(&instructions[pos], instr, sizeof(*instr));
+                       memcpy(&instruction_data[pos], data, sizeof(*data));
+               }
+
+               pos++;
        }
 
-       return 0;
+       return pos;
 }
 
 static int
@@ -4166,14 +7025,13 @@ instruction_config(struct rte_swx_pipeline *p,
 {
        struct instruction *instr = NULL;
        struct instruction_data *data = NULL;
-       char *string = NULL;
        int err = 0;
        uint32_t i;
 
        CHECK(n_instructions, EINVAL);
        CHECK(instructions, EINVAL);
        for (i = 0; i < n_instructions; i++)
-               CHECK(instructions[i], EINVAL);
+               CHECK_INSTRUCTION(instructions[i], EINVAL);
 
        /* Memory allocation. */
        instr = calloc(n_instructions, sizeof(struct instruction));
@@ -4189,15 +7047,17 @@ instruction_config(struct rte_swx_pipeline *p,
        }
 
        for (i = 0; i < n_instructions; i++) {
-               string = strdup(instructions[i]);
+               char *string = strdup(instructions[i]);
                if (!string) {
                        err = ENOMEM;
                        goto error;
                }
 
                err = instr_translate(p, a, string, &instr[i], &data[i]);
-               if (err)
+               if (err) {
+                       free(string);
                        goto error;
+               }
 
                free(string);
        }
@@ -4206,7 +7066,15 @@ instruction_config(struct rte_swx_pipeline *p,
        if (err)
                goto error;
 
-       free(data);
+       err = instr_verify(p, a, instr, data, n_instructions);
+       if (err)
+               goto error;
+
+       n_instructions = instr_optimize(instr, data, n_instructions);
+
+       err = instr_jmp_resolve(instr, data, n_instructions);
+       if (err)
+               goto error;
 
        if (a) {
                a->instructions = instr;
@@ -4216,10 +7084,10 @@ instruction_config(struct rte_swx_pipeline *p,
                p->n_instructions = n_instructions;
        }
 
+       free(data);
        return 0;
 
