[PATCH 1/3] config/riscv: detect V extension

[PATCH 1/3] config/riscv: detect V extension

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch is derived from "config/riscv: detect presence of Zbc
extension with modifications".

The RISC-V C api defines architecture extension test macros
These let us detect whether the V extension is supported on the
compiler and -march we're building with. The C api also defines V
intrinsics we can use rather than inline assembly on newer versions of
GCC (14.1.0+) and Clang (18.1.0+).

If the V extension and intrinsics are both present and we can detect
the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index 7562c6cb99..e3694cf2e6 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -119,6 +119,31 @@ foreach flag: arch_config['machine_args']
     endif
 endforeach
 
+# check if we can do buildtime detection of extensions supported by the target
+riscv_extension_macros = false
+if (cc.get_define('__riscv_arch_test', args: machine_args) == '1')
+  message('Detected architecture extension test macros')
+  riscv_extension_macros = true
+else
+  warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible')
+endif
+
+# detect extensions
+# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+
+if (riscv_extension_macros and
+    (cc.get_define('__riscv_vector', args: machine_args) != ''))
+  if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0'))
+      or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0')))
+    if (cc.compiles('''#include <riscv_vector.h>
+        int main(void) { size_t vl = __riscv_vsetvl_e32m1(1); }''', args: machine_args))
+      message('Compiling with the V extension')
+      machine_args += ['-DRTE_RISCV_FEATURE_V']
+    endif
+  else
+    warning('Detected V extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)')
+  endif
+endif
+
 # apply flags
 foreach flag: dpdk_flags
     if flag.length() > 0
-- 
2.49.0

[PATCH v2 0/3] Add RISC-V V extension detection and LPM optimization

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch series adds support for the RISC-V Vector (V) extension and provides
an optimized implementation of `rte_lpm_lookupx4` using RVV. It includes runtime
detection of the V extension, conditional compilation based on compiler support
for RVV intrinsics.

Test results using lpm_perf_autotest on BPI-F3:
    scalar: 5.7 cycles
    rvv:    2.4 cycles

### v2:
- Improved commit messages.

sunyuechi (3):
  config/riscv: detect V extension
  lib/lpm: R-V V rte_lpm_lookupx4
  riscv: override machine_args only when default

 MAINTAINERS              |  2 +
 config/riscv/meson.build | 27 ++++++++++++
 lib/lpm/meson.build      |  1 +
 lib/lpm/rte_lpm.h        |  2 +
 lib/lpm/rte_lpm_rvv.h    | 91 ++++++++++++++++++++++++++++++++++++++++
 5 files changed, 123 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

--
2.49.0

[PATCH v2 1/3] config/riscv: detect V extension

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch is derived from "config/riscv: detect presence of Zbc
extension with modifications".

The RISC-V C api defines architecture extension test macros
These let us detect whether the V extension is supported on the
compiler and -march we're building with. The C api also defines V
intrinsics we can use rather than inline assembly on newer versions of
GCC (14.1.0+) and Clang (18.1.0+).

If the V extension and intrinsics are both present and we can detect
the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index 7562c6cb99..e3694cf2e6 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -119,6 +119,31 @@ foreach flag: arch_config['machine_args']
     endif
 endforeach
 
+# check if we can do buildtime detection of extensions supported by the target
+riscv_extension_macros = false
+if (cc.get_define('__riscv_arch_test', args: machine_args) == '1')
+  message('Detected architecture extension test macros')
+  riscv_extension_macros = true
+else
+  warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible')
+endif
+
+# detect extensions
+# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+
+if (riscv_extension_macros and
+    (cc.get_define('__riscv_vector', args: machine_args) != ''))
+  if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0'))
+      or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0')))
+    if (cc.compiles('''#include <riscv_vector.h>
+        int main(void) { size_t vl = __riscv_vsetvl_e32m1(1); }''', args: machine_args))
+      message('Compiling with the V extension')
+      machine_args += ['-DRTE_RISCV_FEATURE_V']
+    endif
+  else
+    warning('Detected V extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)')
+  endif
+endif
+
 # apply flags
 foreach flag: dpdk_flags
     if flag.length() > 0
-- 
2.49.0

[PATCH v2 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

Test results using lpm_perf_autotest on BPI-F3:
    scalar: 5.7 cycles
    rvv:    2.4 cycles

The best way to call this RVV function is to follow the approach used in
lib/fib, where all architectures initialize a function pointer in a
unified way. However, other architectures in lib/lpm do not follow this
pattern. Therefore, to avoid affecting other architectures, this commit
does not modify lib/lpm/rte_lpm.c.

Unifying the code style between lpm and fib may be worth considering
in the future.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 MAINTAINERS           |  2 +
 lib/lpm/meson.build   |  1 +
 lib/lpm/rte_lpm.h     |  2 +
 lib/lpm/rte_lpm_rvv.h | 91 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 96 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 3e16789250..0f207ac129 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
 F: config/riscv/
 F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
 F: lib/eal/riscv/
+M: sunyuechi <sunyuechi@iscas.ac.cn>
+F: lib/**/*rvv*
 
 Intel x86
 M: Bruce Richardson <bruce.richardson@intel.com>
diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
index fae4f79fb9..09133061e5 100644
--- a/lib/lpm/meson.build
+++ b/lib/lpm/meson.build
@@ -17,6 +17,7 @@ indirect_headers += files(
         'rte_lpm_scalar.h',
         'rte_lpm_sse.h',
         'rte_lpm_sve.h',
+        'rte_lpm_rvv.h',
 )
 deps += ['hash']
 deps += ['rcu']
diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
index 7df64f06b1..b06517206f 100644
--- a/lib/lpm/rte_lpm.h
+++ b/lib/lpm/rte_lpm.h
@@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
 #include "rte_lpm_altivec.h"
 #elif defined(RTE_ARCH_X86)
 #include "rte_lpm_sse.h"
+#elif defined(RTE_ARCH_RISCV) && defined(RTE_RISCV_FEATURE_V)
+#include "rte_lpm_rvv.h"
 #else
 #include "rte_lpm_scalar.h"
 #endif
diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
new file mode 100644
index 0000000000..e39ade3f07
--- /dev/null
+++ b/lib/lpm/rte_lpm_rvv.h
@@ -0,0 +1,91 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
+ */
+
+#ifndef _RTE_LPM_RVV_H_
+#define _RTE_LPM_RVV_H_
+
+#include <rte_vect.h>
+
+#include <rte_cpuflags.h>
+#include <riscv_vector.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
+#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
+
+typedef void (*lpm_lookupx4_fn)(const struct rte_lpm *, xmm_t, uint32_t *, uint32_t);
+
+static lpm_lookupx4_fn lpm_lookupx4_impl;
+
+static inline void rte_lpm_lookupx4_scalar(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	uint32_t nh;
+	int ret;
+
+	for (int i = 0; i < 4; i++) {
+		ret = rte_lpm_lookup(lpm, ip[i], &nh);
+		hop[i] = (ret == 0) ? nh : defv;
+	}
+}
+
+static inline void rte_lpm_lookupx4_rvv(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	size_t vl = 4;
+
+	const uint32_t *tbl24_p = (const uint32_t *)lpm->tbl24;
+	uint32_t tbl_entries[4] = {
+		tbl24_p[((uint32_t)ip[0]) >> 8],
+		tbl24_p[((uint32_t)ip[1]) >> 8],
+		tbl24_p[((uint32_t)ip[2]) >> 8],
+		tbl24_p[((uint32_t)ip[3]) >> 8],
+	};
+	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
+
+	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
+	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
+
+	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
+	    __riscv_vadd_vv_u32m1(
+		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
+		__riscv_vand_vx_u32m1(
+		    __riscv_vle32_v_u32m1((const uint32_t *)&ip, vl), 0x000000FF, vl),
+		vl),
+	    2, vl);
+
+	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
+	    mask, vtbl_entry, (const uint32_t *)(lpm->tbl8), vtbl8_index, vl);
+
+	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
+	mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
+
+	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
+
+	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
+}
+
+static inline void rte_lpm_lookupx4(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	lpm_lookupx4_impl(lpm, ip, hop, defv);
+}
+
+RTE_INIT(rte_lpm_init_alg)
+{
+	lpm_lookupx4_impl = rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_ISA_V)
+	    ? rte_lpm_lookupx4_rvv
+	    : rte_lpm_lookupx4_scalar;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_LPM_RVV_H_ */
-- 
2.49.0

