xref: /dpdk/lib/lpm/rte_lpm_altivec.h (revision 64fcadeac0f70c8c9c215196a3d580b4ad37dcdc)
199a2dd95SBruce Richardson /*
299a2dd95SBruce Richardson  * SPDX-License-Identifier: BSD-3-Clause
399a2dd95SBruce Richardson  * Copyright (C) IBM Corporation 2016.
499a2dd95SBruce Richardson  */
599a2dd95SBruce Richardson 
699a2dd95SBruce Richardson #ifndef _RTE_LPM_ALTIVEC_H_
799a2dd95SBruce Richardson #define _RTE_LPM_ALTIVEC_H_
899a2dd95SBruce Richardson 
999a2dd95SBruce Richardson #include <rte_branch_prediction.h>
1099a2dd95SBruce Richardson #include <rte_byteorder.h>
1199a2dd95SBruce Richardson #include <rte_common.h>
1299a2dd95SBruce Richardson #include <rte_vect.h>
1399a2dd95SBruce Richardson 
1499a2dd95SBruce Richardson #ifdef __cplusplus
1599a2dd95SBruce Richardson extern "C" {
1699a2dd95SBruce Richardson #endif
1799a2dd95SBruce Richardson 
1899a2dd95SBruce Richardson static inline void
rte_lpm_lookupx4(const struct rte_lpm * lpm,xmm_t ip,uint32_t hop[4],uint32_t defv)1999a2dd95SBruce Richardson rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
2099a2dd95SBruce Richardson 	uint32_t defv)
2199a2dd95SBruce Richardson {
22*64fcadeaSThomas Monjalon 	xmm_t i24;
2399a2dd95SBruce Richardson 	rte_xmm_t i8;
2499a2dd95SBruce Richardson 	uint32_t tbl[4];
2599a2dd95SBruce Richardson 	uint64_t idx, pt, pt2;
2699a2dd95SBruce Richardson 	const uint32_t *ptbl;
2799a2dd95SBruce Richardson 
2899a2dd95SBruce Richardson 	const uint32_t mask = UINT8_MAX;
29*64fcadeaSThomas Monjalon 	const xmm_t mask8 = (xmm_t){mask, mask, mask, mask};
3099a2dd95SBruce Richardson 
3199a2dd95SBruce Richardson 	/*
3299a2dd95SBruce Richardson 	 * RTE_LPM_VALID_EXT_ENTRY_BITMASK for 2 LPM entries
3399a2dd95SBruce Richardson 	 * as one 64-bit value (0x0300000003000000).
3499a2dd95SBruce Richardson 	 */
3599a2dd95SBruce Richardson 	const uint64_t mask_xv =
3699a2dd95SBruce Richardson 		((uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK |
3799a2dd95SBruce Richardson 		(uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 32);
3899a2dd95SBruce Richardson 
3999a2dd95SBruce Richardson 	/*
4099a2dd95SBruce Richardson 	 * RTE_LPM_LOOKUP_SUCCESS for 2 LPM entries
4199a2dd95SBruce Richardson 	 * as one 64-bit value (0x0100000001000000).
4299a2dd95SBruce Richardson 	 */
4399a2dd95SBruce Richardson 	const uint64_t mask_v =
4499a2dd95SBruce Richardson 		((uint64_t)RTE_LPM_LOOKUP_SUCCESS |
4599a2dd95SBruce Richardson 		(uint64_t)RTE_LPM_LOOKUP_SUCCESS << 32);
4699a2dd95SBruce Richardson 
4799a2dd95SBruce Richardson 	/* get 4 indexes for tbl24[]. */
4899a2dd95SBruce Richardson 	i24 = vec_sr((xmm_t) ip,
49*64fcadeaSThomas Monjalon 		(__vector unsigned int){CHAR_BIT, CHAR_BIT, CHAR_BIT, CHAR_BIT});
5099a2dd95SBruce Richardson 
5199a2dd95SBruce Richardson 	/* extract values from tbl24[] */
5299a2dd95SBruce Richardson 	idx = (uint32_t)i24[0];
5399a2dd95SBruce Richardson 	idx = idx < (1<<24) ? idx : (1<<24)-1;
5499a2dd95SBruce Richardson 	ptbl = (const uint32_t *)&lpm->tbl24[idx];
5599a2dd95SBruce Richardson 	tbl[0] = *ptbl;
5699a2dd95SBruce Richardson 
5799a2dd95SBruce Richardson 	idx = (uint32_t) i24[1];
5899a2dd95SBruce Richardson 	idx = idx < (1<<24) ? idx : (1<<24)-1;
5999a2dd95SBruce Richardson 	ptbl = (const uint32_t *)&lpm->tbl24[idx];
6099a2dd95SBruce Richardson 	tbl[1] = *ptbl;
6199a2dd95SBruce Richardson 
6299a2dd95SBruce Richardson 	idx = (uint32_t) i24[2];
6399a2dd95SBruce Richardson 	idx = idx < (1<<24) ? idx : (1<<24)-1;
6499a2dd95SBruce Richardson 	ptbl = (const uint32_t *)&lpm->tbl24[idx];
6599a2dd95SBruce Richardson 	tbl[2] = *ptbl;
6699a2dd95SBruce Richardson 
6799a2dd95SBruce Richardson 	idx = (uint32_t) i24[3];
6899a2dd95SBruce Richardson 	idx = idx < (1<<24) ? idx : (1<<24)-1;
6999a2dd95SBruce Richardson 	ptbl = (const uint32_t *)&lpm->tbl24[idx];