 error:
-       free(string);
        free(data);
        free(instr);
        return err;
@@ -4303,6 +7171,80 @@ static instr_exec_t instruction_table[] = {
        [INSTR_ALU_SHL_HH] = instr_alu_shl_hh_exec,
        [INSTR_ALU_SHL_MI] = instr_alu_shl_mi_exec,
        [INSTR_ALU_SHL_HI] = instr_alu_shl_hi_exec,
+
+       [INSTR_ALU_SHR] = instr_alu_shr_exec,
+       [INSTR_ALU_SHR_MH] = instr_alu_shr_mh_exec,
+       [INSTR_ALU_SHR_HM] = instr_alu_shr_hm_exec,
+       [INSTR_ALU_SHR_HH] = instr_alu_shr_hh_exec,
+       [INSTR_ALU_SHR_MI] = instr_alu_shr_mi_exec,
+       [INSTR_ALU_SHR_HI] = instr_alu_shr_hi_exec,
+
+       [INSTR_REGPREFETCH_RH] = instr_regprefetch_rh_exec,
+       [INSTR_REGPREFETCH_RM] = instr_regprefetch_rm_exec,
+       [INSTR_REGPREFETCH_RI] = instr_regprefetch_ri_exec,
+
+       [INSTR_REGRD_HRH] = instr_regrd_hrh_exec,
+       [INSTR_REGRD_HRM] = instr_regrd_hrm_exec,
+       [INSTR_REGRD_MRH] = instr_regrd_mrh_exec,
+       [INSTR_REGRD_MRM] = instr_regrd_mrm_exec,
+       [INSTR_REGRD_HRI] = instr_regrd_hri_exec,
+       [INSTR_REGRD_MRI] = instr_regrd_mri_exec,
+
+       [INSTR_REGWR_RHH] = instr_regwr_rhh_exec,
+       [INSTR_REGWR_RHM] = instr_regwr_rhm_exec,
+       [INSTR_REGWR_RMH] = instr_regwr_rmh_exec,
+       [INSTR_REGWR_RMM] = instr_regwr_rmm_exec,
+       [INSTR_REGWR_RHI] = instr_regwr_rhi_exec,
+       [INSTR_REGWR_RMI] = instr_regwr_rmi_exec,
+       [INSTR_REGWR_RIH] = instr_regwr_rih_exec,
+       [INSTR_REGWR_RIM] = instr_regwr_rim_exec,
+       [INSTR_REGWR_RII] = instr_regwr_rii_exec,
+
+       [INSTR_REGADD_RHH] = instr_regadd_rhh_exec,
+       [INSTR_REGADD_RHM] = instr_regadd_rhm_exec,
+       [INSTR_REGADD_RMH] = instr_regadd_rmh_exec,
+       [INSTR_REGADD_RMM] = instr_regadd_rmm_exec,
+       [INSTR_REGADD_RHI] = instr_regadd_rhi_exec,
+       [INSTR_REGADD_RMI] = instr_regadd_rmi_exec,
+       [INSTR_REGADD_RIH] = instr_regadd_rih_exec,
+       [INSTR_REGADD_RIM] = instr_regadd_rim_exec,
+       [INSTR_REGADD_RII] = instr_regadd_rii_exec,
+
+       [INSTR_TABLE] = instr_table_exec,
+       [INSTR_EXTERN_OBJ] = instr_extern_obj_exec,
+       [INSTR_EXTERN_FUNC] = instr_extern_func_exec,
+
+       [INSTR_JMP] = instr_jmp_exec,
+       [INSTR_JMP_VALID] = instr_jmp_valid_exec,
+       [INSTR_JMP_INVALID] = instr_jmp_invalid_exec,
+       [INSTR_JMP_HIT] = instr_jmp_hit_exec,
+       [INSTR_JMP_MISS] = instr_jmp_miss_exec,
+       [INSTR_JMP_ACTION_HIT] = instr_jmp_action_hit_exec,
+       [INSTR_JMP_ACTION_MISS] = instr_jmp_action_miss_exec,
+
+       [INSTR_JMP_EQ] = instr_jmp_eq_exec,
+       [INSTR_JMP_EQ_S] = instr_jmp_eq_s_exec,
+       [INSTR_JMP_EQ_I] = instr_jmp_eq_i_exec,
+
+       [INSTR_JMP_NEQ] = instr_jmp_neq_exec,
+       [INSTR_JMP_NEQ_S] = instr_jmp_neq_s_exec,
+       [INSTR_JMP_NEQ_I] = instr_jmp_neq_i_exec,
+
+       [INSTR_JMP_LT] = instr_jmp_lt_exec,
+       [INSTR_JMP_LT_MH] = instr_jmp_lt_mh_exec,
+       [INSTR_JMP_LT_HM] = instr_jmp_lt_hm_exec,
+       [INSTR_JMP_LT_HH] = instr_jmp_lt_hh_exec,
+       [INSTR_JMP_LT_MI] = instr_jmp_lt_mi_exec,
+       [INSTR_JMP_LT_HI] = instr_jmp_lt_hi_exec,
+
+       [INSTR_JMP_GT] = instr_jmp_gt_exec,
+       [INSTR_JMP_GT_MH] = instr_jmp_gt_mh_exec,
+       [INSTR_JMP_GT_HM] = instr_jmp_gt_hm_exec,
+       [INSTR_JMP_GT_HH] = instr_jmp_gt_hh_exec,
+       [INSTR_JMP_GT_MI] = instr_jmp_gt_mi_exec,
+       [INSTR_JMP_GT_HI] = instr_jmp_gt_hi_exec,
+
+       [INSTR_RETURN] = instr_return_exec,
 };
 
 static inline void
@@ -4333,6 +7275,18 @@ action_find(struct rte_swx_pipeline *p, const char *name)
        return NULL;
 }
 
+static struct action *
+action_find_by_id(struct rte_swx_pipeline *p, uint32_t id)
+{
+       struct action *action = NULL;
+
+       TAILQ_FOREACH(action, &p->actions, node)
+               if (action->id == id)
+                       return action;
+
+       return NULL;
+}
+
 static struct field *
 action_field_find(struct action *a, const char *name)
 {
@@ -4612,7 +7566,7 @@ rte_swx_pipeline_table_config(struct rte_swx_pipeline *p,
                struct action *a;
                uint32_t action_data_size;
 
-               CHECK(action_name, EINVAL);
+               CHECK_NAME(action_name, EINVAL);
 
                a = action_find(p, action_name);
                CHECK(a, EINVAL);
@@ -4622,7 +7576,7 @@ rte_swx_pipeline_table_config(struct rte_swx_pipeline *p,
                        action_data_size_max = action_data_size;
        }
 
-       CHECK(params->default_action_name, EINVAL);
+       CHECK_NAME(params->default_action_name, EINVAL);
        for (i = 0; i < p->n_actions; i++)
                if (!strcmp(params->action_names[i],
                            params->default_action_name))
@@ -4633,6 +7587,9 @@ rte_swx_pipeline_table_config(struct rte_swx_pipeline *p,
              !params->default_action_data, EINVAL);
 