[PATCH v2 3/3] riscv: override machine_args only when default

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

Support using -Dcpu_instruction_set=rv64gcv to enable V extension.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index e3694cf2e6..1036a86d05 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -111,6 +111,7 @@ arch_config = arch_config[arch_id]
 # Concatenate flags respecting priorities.
 dpdk_flags = flags_common + vendor_config['flags'] + arch_config.get('flags', [])
 
+if (cpu_instruction_set == 'rv64gc')
 # apply supported machine args
 machine_args = [] # Clear previous machine args
 foreach flag: arch_config['machine_args']
@@ -118,6 +119,7 @@ foreach flag: arch_config['machine_args']
         machine_args += flag
     endif
 endforeach
+endif
 
 # check if we can do buildtime detection of extensions supported by the target
 riscv_extension_macros = false
-- 
2.49.0

[PATCH v2 0/3] Add RISC-V V extension detection and LPM optimization

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch series adds support for the RISC-V Vector (V) extension and provides
an optimized implementation of `rte_lpm_lookupx4` using RVV. It includes runtime
detection of the V extension, conditional compilation based on compiler support
for RVV intrinsics.

Test results using lpm_perf_autotest on BPI-F3:
    scalar: 5.7 cycles
    rvv:    2.4 cycles

### v2:
- Improved commit messages.

sunyuechi (3):
  config/riscv: detect V extension
  lib/lpm: R-V V rte_lpm_lookupx4
  riscv: override machine_args only when default

 MAINTAINERS              |  2 +
 config/riscv/meson.build | 27 ++++++++++++
 lib/lpm/meson.build      |  1 +
 lib/lpm/rte_lpm.h        |  2 +
 lib/lpm/rte_lpm_rvv.h    | 91 ++++++++++++++++++++++++++++++++++++++++
 5 files changed, 123 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

--
2.49.0

[PATCH v2 1/3] config/riscv: detect V extension

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch is derived from "config/riscv: detect presence of Zbc
extension with modifications".

The RISC-V C api defines architecture extension test macros
These let us detect whether the V extension is supported on the
compiler and -march we're building with. The C api also defines V
intrinsics we can use rather than inline assembly on newer versions of
GCC (14.1.0+) and Clang (18.1.0+).

If the V extension and intrinsics are both present and we can detect
the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index 7562c6cb99..e3694cf2e6 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -119,6 +119,31 @@ foreach flag: arch_config['machine_args']
     endif
 endforeach
 
+# check if we can do buildtime detection of extensions supported by the target
+riscv_extension_macros = false
+if (cc.get_define('__riscv_arch_test', args: machine_args) == '1')
+  message('Detected architecture extension test macros')
+  riscv_extension_macros = true
+else
+  warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible')
+endif
+
+# detect extensions
+# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+
+if (riscv_extension_macros and
+    (cc.get_define('__riscv_vector', args: machine_args) != ''))
+  if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0'))
+      or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0')))
+    if (cc.compiles('''#include <riscv_vector.h>
+        int main(void) { size_t vl = __riscv_vsetvl_e32m1(1); }''', args: machine_args))
+      message('Compiling with the V extension')
+      machine_args += ['-DRTE_RISCV_FEATURE_V']
+    endif
+  else
+    warning('Detected V extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)')
+  endif
+endif
+
 # apply flags
 foreach flag: dpdk_flags
     if flag.length() > 0
-- 
2.49.0

[PATCH v2 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

Test results using lpm_perf_autotest on BPI-F3:
    scalar: 5.7 cycles
    rvv:    2.4 cycles

The best way to call this RVV function is to follow the approach used in
lib/fib, where all architectures initialize a function pointer in a
unified way. However, other architectures in lib/lpm do not follow this
pattern. Therefore, to avoid affecting other architectures, this commit
does not modify lib/lpm/rte_lpm.c.

Unifying the code style between lpm and fib may be worth considering
in the future.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 MAINTAINERS           |  2 +
 lib/lpm/meson.build   |  1 +
 lib/lpm/rte_lpm.h     |  2 +
 lib/lpm/rte_lpm_rvv.h | 91 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 96 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 3e16789250..0f207ac129 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
 F: config/riscv/
 F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
 F: lib/eal/riscv/
+M: sunyuechi <sunyuechi@iscas.ac.cn>
+F: lib/**/*rvv*
 
 Intel x86
 M: Bruce Richardson <bruce.richardson@intel.com>
diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
index fae4f79fb9..09133061e5 100644
--- a/lib/lpm/meson.build
+++ b/lib/lpm/meson.build
@@ -17,6 +17,7 @@ indirect_headers += files(
         'rte_lpm_scalar.h',
         'rte_lpm_sse.h',
         'rte_lpm_sve.h',
+        'rte_lpm_rvv.h',
 )
 deps += ['hash']
 deps += ['rcu']
diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
index 7df64f06b1..b06517206f 100644
--- a/lib/lpm/rte_lpm.h
+++ b/lib/lpm/rte_lpm.h
@@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
 #include "rte_lpm_altivec.h"
 #elif defined(RTE_ARCH_X86)
 #include "rte_lpm_sse.h"
+#elif defined(RTE_ARCH_RISCV) && defined(RTE_RISCV_FEATURE_V)
+#include "rte_lpm_rvv.h"
 #else
 #include "rte_lpm_scalar.h"
 #endif
diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
new file mode 100644
index 0000000000..e39ade3f07
--- /dev/null
+++ b/lib/lpm/rte_lpm_rvv.h
@@ -0,0 +1,91 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
+ */
+
+#ifndef _RTE_LPM_RVV_H_
+#define _RTE_LPM_RVV_H_
+
+#include <rte_vect.h>
+
+#include <rte_cpuflags.h>
+#include <riscv_vector.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
+#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
+
+typedef void (*lpm_lookupx4_fn)(const struct rte_lpm *, xmm_t, uint32_t *, uint32_t);
+
+static lpm_lookupx4_fn lpm_lookupx4_impl;
+
+static inline void rte_lpm_lookupx4_scalar(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	uint32_t nh;
+	int ret;
+
+	for (int i = 0; i < 4; i++) {
+		ret = rte_lpm_lookup(lpm, ip[i], &nh);
+		hop[i] = (ret == 0) ? nh : defv;
+	}
+}
+
+static inline void rte_lpm_lookupx4_rvv(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	size_t vl = 4;
+
+	const uint32_t *tbl24_p = (const uint32_t *)lpm->tbl24;
+	uint32_t tbl_entries[4] = {
+		tbl24_p[((uint32_t)ip[0]) >> 8],
+		tbl24_p[((uint32_t)ip[1]) >> 8],
+		tbl24_p[((uint32_t)ip[2]) >> 8],
+		tbl24_p[((uint32_t)ip[3]) >> 8],
+	};
+	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
+
+	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
+	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
+
+	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
+	    __riscv_vadd_vv_u32m1(
+		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
+		__riscv_vand_vx_u32m1(
+		    __riscv_vle32_v_u32m1((const uint32_t *)&ip, vl), 0x000000FF, vl),
+		vl),
+	    2, vl);
+
+	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
+	    mask, vtbl_entry, (const uint32_t *)(lpm->tbl8), vtbl8_index, vl);
+
+	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
+	mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
+
+	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
+
+	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
+}
+
+static inline void rte_lpm_lookupx4(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	lpm_lookupx4_impl(lpm, ip, hop, defv);
+}
+
+RTE_INIT(rte_lpm_init_alg)
+{
+	lpm_lookupx4_impl = rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_ISA_V)
+	    ? rte_lpm_lookupx4_rvv
+	    : rte_lpm_lookupx4_scalar;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_LPM_RVV_H_ */
-- 
2.49.0

