lpm: add AltiVec for ppc64
authorGowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
Thu, 8 Sep 2016 16:48:03 +0000 (22:18 +0530)
committerThomas Monjalon <thomas.monjalon@6wind.com>
Fri, 9 Sep 2016 15:56:08 +0000 (17:56 +0200)
This patch adds ppc64le port for LPM library in DPDK.

Signed-off-by: Gowrishankar Muthukrishnan <gowrishankar.m@linux.vnet.ibm.com>
Acked-by: Chao Zhu <chaozhu@linux.vnet.ibm.com>
app/test/test_xmmt_ops.h
config/defconfig_ppc_64-power8-linuxapp-gcc
lib/librte_eal/common/include/arch/ppc_64/rte_vect.h [new file with mode: 0644]
lib/librte_lpm/Makefile
lib/librte_lpm/rte_lpm.h
lib/librte_lpm/rte_lpm_altivec.h [new file with mode: 0644]

index de9c16f..42174d2 100644 (file)
@@ -62,6 +62,22 @@ vect_set_epi32(int i3, int i2, int i1, int i0)
 /* sets the 4 signed 32-bit integer values and returns the xmm_t variable */
 #define vect_set_epi32(i3, i2, i1, i0) _mm_set_epi32(i3, i2, i1, i0)
 
+#elif defined(RTE_ARCH_PPC_64)
+
+/* vect_* abstraction implementation using ALTIVEC */
+
+/* loads the xmm_t value from address p(does not need to be 16-byte aligned)*/
+#define vect_loadu_sil128(p) vec_ld(0, p)
+
+/* sets the 4 signed 32-bit integer values and returns the xmm_t variable */
+static inline xmm_t  __attribute__((always_inline))
+vect_set_epi32(int i3, int i2, int i1, int i0)
+{
+       xmm_t data = (xmm_t){i0, i1, i2, i3};
+
+       return data;
+}
+
 #endif
 
 #endif /* _TEST_XMMT_OPS_H_ */
index bef8f49..9ddf3c5 100644 (file)
@@ -57,7 +57,6 @@ CONFIG_RTE_LIBRTE_ENIC_PMD=n
 CONFIG_RTE_LIBRTE_FM10K_PMD=n
 
 # This following libraries are not available on Power. So they're turned off.
-CONFIG_RTE_LIBRTE_LPM=n
 CONFIG_RTE_LIBRTE_ACL=n
 CONFIG_RTE_LIBRTE_SCHED=n
 CONFIG_RTE_LIBRTE_PORT=n
diff --git a/lib/librte_eal/common/include/arch/ppc_64/rte_vect.h b/lib/librte_eal/common/include/arch/ppc_64/rte_vect.h
new file mode 100644 (file)
index 0000000..05209e5
--- /dev/null
@@ -0,0 +1,60 @@
+/*
+ *   BSD LICENSE
+ *
+ *   Copyright (C) IBM Corporation 2016.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of IBM Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef _RTE_VECT_PPC_64_H_
+#define _RTE_VECT_PPC_64_H_
+
+#include <altivec.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef vector signed int xmm_t;
+
+#define        XMM_SIZE        (sizeof(xmm_t))
+#define        XMM_MASK        (XMM_SIZE - 1)
+
+typedef union rte_xmm {
+       xmm_t    x;
+       uint8_t  u8[XMM_SIZE / sizeof(uint8_t)];
+       uint16_t u16[XMM_SIZE / sizeof(uint16_t)];
+       uint32_t u32[XMM_SIZE / sizeof(uint32_t)];
+       uint64_t u64[XMM_SIZE / sizeof(uint64_t)];
+       double   pd[XMM_SIZE / sizeof(double)];
+} __attribute__((aligned(16))) rte_xmm_t;
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_VECT_PPC_64_H_ */
index 656ade2..3dc549d 100644 (file)
@@ -51,6 +51,8 @@ ifneq ($(filter y,$(CONFIG_RTE_ARCH_ARM) $(CONFIG_RTE_ARCH_ARM64)),)
 SYMLINK-$(CONFIG_RTE_LIBRTE_LPM)-include += rte_lpm_neon.h
 else ifeq ($(CONFIG_RTE_ARCH_X86),y)
 SYMLINK-$(CONFIG_RTE_LIBRTE_LPM)-include += rte_lpm_sse.h
+else ifeq ($(CONFIG_RTE_ARCH_PPC_64),y)
+SYMLINK-$(CONFIG_RTE_LIBRTE_LPM)-include += rte_lpm_altivec.h
 endif
 
 # this lib needs eal
index 2df1d67..dbe5483 100644 (file)
@@ -480,6 +480,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
 