        /* Table type checks. */
+       if (recommended_table_type_name)
+               CHECK_NAME(recommended_table_type_name, EINVAL);
+
        if (params->n_fields) {
                enum rte_swx_table_match_type match_type;
 
@@ -4965,6 +7922,132 @@ table_free(struct rte_swx_pipeline *p)
        }
 }
 
+/*
+ * Register array.
+ */
+static struct regarray *
+regarray_find(struct rte_swx_pipeline *p, const char *name)
+{
+       struct regarray *elem;
+
+       TAILQ_FOREACH(elem, &p->regarrays, node)
+               if (!strcmp(elem->name, name))
+                       return elem;
+
+       return NULL;
+}
+
+static struct regarray *
+regarray_find_by_id(struct rte_swx_pipeline *p, uint32_t id)
+{
+       struct regarray *elem = NULL;
+
+       TAILQ_FOREACH(elem, &p->regarrays, node)
+               if (elem->id == id)
+                       return elem;
+
+       return NULL;
+}
+
+int
+rte_swx_pipeline_regarray_config(struct rte_swx_pipeline *p,
+                             const char *name,
+                             uint32_t size,
+                             uint64_t init_val)
+{
+       struct regarray *r;
+
+       CHECK(p, EINVAL);
+
+       CHECK_NAME(name, EINVAL);
+       CHECK(!regarray_find(p, name), EEXIST);
+
+       CHECK(size, EINVAL);
+       size = rte_align32pow2(size);
+
+       /* Memory allocation. */
+       r = calloc(1, sizeof(struct regarray));
+       CHECK(r, ENOMEM);
+
+       /* Node initialization. */
+       strcpy(r->name, name);
+       r->init_val = init_val;
+       r->size = size;
+       r->id = p->n_regarrays;
+
+       /* Node add to tailq. */
+       TAILQ_INSERT_TAIL(&p->regarrays, r, node);
+       p->n_regarrays++;
+
+       return 0;
+}
+
+static int
+regarray_build(struct rte_swx_pipeline *p)
+{
+       struct regarray *regarray;
+
+       if (!p->n_regarrays)
+               return 0;
+
+       p->regarray_runtime = calloc(p->n_regarrays, sizeof(struct regarray_runtime));
+       CHECK(p->regarray_runtime, ENOMEM);
+
+       TAILQ_FOREACH(regarray, &p->regarrays, node) {
+               struct regarray_runtime *r = &p->regarray_runtime[regarray->id];
+               uint32_t i;
+
+               r->regarray = env_malloc(regarray->size * sizeof(uint64_t),
+                                        RTE_CACHE_LINE_SIZE,
+                                        p->numa_node);
+               CHECK(r->regarray, ENOMEM);
+
+               if (regarray->init_val)
+                       for (i = 0; i < regarray->size; i++)
+                               r->regarray[i] = regarray->init_val;
+
+               r->size_mask = regarray->size - 1;
+       }
+
+       return 0;
+}
+
+static void
+regarray_build_free(struct rte_swx_pipeline *p)
+{
+       uint32_t i;
+
+       if (!p->regarray_runtime)
+               return;
+
+       for (i = 0; i < p->n_regarrays; i++) {
+               struct regarray *regarray = regarray_find_by_id(p, i);
+               struct regarray_runtime *r = &p->regarray_runtime[i];
+
+               env_free(r->regarray, regarray->size * sizeof(uint64_t));
+       }
+
+       free(p->regarray_runtime);
+       p->regarray_runtime = NULL;
+}
+
+static void
+regarray_free(struct rte_swx_pipeline *p)
+{
+       regarray_build_free(p);
+
+       for ( ; ; ) {
+               struct regarray *elem;
+
+               elem = TAILQ_FIRST(&p->regarrays);
+               if (!elem)
+                       break;
+
+               TAILQ_REMOVE(&p->regarrays, elem, node);
+               free(elem);
+       }
+}
+
 /*
  * Pipeline.
  */
@@ -4993,6 +8076,7 @@ rte_swx_pipeline_config(struct rte_swx_pipeline **p, int numa_node)
        TAILQ_INIT(&pipeline->actions);
        TAILQ_INIT(&pipeline->table_types);
        TAILQ_INIT(&pipeline->tables);
+       TAILQ_INIT(&pipeline->regarrays);
 
        pipeline->n_structs = 1; /* Struct 0 is reserved for action_data. */
        pipeline->numa_node = numa_node;
@@ -5009,6 +8093,7 @@ rte_swx_pipeline_free(struct rte_swx_pipeline *p)
 
        free(p->instructions);
 