[PATCH v2 3/3] riscv: override machine_args only when default

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

Support using -Dcpu_instruction_set=rv64gcv to enable V extension.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index e3694cf2e6..1036a86d05 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -111,6 +111,7 @@ arch_config = arch_config[arch_id]
 # Concatenate flags respecting priorities.
 dpdk_flags = flags_common + vendor_config['flags'] + arch_config.get('flags', [])
 
+if (cpu_instruction_set == 'rv64gc')
 # apply supported machine args
 machine_args = [] # Clear previous machine args
 foreach flag: arch_config['machine_args']
@@ -118,6 +119,7 @@ foreach flag: arch_config['machine_args']
         machine_args += flag
     endif
 endforeach
+endif
 
 # check if we can do buildtime detection of extensions supported by the target
 riscv_extension_macros = false
-- 
2.49.0

Re: [PATCH v2 0/3] Add RISC-V V extension detection and LPM optimization

[孙越池]

The previous v2 patch incorrectly used --in-reply-to, which caused a warning on Patchwork. This update only corrects the Reply-To field.


> -----原始邮件-----
> 发件人: uk7b@foxmail.com
> 发送时间: 2025-06-04 21:07:33 (星期三)
> 收件人: dev@dpdk.org
&gt; 抄送: sunyuechi <sunyuechi@iscas.ac.cn>
&gt; 主题: [PATCH v2 0/3] Add RISC-V V extension detection and LPM optimization
&gt; 
&gt; From: sunyuechi <sunyuechi@iscas.ac.cn>
&gt; 
&gt; This patch series adds support for the RISC-V Vector (V) extension and provides
&gt; an optimized implementation of `rte_lpm_lookupx4` using RVV. It includes runtime
&gt; detection of the V extension, conditional compilation based on compiler support
&gt; for RVV intrinsics.
&gt; 
&gt; Test results using lpm_perf_autotest on BPI-F3:
&gt;     scalar: 5.7 cycles
&gt;     rvv:    2.4 cycles
&gt; 
&gt; ### v2:
&gt; - Improved commit messages.
&gt; 
&gt; sunyuechi (3):
&gt;   config/riscv: detect V extension
&gt;   lib/lpm: R-V V rte_lpm_lookupx4
&gt;   riscv: override machine_args only when default
&gt; 
&gt;  MAINTAINERS              |  2 +
&gt;  config/riscv/meson.build | 27 ++++++++++++
&gt;  lib/lpm/meson.build      |  1 +
&gt;  lib/lpm/rte_lpm.h        |  2 +
&gt;  lib/lpm/rte_lpm_rvv.h    | 91 ++++++++++++++++++++++++++++++++++++++++
&gt;  5 files changed, 123 insertions(+)
&gt;  create mode 100644 lib/lpm/rte_lpm_rvv.h
&gt; 
&gt; --
&gt; 2.49.0
</sunyuechi@iscas.ac.cn></sunyuechi@iscas.ac.cn>

[PATCH v3 0/3] Add RISC-V V extension detection and LPM optimization

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch series adds support for the RISC-V Vector (V) extension and provides
an optimized implementation of `rte_lpm_lookupx4` using RVV. It includes runtime
detection of the V extension, conditional compilation based on compiler support
for RVV intrinsics.

Test results using lpm_perf_autotest on BPI-F3:
    scalar: 5.7 cycles
    rvv:    2.4 cycles

v3:
- Due to previous SMTP server restrictions, patches could not be sent as a proper series.
This version re-sends the same patches as a series. No code changes.
v2:
- Improved commit messages.

sunyuechi (3):
  config/riscv: detect V extension
  lib/lpm: R-V V rte_lpm_lookupx4
  riscv: override machine_args only when default

 MAINTAINERS              |  2 +
 config/riscv/meson.build | 27 ++++++++++++
 lib/lpm/meson.build      |  1 +
 lib/lpm/rte_lpm.h        |  2 +
 lib/lpm/rte_lpm_rvv.h    | 91 ++++++++++++++++++++++++++++++++++++++++
 5 files changed, 123 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

--
2.49.0

[PATCH v3 1/3] config/riscv: detect V extension

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

This patch is derived from "config/riscv: detect presence of Zbc
extension with modifications".

The RISC-V C api defines architecture extension test macros
These let us detect whether the V extension is supported on the
compiler and -march we're building with. The C api also defines V
intrinsics we can use rather than inline assembly on newer versions of
GCC (14.1.0+) and Clang (18.1.0+).

If the V extension and intrinsics are both present and we can detect
the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index 7562c6cb99..e3694cf2e6 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -119,6 +119,31 @@ foreach flag: arch_config['machine_args']
     endif
 endforeach
 
+# check if we can do buildtime detection of extensions supported by the target
+riscv_extension_macros = false
+if (cc.get_define('__riscv_arch_test', args: machine_args) == '1')
+  message('Detected architecture extension test macros')
+  riscv_extension_macros = true
+else
+  warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible')
+endif
+
+# detect extensions
+# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+
+if (riscv_extension_macros and
+    (cc.get_define('__riscv_vector', args: machine_args) != ''))
+  if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0'))
+      or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0')))
+    if (cc.compiles('''#include <riscv_vector.h>
+        int main(void) { size_t vl = __riscv_vsetvl_e32m1(1); }''', args: machine_args))
+      message('Compiling with the V extension')
+      machine_args += ['-DRTE_RISCV_FEATURE_V']
+    endif
+  else
+    warning('Detected V extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)')
+  endif
+endif
+
 # apply flags
 foreach flag: dpdk_flags
     if flag.length() > 0
-- 
2.49.0

[PATCH v3 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

Test results using lpm_perf_autotest on BPI-F3:
    scalar: 5.7 cycles
    rvv:    2.4 cycles

The best way to call this RVV function is to follow the approach used in
lib/fib, where all architectures initialize a function pointer in a
unified way. However, other architectures in lib/lpm do not follow this
pattern. Therefore, to avoid affecting other architectures, this commit
does not modify lib/lpm/rte_lpm.c.