 #if defined(RTE_ARCH_ARM) || defined(RTE_ARCH_ARM64)
 #include "rte_lpm_neon.h"
+#elif defined(RTE_ARCH_PPC_64)
+#include "rte_lpm_altivec.h"
 #else
 #include "rte_lpm_sse.h"
 #endif
diff --git a/lib/librte_lpm/rte_lpm_altivec.h b/lib/librte_lpm/rte_lpm_altivec.h
new file mode 100644 (file)
index 0000000..e26e087
--- /dev/null
@@ -0,0 +1,154 @@
+/*
+ *   BSD LICENSE
+ *
+ *   Copyright (C) IBM Corporation 2016.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of IBM Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef _RTE_LPM_ALTIVEC_H_
+#define _RTE_LPM_ALTIVEC_H_
+
+#include <rte_branch_prediction.h>
+#include <rte_byteorder.h>
+#include <rte_common.h>
+#include <rte_vect.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static inline void
+rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
+       uint32_t defv)
+{
+       vector signed int i24;
+       rte_xmm_t i8;
+       uint32_t tbl[4];
+       uint64_t idx, pt, pt2;
+       const uint32_t *ptbl;
+
+       const uint32_t mask = UINT8_MAX;
+       const vector signed int mask8 = (xmm_t){mask, mask, mask, mask};
+
+       /*
+        * RTE_LPM_VALID_EXT_ENTRY_BITMASK for 2 LPM entries
+        * as one 64-bit value (0x0300000003000000).
+        */
+       const uint64_t mask_xv =
+               ((uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK |
+               (uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 32);
+
+       /*
+        * RTE_LPM_LOOKUP_SUCCESS for 2 LPM entries
+        * as one 64-bit value (0x0100000001000000).
+        */
+       const uint64_t mask_v =
+               ((uint64_t)RTE_LPM_LOOKUP_SUCCESS |
+               (uint64_t)RTE_LPM_LOOKUP_SUCCESS << 32);
+
+       /* get 4 indexes for tbl24[]. */
+       i24 = vec_sr((xmm_t) ip,
+               (vector unsigned int){CHAR_BIT, CHAR_BIT, CHAR_BIT, CHAR_BIT});
+
+       /* extract values from tbl24[] */
+       idx = (uint32_t)i24[0];
+       idx = idx < (1<<24) ? idx : (1<<24)-1;
+       ptbl = (const uint32_t *)&lpm->tbl24[idx];
+       tbl[0] = *ptbl;
+
+       idx = (uint32_t) i24[1];
+       idx = idx < (1<<24) ? idx : (1<<24)-1;
+       ptbl = (const uint32_t *)&lpm->tbl24[idx];
+       tbl[1] = *ptbl;
+
+       idx = (uint32_t) i24[2];
+       idx = idx < (1<<24) ? idx : (1<<24)-1;
+       ptbl = (const uint32_t *)&lpm->tbl24[idx];
+       tbl[2] = *ptbl;
+
+       idx = (uint32_t) i24[3];
+       idx = idx < (1<<24) ? idx : (1<<24)-1;
+       ptbl = (const uint32_t *)&lpm->tbl24[idx];
+       tbl[3] = *ptbl;
+
+       /* get 4 indexes for tbl8[]. */
+       i8.x = vec_and(ip, mask8);
+
+       pt = (uint64_t)tbl[0] |
+               (uint64_t)tbl[1] << 32;
+       pt2 = (uint64_t)tbl[2] |
+               (uint64_t)tbl[3] << 32;
+
+       /* search successfully finished for all 4 IP addresses. */
+       if (likely((pt & mask_xv) == mask_v) &&
+                       likely((pt2 & mask_xv) == mask_v)) {
+               *(uint64_t *)hop = pt & RTE_LPM_MASKX4_RES;
+               *(uint64_t *)(hop + 2) = pt2 & RTE_LPM_MASKX4_RES;
+               return;
+       }
+
+       if (unlikely((pt & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+                       RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+               i8.u32[0] = i8.u32[0] +
+                       (uint8_t)tbl[0] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+               ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[0]];
+               tbl[0] = *ptbl;
+       }
+       if (unlikely((pt >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+                       RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+               i8.u32[1] = i8.u32[1] +
+                       (uint8_t)tbl[1] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+               ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[1]];
+               tbl[1] = *ptbl;
+       }
+       if (unlikely((pt2 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+                       RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+               i8.u32[2] = i8.u32[2] +
+                       (uint8_t)tbl[2] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+               ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[2]];
+               tbl[2] = *ptbl;
+       }
+       if (unlikely((pt2 >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+                       RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+               i8.u32[3] = i8.u32[3] +
+                       (uint8_t)tbl[3] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+               ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[3]];
+               tbl[3] = *ptbl;
+       }
+
+       hop[0] = (tbl[0] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[0] & 0x00FFFFFF : defv;
+       hop[1] = (tbl[1] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[1] & 0x00FFFFFF : defv;
+       hop[2] = (tbl[2] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[2] & 0x00FFFFFF : defv;
+       hop[3] = (tbl[3] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[3] & 0x00FFFFFF : defv;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_LPM_ALTIVEC_H_ */