+       regarray_free(p);
        table_state_free(p);
        table_free(p);
        action_free(p);
@@ -5093,10 +8178,15 @@ rte_swx_pipeline_build(struct rte_swx_pipeline *p)
        if (status)
                goto error;
 
+       status = regarray_build(p);
+       if (status)
+               goto error;
+
        p->build_done = 1;
        return 0;
 
 error:
+       regarray_build_free(p);
        table_state_build_free(p);
        table_build_free(p);
        action_build_free(p);
@@ -5120,9 +8210,194 @@ rte_swx_pipeline_run(struct rte_swx_pipeline *p, uint32_t n_instructions)
                instr_exec(p);
 }
 
+void
+rte_swx_pipeline_flush(struct rte_swx_pipeline *p)
+{
+       uint32_t i;
+
+       for (i = 0; i < p->n_ports_out; i++) {
+               struct port_out_runtime *port = &p->out[i];
+
+               if (port->flush)
+                       port->flush(port->obj);
+       }
+}
+
 /*
  * Control.
  */
+int
+rte_swx_ctl_pipeline_info_get(struct rte_swx_pipeline *p,
+                             struct rte_swx_ctl_pipeline_info *pipeline)
+{
+       struct action *action;
+       struct table *table;
+       uint32_t n_actions = 0, n_tables = 0;
+
+       if (!p || !pipeline)
+               return -EINVAL;
+
+       TAILQ_FOREACH(action, &p->actions, node)
+               n_actions++;
+
+       TAILQ_FOREACH(table, &p->tables, node)
+               n_tables++;
+
+       pipeline->n_ports_in = p->n_ports_in;
+       pipeline->n_ports_out = p->n_ports_out;
+       pipeline->n_actions = n_actions;
+       pipeline->n_tables = n_tables;
+       pipeline->n_regarrays = p->n_regarrays;
+
+       return 0;
+}
+
+int
+rte_swx_ctl_pipeline_numa_node_get(struct rte_swx_pipeline *p, int *numa_node)
+{
+       if (!p || !numa_node)
+               return -EINVAL;
+
+       *numa_node = p->numa_node;
+       return 0;
+}
+
+int
+rte_swx_ctl_action_info_get(struct rte_swx_pipeline *p,
+                           uint32_t action_id,
+                           struct rte_swx_ctl_action_info *action)
+{
+       struct action *a = NULL;
+
+       if (!p || (action_id >= p->n_actions) || !action)
+               return -EINVAL;
+
+       a = action_find_by_id(p, action_id);
+       if (!a)
+               return -EINVAL;
+
+       strcpy(action->name, a->name);
+       action->n_args = a->st ? a->st->n_fields : 0;
+       return 0;
+}
+
+int
+rte_swx_ctl_action_arg_info_get(struct rte_swx_pipeline *p,
+                               uint32_t action_id,
+                               uint32_t action_arg_id,
+                               struct rte_swx_ctl_action_arg_info *action_arg)
+{
+       struct action *a = NULL;
+       struct field *arg = NULL;
+
+       if (!p || (action_id >= p->n_actions) || !action_arg)
+               return -EINVAL;
+
+       a = action_find_by_id(p, action_id);
+       if (!a || !a->st || (action_arg_id >= a->st->n_fields))
+               return -EINVAL;
+
+       arg = &a->st->fields[action_arg_id];
+       strcpy(action_arg->name, arg->name);
+       action_arg->n_bits = arg->n_bits;
+
+       return 0;
+}
+
+int
+rte_swx_ctl_table_info_get(struct rte_swx_pipeline *p,
+                          uint32_t table_id,
+                          struct rte_swx_ctl_table_info *table)
+{
+       struct table *t = NULL;
+
+       if (!p || !table)
+               return -EINVAL;
+
+       t = table_find_by_id(p, table_id);
+       if (!t)
+               return -EINVAL;
+
+       strcpy(table->name, t->name);
+       strcpy(table->args, t->args);
+       table->n_match_fields = t->n_fields;
+       table->n_actions = t->n_actions;
+       table->default_action_is_const = t->default_action_is_const;
+       table->size = t->size;
+       return 0;
+}
+
+int
+rte_swx_ctl_table_match_field_info_get(struct rte_swx_pipeline *p,
+       uint32_t table_id,
+       uint32_t match_field_id,
+       struct rte_swx_ctl_table_match_field_info *match_field)
+{
+       struct table *t;
+       struct match_field *f;
+
+       if (!p || (table_id >= p->n_tables) || !match_field)
+               return -EINVAL;
+
+       t = table_find_by_id(p, table_id);
+       if (!t || (match_field_id >= t->n_fields))
+               return -EINVAL;
+
+       f = &t->fields[match_field_id];
+       match_field->match_type = f->match_type;
+       match_field->is_header = t->is_header;