Unifying the code style between lpm and fib may be worth considering
in the future.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 MAINTAINERS           |  2 +
 lib/lpm/meson.build   |  1 +
 lib/lpm/rte_lpm.h     |  2 +
 lib/lpm/rte_lpm_rvv.h | 91 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 96 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 3e16789250..0f207ac129 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
 F: config/riscv/
 F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
 F: lib/eal/riscv/
+M: sunyuechi <sunyuechi@iscas.ac.cn>
+F: lib/**/*rvv*
 
 Intel x86
 M: Bruce Richardson <bruce.richardson@intel.com>
diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
index fae4f79fb9..09133061e5 100644
--- a/lib/lpm/meson.build
+++ b/lib/lpm/meson.build
@@ -17,6 +17,7 @@ indirect_headers += files(
         'rte_lpm_scalar.h',
         'rte_lpm_sse.h',
         'rte_lpm_sve.h',
+        'rte_lpm_rvv.h',
 )
 deps += ['hash']
 deps += ['rcu']
diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
index 7df64f06b1..b06517206f 100644
--- a/lib/lpm/rte_lpm.h
+++ b/lib/lpm/rte_lpm.h
@@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
 #include "rte_lpm_altivec.h"
 #elif defined(RTE_ARCH_X86)
 #include "rte_lpm_sse.h"
+#elif defined(RTE_ARCH_RISCV) && defined(RTE_RISCV_FEATURE_V)
+#include "rte_lpm_rvv.h"
 #else
 #include "rte_lpm_scalar.h"
 #endif
diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
new file mode 100644
index 0000000000..e39ade3f07
--- /dev/null
+++ b/lib/lpm/rte_lpm_rvv.h
@@ -0,0 +1,91 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
+ */
+
+#ifndef _RTE_LPM_RVV_H_
+#define _RTE_LPM_RVV_H_
+
+#include <rte_vect.h>
+
+#include <rte_cpuflags.h>
+#include <riscv_vector.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
+#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
+
+typedef void (*lpm_lookupx4_fn)(const struct rte_lpm *, xmm_t, uint32_t *, uint32_t);
+
+static lpm_lookupx4_fn lpm_lookupx4_impl;
+
+static inline void rte_lpm_lookupx4_scalar(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	uint32_t nh;
+	int ret;
+
+	for (int i = 0; i < 4; i++) {
+		ret = rte_lpm_lookup(lpm, ip[i], &nh);
+		hop[i] = (ret == 0) ? nh : defv;
+	}
+}
+
+static inline void rte_lpm_lookupx4_rvv(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	size_t vl = 4;
+
+	const uint32_t *tbl24_p = (const uint32_t *)lpm->tbl24;
+	uint32_t tbl_entries[4] = {
+		tbl24_p[((uint32_t)ip[0]) >> 8],
+		tbl24_p[((uint32_t)ip[1]) >> 8],
+		tbl24_p[((uint32_t)ip[2]) >> 8],
+		tbl24_p[((uint32_t)ip[3]) >> 8],
+	};
+	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
+
+	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
+	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
+
+	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
+	    __riscv_vadd_vv_u32m1(
+		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
+		__riscv_vand_vx_u32m1(
+		    __riscv_vle32_v_u32m1((const uint32_t *)&ip, vl), 0x000000FF, vl),
+		vl),
+	    2, vl);
+
+	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
+	    mask, vtbl_entry, (const uint32_t *)(lpm->tbl8), vtbl8_index, vl);
+
+	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
+	mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
+
+	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
+
+	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
+}
+
+static inline void rte_lpm_lookupx4(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	lpm_lookupx4_impl(lpm, ip, hop, defv);
+}
+
+RTE_INIT(rte_lpm_init_alg)
+{
+	lpm_lookupx4_impl = rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_ISA_V)
+	    ? rte_lpm_lookupx4_rvv
+	    : rte_lpm_lookupx4_scalar;
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_LPM_RVV_H_ */
-- 
2.49.0

[PATCH v3 3/3] riscv: override machine_args only when default

[uk7b]

From: sunyuechi <sunyuechi@iscas.ac.cn>

Support using -Dcpu_instruction_set=rv64gcv to enable V extension.

Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index e3694cf2e6..1036a86d05 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -111,6 +111,7 @@ arch_config = arch_config[arch_id]
 # Concatenate flags respecting priorities.
 dpdk_flags = flags_common + vendor_config['flags'] + arch_config.get('flags', [])
 
+if (cpu_instruction_set == 'rv64gc')
 # apply supported machine args
 machine_args = [] # Clear previous machine args
 foreach flag: arch_config['machine_args']
@@ -118,6 +119,7 @@ foreach flag: arch_config['machine_args']
         machine_args += flag
     endif
 endforeach
+endif
 
 # check if we can do buildtime detection of extensions supported by the target
 riscv_extension_macros = false
-- 
2.49.0

Re: [PATCH v2 1/3] config/riscv: detect V extension

[Stephen Hemminger]

On Wed,  4 Jun 2025 19:49:16 +0800
uk7b@foxmail.com wrote:

> From: sunyuechi <sunyuechi@iscas.ac.cn>
> 
> This patch is derived from "config/riscv: detect presence of Zbc
> extension with modifications".
> 
> The RISC-V C api defines architecture extension test macros
> These let us detect whether the V extension is supported on the
> compiler and -march we're building with. The C api also defines V
> intrinsics we can use rather than inline assembly on newer versions of
> GCC (14.1.0+) and Clang (18.1.0+).
> 
> If the V extension and intrinsics are both present and we can detect
> the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.
> 
> Signed-off-by: sunyuechi <sunyuechi@iscas.ac.cn>


Need entry in .mailmap as first patch, we keep track of contributions.

Since Signed-off-by has legal meaning you need to use full legal name.
By giving Signed-off-by you are saying that "yes, I have full legal right
to open source this".

[PATCH v4 0/3] [PATCH v4 0/3] Add RISC-V V extension detection and LPM optimization

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

This patch series adds support for the RISC-V Vector (V) extension and 
provides an optimized implementation of `rte_lpm_lookupx4` using RVV.

- Test: app/test/lpm_perf_autotest
- Platform: Banana Pi BPI-F3
- SoC: Spacemit X60 (8 cores with Vector extension)
- CPU Frequency: up to 1.6 GHz
- Cache: 256 KiB L1d ×8, 256 KiB L1i ×8, 1 MiB L2 ×2
- Memory: 16 GiB
- Kernel: Linux 6.6.36
- Compiler: GCC 14.2.0 (with RVV intrinsic support)

Test results(LPM LookupX4):
    scalar: 5.7 cycles
    rvv:    4.6 cycles

v4:
- Keep the LPM code consistent, use a static inline function instead of runtime detection.
- Update the commit message.
- Update the .mailmap file.
v3:
- Due to previous SMTP server restrictions, patches could not be sent as a proper series.
This version re-sends the same patches as a series. No code changes.
v2:
- Improved commit messages.

Sun Yuechi (3):
  config/riscv: detect V extension
  lib/lpm: R-V V rte_lpm_lookupx4
  riscv: override machine_args only when default

 .mailmap                 |  1 +
 MAINTAINERS              |  2 ++
 config/riscv/meson.build | 27 +++++++++++++++++
 lib/lpm/meson.build      |  1 +
 lib/lpm/rte_lpm.h        |  2 ++
 lib/lpm/rte_lpm_rvv.h    | 62 ++++++++++++++++++++++++++++++++++++++++
 6 files changed, 95 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

-- 
2.49.0

[PATCH v4 1/3] config/riscv: detect V extension

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

This patch is derived from "config/riscv: detect presence of Zbc
extension with modifications".

The RISC-V C api defines architecture extension test macros
These let us detect whether the V extension is supported on the
compiler and -march we're building with. The C api also defines V
intrinsics we can use rather than inline assembly on newer versions of
GCC (14.1.0+) and Clang (18.1.0+).

If the V extension and intrinsics are both present and we can detect
the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.

Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
---
 .mailmap                 |  1 +
 config/riscv/meson.build | 25 +++++++++++++++++++++++++
 2 files changed, 26 insertions(+)

diff --git a/.mailmap b/.mailmap
index c65872cd9f..b635eb645a 100644
--- a/.mailmap
+++ b/.mailmap
@@ -1508,6 +1508,7 @@ Sunil Kumar Kori <skori@marvell.com> <skori@mavell.com> <sunil.kori@nxp.com>
 Sunil Pai G <sunil.pai.g@intel.com>
 Sunil Uttarwar <sunilprakashrao.uttarwar@amd.com>
 Sun Jiajia <sunx.jiajia@intel.com>
+Sun Yuechi <sunyuechi@iscas.ac.cn> <uk7b@foxmail.com>
 Sunyang Wu <sunyang.wu@jaguarmicro.com>
 Surabhi Boob <surabhi.boob@intel.com>
 Suyang Ju <sju@paloaltonetworks.com>
diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index 7562c6cb99..e3694cf2e6 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -119,6 +119,31 @@ foreach flag: arch_config['machine_args']
     endif
 endforeach
 
+# check if we can do buildtime detection of extensions supported by the target
+riscv_extension_macros = false
+if (cc.get_define('__riscv_arch_test', args: machine_args) == '1')
+  message('Detected architecture extension test macros')
+  riscv_extension_macros = true
+else
+  warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible')
+endif
+
+# detect extensions
+# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+
+if (riscv_extension_macros and
+    (cc.get_define('__riscv_vector', args: machine_args) != ''))
+  if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0'))
+      or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0')))
+    if (cc.compiles('''#include <riscv_vector.h>
+        int main(void) { size_t vl = __riscv_vsetvl_e32m1(1); }''', args: machine_args))
+      message('Compiling with the V extension')
+      machine_args += ['-DRTE_RISCV_FEATURE_V']
+    endif
+  else
+    warning('Detected V extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)')
+  endif
+endif
+
 # apply flags
 foreach flag: dpdk_flags
     if flag.length() > 0
-- 
2.49.0

[PATCH v4 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

The initialization of vtbl_entry is not fully vectorized here because
doing so would require __riscv_vluxei32_v_u32m1, which is slower
than the scalar approach in this small-scale scenario.

- Test: app/test/lpm_perf_autotest
- Platform: Banana Pi(BPI-F3)
- SoC: Spacemit X60 (8 cores with Vector extension)
- CPU Frequency: up to 1.6 GHz
- Cache: 256 KiB L1d ×8, 256 KiB L1i ×8, 1 MiB L2 ×2
- Memory: 16 GiB
- Kernel: Linux 6.6.36
- Compiler: GCC 14.2.0 (with RVV intrinsic support)

Test results(LPM LookupX4):
    scalar: 5.7 cycles
    rvv:    4.6 cycles

Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
---
 MAINTAINERS           |  2 ++
 lib/lpm/meson.build   |  1 +
 lib/lpm/rte_lpm.h     |  2 ++
 lib/lpm/rte_lpm_rvv.h | 62 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 67 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 3e16789250..0f207ac129 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
 F: config/riscv/
 F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
 F: lib/eal/riscv/
+M: sunyuechi <sunyuechi@iscas.ac.cn>
+F: lib/**/*rvv*
 
 Intel x86
 M: Bruce Richardson <bruce.richardson@intel.com>
diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
index fae4f79fb9..09133061e5 100644
--- a/lib/lpm/meson.build
+++ b/lib/lpm/meson.build
@@ -17,6 +17,7 @@ indirect_headers += files(
         'rte_lpm_scalar.h',
         'rte_lpm_sse.h',
         'rte_lpm_sve.h',
+        'rte_lpm_rvv.h',
 )
 deps += ['hash']
 deps += ['rcu']
diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
index 7df64f06b1..b06517206f 100644
--- a/lib/lpm/rte_lpm.h
+++ b/lib/lpm/rte_lpm.h
@@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
 #include "rte_lpm_altivec.h"
 #elif defined(RTE_ARCH_X86)
 #include "rte_lpm_sse.h"
+#elif defined(RTE_ARCH_RISCV) && defined(RTE_RISCV_FEATURE_V)
+#include "rte_lpm_rvv.h"
 #else
 #include "rte_lpm_scalar.h"
 #endif
diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
new file mode 100644
index 0000000000..5f48fb2b32
--- /dev/null
+++ b/lib/lpm/rte_lpm_rvv.h
@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
+ */
+
+#ifndef _RTE_LPM_RVV_H_
+#define _RTE_LPM_RVV_H_
+
+#include <rte_vect.h>
+
+#include <rte_cpuflags.h>
+#include <riscv_vector.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
+#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
+
+static inline void rte_lpm_lookupx4(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	size_t vl = 4;
+
+	const uint32_t *tbl24_p = (const uint32_t *)lpm->tbl24;
+	uint32_t tbl_entries[4] = {
+		tbl24_p[((uint32_t)ip[0]) >> 8],
+		tbl24_p[((uint32_t)ip[1]) >> 8],
+		tbl24_p[((uint32_t)ip[2]) >> 8],
+		tbl24_p[((uint32_t)ip[3]) >> 8],
+	};
+	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
+
+	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
+	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
+
+	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
+	    __riscv_vadd_vv_u32m1(
+		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
+		__riscv_vand_vx_u32m1(
+		    __riscv_vle32_v_u32m1((const uint32_t *)&ip, vl), 0x000000FF, vl),
+		vl),
+	    2, vl);
+
+	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
+	    mask, vtbl_entry, (const uint32_t *)(lpm->tbl8), vtbl8_index, vl);
+
+	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
+	mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
+
+	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
+
+	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_LPM_RVV_H_ */
-- 
2.49.0

[PATCH v4 3/3] riscv: override machine_args only when default

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

Support using -Dcpu_instruction_set=rv64gcv to enable V extension.

Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index e3694cf2e6..1036a86d05 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -111,6 +111,7 @@ arch_config = arch_config[arch_id]
 # Concatenate flags respecting priorities.
 dpdk_flags = flags_common + vendor_config['flags'] + arch_config.get('flags', [])
 
+if (cpu_instruction_set == 'rv64gc')
 # apply supported machine args
 machine_args = [] # Clear previous machine args
 foreach flag: arch_config['machine_args']
@@ -118,6 +119,7 @@ foreach flag: arch_config['machine_args']
         machine_args += flag
     endif
 endforeach
+endif
 
 # check if we can do buildtime detection of extensions supported by the target
 riscv_extension_macros = false
-- 
2.49.0

Re: [PATCH v4 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[Medvedkin, Vladimir]

Hi Sun,

You did not address my previous comments regarding commit message. You 
can put everything you've wrote in this commit as a note and add 
meaningful description about what commit generally does, like (please 
correct if needed):

"Implement LPM lookupx4 routine for RISC-V architecture using RISC-V 
Vector Extension instruction set"

Everything else (performance tests, implementation thoughts and 
considerations, etc.) should be in the patch notes. For more information 
on what "patch notes" are, you may want refer to Git documentation [1].

[1] https://git-scm.com/docs/git-notes

On 05/06/2025 11:58, uk7b@foxmail.com wrote:

> From: Sun Yuechi <sunyuechi@iscas.ac.cn>
>
> The initialization of vtbl_entry is not fully vectorized here because
> doing so would require __riscv_vluxei32_v_u32m1, which is slower
> than the scalar approach in this small-scale scenario.
>
> - Test: app/test/lpm_perf_autotest
> - Platform: Banana Pi(BPI-F3)
> - SoC: Spacemit X60 (8 cores with Vector extension)
> - CPU Frequency: up to 1.6 GHz
> - Cache: 256 KiB L1d ×8, 256 KiB L1i ×8, 1 MiB L2 ×2
> - Memory: 16 GiB
> - Kernel: Linux 6.6.36
> - Compiler: GCC 14.2.0 (with RVV intrinsic support)
>
> Test results(LPM LookupX4):
>      scalar: 5.7 cycles
>      rvv:    4.6 cycles
>
> Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
> ---
>   MAINTAINERS           |  2 ++
>   lib/lpm/meson.build   |  1 +
>   lib/lpm/rte_lpm.h     |  2 ++
>   lib/lpm/rte_lpm_rvv.h | 62 +++++++++++++++++++++++++++++++++++++++++++
>   4 files changed, 67 insertions(+)
>   create mode 100644 lib/lpm/rte_lpm_rvv.h
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 3e16789250..0f207ac129 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
>   F: config/riscv/
>   F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
>   F: lib/eal/riscv/
> +M: sunyuechi <sunyuechi@iscas.ac.cn>
> +F: lib/**/*rvv*
>   
>   Intel x86
>   M: Bruce Richardson <bruce.richardson@intel.com>
> diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
> index fae4f79fb9..09133061e5 100644
> --- a/lib/lpm/meson.build
> +++ b/lib/lpm/meson.build
> @@ -17,6 +17,7 @@ indirect_headers += files(
>           'rte_lpm_scalar.h',
>           'rte_lpm_sse.h',
>           'rte_lpm_sve.h',
> +        'rte_lpm_rvv.h',
>   )
>   deps += ['hash']
>   deps += ['rcu']
> diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
> index 7df64f06b1..b06517206f 100644
> --- a/lib/lpm/rte_lpm.h
> +++ b/lib/lpm/rte_lpm.h
> @@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
>   #include "rte_lpm_altivec.h"
>   #elif defined(RTE_ARCH_X86)
>   #include "rte_lpm_sse.h"
> +#elif defined(RTE_ARCH_RISCV) && defined(RTE_RISCV_FEATURE_V)
> +#include "rte_lpm_rvv.h"
>   #else
>   #include "rte_lpm_scalar.h"
>   #endif
> diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
> new file mode 100644
> index 0000000000..5f48fb2b32
> --- /dev/null
> +++ b/lib/lpm/rte_lpm_rvv.h
> @@ -0,0 +1,62 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
> + */
> +
> +#ifndef _RTE_LPM_RVV_H_
> +#define _RTE_LPM_RVV_H_
> +
> +#include <rte_vect.h>
> +
> +#include <rte_cpuflags.h>
> +#include <riscv_vector.h>
> +
> +#ifdef __cplusplus
> +extern "C" {
> +#endif
> +
> +#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
> +#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
> +
> +static inline void rte_lpm_lookupx4(
> +	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
> +{
> +	size_t vl = 4;
> +
> +	const uint32_t *tbl24_p = (const uint32_t *)lpm->tbl24;
> +	uint32_t tbl_entries[4] = {
> +		tbl24_p[((uint32_t)ip[0]) >> 8],
> +		tbl24_p[((uint32_t)ip[1]) >> 8],
> +		tbl24_p[((uint32_t)ip[2]) >> 8],
> +		tbl24_p[((uint32_t)ip[3]) >> 8],
> +	};
> +	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
> +
> +	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
> +	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
> +	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
> +
> +	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
> +	    __riscv_vadd_vv_u32m1(
> +		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
> +		__riscv_vand_vx_u32m1(
> +		    __riscv_vle32_v_u32m1((const uint32_t *)&ip, vl), 0x000000FF, vl),
> +		vl),
> +	    2, vl);
> +
> +	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
> +	    mask, vtbl_entry, (const uint32_t *)(lpm->tbl8), vtbl8_index, vl);
> +
> +	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
> +	mask = __riscv_vmseq_vx_u32m1_b32(
> +	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
> +
> +	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
> +
> +	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
> +}
> +
> +#ifdef __cplusplus
> +}
> +#endif
> +
> +#endif /* _RTE_LPM_RVV_H_ */

-- 
Regards,
Vladimir

[PATCH v5 0/3] Add RISC-V V extension detection and LPM optimization

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

This patch series adds support for the RISC-V Vector (V) extension and
provides an optimized implementation of `rte_lpm_lookupx4` using RVV.

The initialization of vtbl_entry is not fully vectorized here because
doing so would require __riscv_vluxei32_v_u32m1, which is slower
than the scalar approach in this small-scale scenario.

- Test: app/test/lpm_perf_autotest
- Platform: Banana Pi BPI-F3
- SoC: Spacemit X60 (8 cores with Vector extension)
- CPU Frequency: up to 1.6 GHz
- Cache: 256 KiB L1d ×8, 256 KiB L1i ×8, 1 MiB L2 ×2
- Memory: 16 GiB
- Kernel: Linux 6.6.36
- Compiler: GCC 14.2.0 (with RVV intrinsic support)

Test results(LPM LookupX4):
    scalar: 5.7 cycles
    rvv:    4.6 cycles

v5:
- Updated the commit message and moved test details and thoughts to the cover letter.
v4:
- Keep the LPM code consistent, use a static inline function instead of runtime detection.
- Update the commit message.
- Update the .mailmap file.
v3:
- Due to previous SMTP server restrictions, patches could not be sent as a proper series.
This version re-sends the same patches as a series. No code changes.
v2:
- Improved commit messages.

Sun Yuechi (3):
  config/riscv: detect V extension
  lib/lpm: R-V V rte_lpm_lookupx4
  riscv: override machine_args only when default

 .mailmap                 |  1 +
 MAINTAINERS              |  2 ++
 config/riscv/meson.build | 27 +++++++++++++++++
 lib/lpm/meson.build      |  1 +
 lib/lpm/rte_lpm.h        |  2 ++
 lib/lpm/rte_lpm_rvv.h    | 62 ++++++++++++++++++++++++++++++++++++++++
 6 files changed, 95 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

-- 
2.49.0

[PATCH v5 1/3] config/riscv: detect V extension

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

This patch is derived from "config/riscv: detect presence of Zbc
extension with modifications".

The RISC-V C api defines architecture extension test macros
These let us detect whether the V extension is supported on the
compiler and -march we're building with. The C api also defines V
intrinsics we can use rather than inline assembly on newer versions of
GCC (14.1.0+) and Clang (18.1.0+).

If the V extension and intrinsics are both present and we can detect
the V extension at runtime, we define a flag, RTE_RISCV_FEATURE_V.

Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
---
 .mailmap                 |  1 +
 config/riscv/meson.build | 25 +++++++++++++++++++++++++
 2 files changed, 26 insertions(+)

diff --git a/.mailmap b/.mailmap
index c65872cd9f..b635eb645a 100644
--- a/.mailmap
+++ b/.mailmap
@@ -1508,6 +1508,7 @@ Sunil Kumar Kori <skori@marvell.com> <skori@mavell.com> <sunil.kori@nxp.com>
 Sunil Pai G <sunil.pai.g@intel.com>
 Sunil Uttarwar <sunilprakashrao.uttarwar@amd.com>
 Sun Jiajia <sunx.jiajia@intel.com>
+Sun Yuechi <sunyuechi@iscas.ac.cn> <uk7b@foxmail.com>
 Sunyang Wu <sunyang.wu@jaguarmicro.com>
 Surabhi Boob <surabhi.boob@intel.com>
 Suyang Ju <sju@paloaltonetworks.com>
diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index 7562c6cb99..e3694cf2e6 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -119,6 +119,31 @@ foreach flag: arch_config['machine_args']
     endif
 endforeach
 
+# check if we can do buildtime detection of extensions supported by the target
+riscv_extension_macros = false
+if (cc.get_define('__riscv_arch_test', args: machine_args) == '1')
+  message('Detected architecture extension test macros')
+  riscv_extension_macros = true
+else
+  warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible')
+endif
+
+# detect extensions
+# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+
+if (riscv_extension_macros and
+    (cc.get_define('__riscv_vector', args: machine_args) != ''))
+  if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0'))
+      or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0')))
+    if (cc.compiles('''#include <riscv_vector.h>
+        int main(void) { size_t vl = __riscv_vsetvl_e32m1(1); }''', args: machine_args))
+      message('Compiling with the V extension')
+      machine_args += ['-DRTE_RISCV_FEATURE_V']
+    endif
+  else
+    warning('Detected V extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)')
+  endif
+endif
+
 # apply flags
 foreach flag: dpdk_flags
     if flag.length() > 0
-- 
2.49.0