7099a2dd95SBruce Richardson 	tbl[3] = *ptbl;
7199a2dd95SBruce Richardson 
7299a2dd95SBruce Richardson 	/* get 4 indexes for tbl8[]. */
7399a2dd95SBruce Richardson 	i8.x = vec_and(ip, mask8);
7499a2dd95SBruce Richardson 
7599a2dd95SBruce Richardson 	pt = (uint64_t)tbl[0] |
7699a2dd95SBruce Richardson 		(uint64_t)tbl[1] << 32;
7799a2dd95SBruce Richardson 	pt2 = (uint64_t)tbl[2] |
7899a2dd95SBruce Richardson 		(uint64_t)tbl[3] << 32;
7999a2dd95SBruce Richardson 
8099a2dd95SBruce Richardson 	/* search successfully finished for all 4 IP addresses. */
8199a2dd95SBruce Richardson 	if (likely((pt & mask_xv) == mask_v) &&
8299a2dd95SBruce Richardson 			likely((pt2 & mask_xv) == mask_v)) {
8399a2dd95SBruce Richardson 		*(uint64_t *)hop = pt & RTE_LPM_MASKX4_RES;
8499a2dd95SBruce Richardson 		*(uint64_t *)(hop + 2) = pt2 & RTE_LPM_MASKX4_RES;
8599a2dd95SBruce Richardson 		return;
8699a2dd95SBruce Richardson 	}
8799a2dd95SBruce Richardson 
8899a2dd95SBruce Richardson 	if (unlikely((pt & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
8999a2dd95SBruce Richardson 			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
9099a2dd95SBruce Richardson 		i8.u32[0] = i8.u32[0] +
9199a2dd95SBruce Richardson 			(tbl[0] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
9299a2dd95SBruce Richardson 		ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[0]];
9399a2dd95SBruce Richardson 		tbl[0] = *ptbl;
9499a2dd95SBruce Richardson 	}
9599a2dd95SBruce Richardson 	if (unlikely((pt >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
9699a2dd95SBruce Richardson 			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
9799a2dd95SBruce Richardson 		i8.u32[1] = i8.u32[1] +
9899a2dd95SBruce Richardson 			(tbl[1] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
9999a2dd95SBruce Richardson 		ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[1]];
10099a2dd95SBruce Richardson 		tbl[1] = *ptbl;
10199a2dd95SBruce Richardson 	}
10299a2dd95SBruce Richardson 	if (unlikely((pt2 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
10399a2dd95SBruce Richardson 			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
10499a2dd95SBruce Richardson 		i8.u32[2] = i8.u32[2] +
10599a2dd95SBruce Richardson 			(tbl[2] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
10699a2dd95SBruce Richardson 		ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[2]];
10799a2dd95SBruce Richardson 		tbl[2] = *ptbl;
10899a2dd95SBruce Richardson 	}
10999a2dd95SBruce Richardson 	if (unlikely((pt2 >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
11099a2dd95SBruce Richardson 			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
11199a2dd95SBruce Richardson 		i8.u32[3] = i8.u32[3] +
11299a2dd95SBruce Richardson 			(tbl[3] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
11399a2dd95SBruce Richardson 		ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[3]];
11499a2dd95SBruce Richardson 		tbl[3] = *ptbl;
11599a2dd95SBruce Richardson 	}
11699a2dd95SBruce Richardson 
11799a2dd95SBruce Richardson 	hop[0] = (tbl[0] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[0] & 0x00FFFFFF : defv;
11899a2dd95SBruce Richardson 	hop[1] = (tbl[1] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[1] & 0x00FFFFFF : defv;
11999a2dd95SBruce Richardson 	hop[2] = (tbl[2] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[2] & 0x00FFFFFF : defv;
12099a2dd95SBruce Richardson 	hop[3] = (tbl[3] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[3] & 0x00FFFFFF : defv;
12199a2dd95SBruce Richardson }
12299a2dd95SBruce Richardson 
12399a2dd95SBruce Richardson #ifdef __cplusplus
12499a2dd95SBruce Richardson }
12599a2dd95SBruce Richardson #endif
12699a2dd95SBruce Richardson 
12799a2dd95SBruce Richardson #endif /* _RTE_LPM_ALTIVEC_H_ */
128