+       match_field->n_bits = f->field->n_bits;
+       match_field->offset = f->field->offset;
+
+       return 0;
+}
+
+int
+rte_swx_ctl_table_action_info_get(struct rte_swx_pipeline *p,
+       uint32_t table_id,
+       uint32_t table_action_id,
+       struct rte_swx_ctl_table_action_info *table_action)
+{
+       struct table *t;
+
+       if (!p || (table_id >= p->n_tables) || !table_action)
+               return -EINVAL;
+
+       t = table_find_by_id(p, table_id);
+       if (!t || (table_action_id >= t->n_actions))
+               return -EINVAL;
+
+       table_action->action_id = t->actions[table_action_id]->id;
+
+       return 0;
+}
+
+int
+rte_swx_ctl_table_ops_get(struct rte_swx_pipeline *p,
+                         uint32_t table_id,
+                         struct rte_swx_table_ops *table_ops,
+                         int *is_stub)
+{
+       struct table *t;
+
+       if (!p || (table_id >= p->n_tables))
+               return -EINVAL;
+
+       t = table_find_by_id(p, table_id);
+       if (!t)
+               return -EINVAL;
+
+       if (t->type) {
+               if (table_ops)
+                       memcpy(table_ops, &t->type->ops, sizeof(*table_ops));
+               *is_stub = 0;
+       } else {
+               *is_stub = 1;
+       }
+
+       return 0;
+}
+
 int
 rte_swx_pipeline_table_state_get(struct rte_swx_pipeline *p,
                                 struct rte_swx_table_state **table_state)
@@ -5144,3 +8419,100 @@ rte_swx_pipeline_table_state_set(struct rte_swx_pipeline *p,
        p->table_state = table_state;
        return 0;
 }
+
+int
+rte_swx_ctl_pipeline_port_in_stats_read(struct rte_swx_pipeline *p,
+                                       uint32_t port_id,
+                                       struct rte_swx_port_in_stats *stats)
+{
+       struct port_in *port;
+
+       if (!p || !stats)
+               return -EINVAL;
+
+       port = port_in_find(p, port_id);
+       if (!port)
+               return -EINVAL;
+
+       port->type->ops.stats_read(port->obj, stats);
+       return 0;
+}
+
+int
+rte_swx_ctl_pipeline_port_out_stats_read(struct rte_swx_pipeline *p,
+                                        uint32_t port_id,
+                                        struct rte_swx_port_out_stats *stats)
+{
+       struct port_out *port;
+
+       if (!p || !stats)
+               return -EINVAL;
+
+       port = port_out_find(p, port_id);
+       if (!port)
+               return -EINVAL;
+
+       port->type->ops.stats_read(port->obj, stats);
+       return 0;
+}
+
+int
+rte_swx_ctl_regarray_info_get(struct rte_swx_pipeline *p,
+                             uint32_t regarray_id,
+                             struct rte_swx_ctl_regarray_info *regarray)
+{
+       struct regarray *r;
+
+       if (!p || !regarray)
+               return -EINVAL;
+
+       r = regarray_find_by_id(p, regarray_id);
+       if (!r)
+               return -EINVAL;
+
+       strcpy(regarray->name, r->name);
+       regarray->size = r->size;
+       return 0;
+}
+
+int
+rte_swx_ctl_pipeline_regarray_read(struct rte_swx_pipeline *p,
+                                  const char *regarray_name,
+                                  uint32_t regarray_index,
+                                  uint64_t *value)
+{
+       struct regarray *regarray;
+       struct regarray_runtime *r;
+
+       if (!p || !regarray_name || !value)
+               return -EINVAL;
+
+       regarray = regarray_find(p, regarray_name);
+       if (!regarray || (regarray_index >= regarray->size))
+               return -EINVAL;
+
+       r = &p->regarray_runtime[regarray->id];
+       *value = r->regarray[regarray_index];
+       return 0;
+}
+
+int
+rte_swx_ctl_pipeline_regarray_write(struct rte_swx_pipeline *p,
+                                  const char *regarray_name,
+                                  uint32_t regarray_index,
+                                  uint64_t value)
+{
+       struct regarray *regarray;
+       struct regarray_runtime *r;
+
+       if (!p || !regarray_name)
+               return -EINVAL;
+
+       regarray = regarray_find(p, regarray_name);
+       if (!regarray || (regarray_index >= regarray->size))
+               return -EINVAL;
+
+       r = &p->regarray_runtime[regarray->id];
+       r->regarray[regarray_index] = value;
+       return 0;
+}