[PATCH v5 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

Implement LPM lookupx4 function for RISC-V architecture using RISC-V
Vector Extension instruction set

Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
---
 MAINTAINERS           |  2 ++
 lib/lpm/meson.build   |  1 +
 lib/lpm/rte_lpm.h     |  2 ++
 lib/lpm/rte_lpm_rvv.h | 62 +++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 67 insertions(+)
 create mode 100644 lib/lpm/rte_lpm_rvv.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 3e16789250..0f207ac129 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
 F: config/riscv/
 F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
 F: lib/eal/riscv/
+M: sunyuechi <sunyuechi@iscas.ac.cn>
+F: lib/**/*rvv*
 
 Intel x86
 M: Bruce Richardson <bruce.richardson@intel.com>
diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
index fae4f79fb9..09133061e5 100644
--- a/lib/lpm/meson.build
+++ b/lib/lpm/meson.build
@@ -17,6 +17,7 @@ indirect_headers += files(
         'rte_lpm_scalar.h',
         'rte_lpm_sse.h',
         'rte_lpm_sve.h',
+        'rte_lpm_rvv.h',
 )
 deps += ['hash']
 deps += ['rcu']
diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
index 7df64f06b1..b06517206f 100644
--- a/lib/lpm/rte_lpm.h
+++ b/lib/lpm/rte_lpm.h
@@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
 #include "rte_lpm_altivec.h"
 #elif defined(RTE_ARCH_X86)
 #include "rte_lpm_sse.h"
+#elif defined(RTE_ARCH_RISCV) && defined(RTE_RISCV_FEATURE_V)
+#include "rte_lpm_rvv.h"
 #else
 #include "rte_lpm_scalar.h"
 #endif
diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
new file mode 100644
index 0000000000..5f48fb2b32
--- /dev/null
+++ b/lib/lpm/rte_lpm_rvv.h
@@ -0,0 +1,62 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
+ */
+
+#ifndef _RTE_LPM_RVV_H_
+#define _RTE_LPM_RVV_H_
+
+#include <rte_vect.h>
+
+#include <rte_cpuflags.h>
+#include <riscv_vector.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
+#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
+
+static inline void rte_lpm_lookupx4(
+	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
+{
+	size_t vl = 4;
+
+	const uint32_t *tbl24_p = (const uint32_t *)lpm->tbl24;
+	uint32_t tbl_entries[4] = {
+		tbl24_p[((uint32_t)ip[0]) >> 8],
+		tbl24_p[((uint32_t)ip[1]) >> 8],
+		tbl24_p[((uint32_t)ip[2]) >> 8],
+		tbl24_p[((uint32_t)ip[3]) >> 8],
+	};
+	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
+
+	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
+	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
+
+	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
+	    __riscv_vadd_vv_u32m1(
+		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
+		__riscv_vand_vx_u32m1(
+		    __riscv_vle32_v_u32m1((const uint32_t *)&ip, vl), 0x000000FF, vl),
+		vl),
+	    2, vl);
+
+	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
+	    mask, vtbl_entry, (const uint32_t *)(lpm->tbl8), vtbl8_index, vl);
+
+	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
+	mask = __riscv_vmseq_vx_u32m1_b32(
+	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
+
+	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
+
+	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_LPM_RVV_H_ */
-- 
2.49.0

[PATCH v5 3/3] riscv: override machine_args only when default

[uk7b]

From: Sun Yuechi <sunyuechi@iscas.ac.cn>

Support using -Dcpu_instruction_set=rv64gcv to enable V extension.

Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
---
 config/riscv/meson.build | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/config/riscv/meson.build b/config/riscv/meson.build
index e3694cf2e6..1036a86d05 100644
--- a/config/riscv/meson.build
+++ b/config/riscv/meson.build
@@ -111,6 +111,7 @@ arch_config = arch_config[arch_id]
 # Concatenate flags respecting priorities.
 dpdk_flags = flags_common + vendor_config['flags'] + arch_config.get('flags', [])
 
+if (cpu_instruction_set == 'rv64gc')
 # apply supported machine args
 machine_args = [] # Clear previous machine args
 foreach flag: arch_config['machine_args']
@@ -118,6 +119,7 @@ foreach flag: arch_config['machine_args']
         machine_args += flag
     endif
 endforeach
+endif
 
 # check if we can do buildtime detection of extensions supported by the target
 riscv_extension_macros = false
-- 
2.49.0

Re: Re: [PATCH v4 2/3] lib/lpm: R-V V rte_lpm_lookupx4

[孙越池]

Thanks, they have been updated in v5.


> -----原始邮件-----
&gt; 发件人: "Medvedkin, Vladimir" <vladimir.medvedkin@intel.com>
&gt; 发送时间: 2025-06-11 19:52:57 (星期三)
&gt; 收件人: uk7b@foxmail.com, dev@dpdk.org
&gt; 抄送: "Sun Yuechi" <sunyuechi@iscas.ac.cn>, "Thomas Monjalon" <thomas@monjalon.net>, "Bruce Richardson" <bruce.richardson@intel.com>, "Stanislaw Kardach" <stanislaw.kardach@gmail.com>
&gt; 主题: Re: [PATCH v4 2/3] lib/lpm: R-V V rte_lpm_lookupx4
&gt; 
&gt; Hi Sun,
&gt; 
&gt; You did not address my previous comments regarding commit message. You 
&gt; can put everything you've wrote in this commit as a note and add 
&gt; meaningful description about what commit generally does, like (please 
&gt; correct if needed):
&gt; 
&gt; "Implement LPM lookupx4 routine for RISC-V architecture using RISC-V 
&gt; Vector Extension instruction set"
&gt; 
&gt; Everything else (performance tests, implementation thoughts and 
&gt; considerations, etc.) should be in the patch notes. For more information 
&gt; on what "patch notes" are, you may want refer to Git documentation [1].
&gt; 
&gt; [1] https://git-scm.com/docs/git-notes
&gt; 
&gt; On 05/06/2025 11:58, uk7b@foxmail.com wrote:
&gt; 
&gt; &gt; From: Sun Yuechi <sunyuechi@iscas.ac.cn>
&gt; &gt;
&gt; &gt; The initialization of vtbl_entry is not fully vectorized here because
&gt; &gt; doing so would require __riscv_vluxei32_v_u32m1, which is slower
&gt; &gt; than the scalar approach in this small-scale scenario.
&gt; &gt;
&gt; &gt; - Test: app/test/lpm_perf_autotest
&gt; &gt; - Platform: Banana Pi(BPI-F3)
&gt; &gt; - SoC: Spacemit X60 (8 cores with Vector extension)
&gt; &gt; - CPU Frequency: up to 1.6 GHz
&gt; &gt; - Cache: 256 KiB L1d ×8, 256 KiB L1i ×8, 1 MiB L2 ×2
&gt; &gt; - Memory: 16 GiB
&gt; &gt; - Kernel: Linux 6.6.36
&gt; &gt; - Compiler: GCC 14.2.0 (with RVV intrinsic support)
&gt; &gt;
&gt; &gt; Test results(LPM LookupX4):
&gt; &gt;      scalar: 5.7 cycles
&gt; &gt;      rvv:    4.6 cycles
&gt; &gt;
&gt; &gt; Signed-off-by: Sun Yuechi <sunyuechi@iscas.ac.cn>
&gt; &gt; ---
&gt; &gt;   MAINTAINERS           |  2 ++
&gt; &gt;   lib/lpm/meson.build   |  1 +
&gt; &gt;   lib/lpm/rte_lpm.h     |  2 ++
&gt; &gt;   lib/lpm/rte_lpm_rvv.h | 62 +++++++++++++++++++++++++++++++++++++++++++
&gt; &gt;   4 files changed, 67 insertions(+)
&gt; &gt;   create mode 100644 lib/lpm/rte_lpm_rvv.h
&gt; &gt;
&gt; &gt; diff --git a/MAINTAINERS b/MAINTAINERS
&gt; &gt; index 3e16789250..0f207ac129 100644
&gt; &gt; --- a/MAINTAINERS
&gt; &gt; +++ b/MAINTAINERS
&gt; &gt; @@ -340,6 +340,8 @@ M: Stanislaw Kardach <stanislaw.kardach@gmail.com>
&gt; &gt;   F: config/riscv/
&gt; &gt;   F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst
&gt; &gt;   F: lib/eal/riscv/
&gt; &gt; +M: sunyuechi <sunyuechi@iscas.ac.cn>
&gt; &gt; +F: lib/**/*rvv*
&gt; &gt;   
&gt; &gt;   Intel x86
&gt; &gt;   M: Bruce Richardson <bruce.richardson@intel.com>
&gt; &gt; diff --git a/lib/lpm/meson.build b/lib/lpm/meson.build
&gt; &gt; index fae4f79fb9..09133061e5 100644
&gt; &gt; --- a/lib/lpm/meson.build
&gt; &gt; +++ b/lib/lpm/meson.build
&gt; &gt; @@ -17,6 +17,7 @@ indirect_headers += files(
&gt; &gt;           'rte_lpm_scalar.h',
&gt; &gt;           'rte_lpm_sse.h',
&gt; &gt;           'rte_lpm_sve.h',
&gt; &gt; +        'rte_lpm_rvv.h',
&gt; &gt;   )
&gt; &gt;   deps += ['hash']
&gt; &gt;   deps += ['rcu']
&gt; &gt; diff --git a/lib/lpm/rte_lpm.h b/lib/lpm/rte_lpm.h
&gt; &gt; index 7df64f06b1..b06517206f 100644
&gt; &gt; --- a/lib/lpm/rte_lpm.h
&gt; &gt; +++ b/lib/lpm/rte_lpm.h
&gt; &gt; @@ -408,6 +408,8 @@ rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
&gt; &gt;   #include "rte_lpm_altivec.h"
&gt; &gt;   #elif defined(RTE_ARCH_X86)
&gt; &gt;   #include "rte_lpm_sse.h"
&gt; &gt; +#elif defined(RTE_ARCH_RISCV) &amp;&amp; defined(RTE_RISCV_FEATURE_V)
&gt; &gt; +#include "rte_lpm_rvv.h"
&gt; &gt;   #else
&gt; &gt;   #include "rte_lpm_scalar.h"
&gt; &gt;   #endif
&gt; &gt; diff --git a/lib/lpm/rte_lpm_rvv.h b/lib/lpm/rte_lpm_rvv.h
&gt; &gt; new file mode 100644
&gt; &gt; index 0000000000..5f48fb2b32
&gt; &gt; --- /dev/null
&gt; &gt; +++ b/lib/lpm/rte_lpm_rvv.h
&gt; &gt; @@ -0,0 +1,62 @@
&gt; &gt; +/* SPDX-License-Identifier: BSD-3-Clause
&gt; &gt; + * Copyright (c) 2025 Institute of Software Chinese Academy of Sciences (ISCAS).
&gt; &gt; + */
&gt; &gt; +
&gt; &gt; +#ifndef _RTE_LPM_RVV_H_
&gt; &gt; +#define _RTE_LPM_RVV_H_
&gt; &gt; +
&gt; &gt; +#include <rte_vect.h>
&gt; &gt; +
&gt; &gt; +#include <rte_cpuflags.h>
&gt; &gt; +#include <riscv_vector.h>
&gt; &gt; +
&gt; &gt; +#ifdef __cplusplus
&gt; &gt; +extern "C" {
&gt; &gt; +#endif
&gt; &gt; +
&gt; &gt; +#define RTE_LPM_LOOKUP_SUCCESS 0x01000000
&gt; &gt; +#define RTE_LPM_VALID_EXT_ENTRY_BITMASK 0x03000000
&gt; &gt; +
&gt; &gt; +static inline void rte_lpm_lookupx4(
&gt; &gt; +	const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4], uint32_t defv)
&gt; &gt; +{
&gt; &gt; +	size_t vl = 4;
&gt; &gt; +
&gt; &gt; +	const uint32_t *tbl24_p = (const uint32_t *)lpm-&gt;tbl24;
&gt; &gt; +	uint32_t tbl_entries[4] = {
&gt; &gt; +		tbl24_p[((uint32_t)ip[0]) &gt;&gt; 8],
&gt; &gt; +		tbl24_p[((uint32_t)ip[1]) &gt;&gt; 8],
&gt; &gt; +		tbl24_p[((uint32_t)ip[2]) &gt;&gt; 8],
&gt; &gt; +		tbl24_p[((uint32_t)ip[3]) &gt;&gt; 8],
&gt; &gt; +	};
&gt; &gt; +	vuint32m1_t vtbl_entry = __riscv_vle32_v_u32m1(tbl_entries, vl);
&gt; &gt; +
&gt; &gt; +	vbool32_t mask = __riscv_vmseq_vx_u32m1_b32(
&gt; &gt; +	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl),
&gt; &gt; +	    RTE_LPM_VALID_EXT_ENTRY_BITMASK, vl);
&gt; &gt; +
&gt; &gt; +	vuint32m1_t vtbl8_index = __riscv_vsll_vx_u32m1(
&gt; &gt; +	    __riscv_vadd_vv_u32m1(
&gt; &gt; +		__riscv_vsll_vx_u32m1(__riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl), 8, vl),
&gt; &gt; +		__riscv_vand_vx_u32m1(
&gt; &gt; +		    __riscv_vle32_v_u32m1((const uint32_t *)&amp;ip, vl), 0x000000FF, vl),
&gt; &gt; +		vl),
&gt; &gt; +	    2, vl);
&gt; &gt; +
&gt; &gt; +	vtbl_entry = __riscv_vluxei32_v_u32m1_mu(
&gt; &gt; +	    mask, vtbl_entry, (const uint32_t *)(lpm-&gt;tbl8), vtbl8_index, vl);
&gt; &gt; +
&gt; &gt; +	vuint32m1_t vnext_hop = __riscv_vand_vx_u32m1(vtbl_entry, 0x00FFFFFF, vl);
&gt; &gt; +	mask = __riscv_vmseq_vx_u32m1_b32(
&gt; &gt; +	    __riscv_vand_vx_u32m1(vtbl_entry, RTE_LPM_LOOKUP_SUCCESS, vl), 0, vl);
&gt; &gt; +
&gt; &gt; +	vnext_hop = __riscv_vmerge_vxm_u32m1(vnext_hop, defv, mask, vl);
&gt; &gt; +
&gt; &gt; +	__riscv_vse32_v_u32m1(hop, vnext_hop, vl);
&gt; &gt; +}
&gt; &gt; +
&gt; &gt; +#ifdef __cplusplus
&gt; &gt; +}
&gt; &gt; +#endif
&gt; &gt; +
&gt; &gt; +#endif /* _RTE_LPM_RVV_H_ */
&gt; 
&gt; -- 
&gt; Regards,
&gt; Vladimir
</riscv_vector.h></rte_cpuflags.h></rte_vect.h></bruce.richardson@intel.com></sunyuechi@iscas.ac.cn></stanislaw.kardach@gmail.com></sunyuechi@iscas.ac.cn></sunyuechi@iscas.ac.cn></stanislaw.kardach@gmail.com></bruce.richardson@intel.com></thomas@monjalon.net></sunyuechi@iscas.ac.cn></vladimir.medvedkin@intel.com>