[PATCH 0/3] remove driver-specific logic for AVX builds

[PATCH 0/3] remove driver-specific logic for AVX builds

[Bruce Richardson]

A number of drivers had special optimized AVX2 and AVX512 code paths
for performance reasons, and these tended to have copy-pasted logic
to build those files. Centralise that logic in the main
drivers/meson.build file to avoid duplication.

Bruce Richardson (3):
  build: add generalized AVX handling for drivers
  net/intel: use common AVX build code
  drivers/net: build use common AVX handling

 drivers/meson.build                | 30 ++++++++++++++++++++++++++++++
 drivers/net/bnxt/meson.build       | 10 +---------
 drivers/net/enic/meson.build       | 10 +---------
 drivers/net/intel/i40e/meson.build | 26 ++------------------------
 drivers/net/intel/iavf/meson.build | 25 ++-----------------------
 drivers/net/intel/ice/meson.build  | 25 ++-----------------------
 drivers/net/intel/idpf/meson.build | 25 ++-----------------------
 drivers/net/nfp/meson.build        | 10 +---------
 drivers/net/octeon_ep/meson.build  | 14 ++------------
 drivers/net/virtio/meson.build     |  9 +--------
 10 files changed, 44 insertions(+), 140 deletions(-)

--
2.43.0

[PATCH 1/3] build: add generalized AVX handling for drivers

[Bruce Richardson]

Add support to the top-level driver build file for AVX2 and AVX512
specific sources. This should simplify driver builds by avoiding the
need to constantly reimplement the same build logic

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/meson.build | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/meson.build b/drivers/meson.build
index 05391a575d..c42c7764bf 100644
--- a/drivers/meson.build
+++ b/drivers/meson.build
@@ -126,6 +126,8 @@ foreach subpath:subdirs
         name = drv
         annotate_locks = true
         sources = []
+        sources_avx2 = []
+        sources_avx512 = []
         headers = []
         driver_sdk_headers = [] # public headers included by drivers
         objs = []
@@ -235,6 +237,34 @@ foreach subpath:subdirs
             dpdk_includes += include_directories(drv_path)
         endif
 
+        # handle avx2 and avx512 source files
+        if arch_subdir == 'x86'
+            if sources_avx2.length() > 0
+                avx2_lib = static_library(lib_name + '_avx2_lib',
+                        sources_avx2,
+                        dependencies: static_deps,
+                        include_directories: includes,
+                        c_args: [cflags, cc_avx2_flags])
+                objs += avx2_lib.extract_objects(sources_avx2)
+            endif
+            if sources_avx512.length() > 0
+                cflags += '-DCC_AVX512_SUPPORT'
+                avx512_args = [cflags, cc_avx512_flags]
+                if cc.has_argument('-march=skylake-avx512')
+                    avx512_args += '-march=skylake-avx512'
+                    if cc.has_argument('-Wno-overriding-option')
+                        avx512_args += '-Wno-overriding-option'
+                    endif
+                endif
+                avx512_lib = static_library(lib_name + '_avx512_lib',
+                        sources_avx512,
+                        dependencies: static_deps,
+                        include_directories: includes,
+                        c_args: avx512_args)
+                objs += avx512_lib.extract_objects(sources_avx512)
+            endif
+        endif
+
         # generate pmdinfo sources by building a temporary
         # lib and then running pmdinfogen on the contents of
         # that lib. The final lib reuses the object files and
-- 
2.43.0

[PATCH 2/3] net/intel: use common AVX build code

[Bruce Richardson]

Remove driver-specific build instructions for the AVX2 and AVX-512 code,
and rely instead on the generic driver build file.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/intel/i40e/meson.build | 26 ++------------------------
 drivers/net/intel/iavf/meson.build | 25 ++-----------------------
 drivers/net/intel/ice/meson.build  | 25 ++-----------------------
 drivers/net/intel/idpf/meson.build | 25 ++-----------------------
 4 files changed, 8 insertions(+), 93 deletions(-)

diff --git a/drivers/net/intel/i40e/meson.build b/drivers/net/intel/i40e/meson.build
index 2973ed1a01..25a3d72714 100644
--- a/drivers/net/intel/i40e/meson.build
+++ b/drivers/net/intel/i40e/meson.build
@@ -40,35 +40,13 @@ includes += include_directories('base')
 
 if arch_subdir == 'x86'
     sources += files('i40e_rxtx_vec_sse.c')
+    sources_avx2 = files('i40e_rxtx_vec_avx2.c')
+    sources_avx512 = files('i40e_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    i40e_avx2_lib = static_library('i40e_avx2_lib',
-            'i40e_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_kvargs, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += i40e_avx2_lib.extract_objects('i40e_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        i40e_avx512_lib = static_library('i40e_avx512_lib',
-                'i40e_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev,
-                    static_rte_kvargs, static_rte_hash],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += i40e_avx512_lib.extract_objects('i40e_rxtx_vec_avx512.c')
-    endif
 elif arch_subdir == 'ppc'
        sources += files('i40e_rxtx_vec_altivec.c')
 elif arch_subdir == 'arm'
diff --git a/drivers/net/intel/iavf/meson.build b/drivers/net/intel/iavf/meson.build
index f7eac7c57a..fc3e4d2f28 100644
--- a/drivers/net/intel/iavf/meson.build
+++ b/drivers/net/intel/iavf/meson.build
@@ -28,34 +28,13 @@ includes += include_directories('base')
 
 if arch_subdir == 'x86'
     sources += files('iavf_rxtx_vec_sse.c')
+    sources_avx2 = files('iavf_rxtx_vec_avx2.c')
+    sources_avx512 = files('iavf_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    iavf_avx2_lib = static_library('iavf_avx2_lib',
-            'iavf_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += iavf_avx2_lib.extract_objects('iavf_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        iavf_avx512_lib = static_library('iavf_avx512_lib',
-                'iavf_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += iavf_avx512_lib.extract_objects('iavf_rxtx_vec_avx512.c')
-    endif
 elif arch_subdir == 'arm'
     sources += files('iavf_rxtx_vec_neon.c')
 endif
diff --git a/drivers/net/intel/ice/meson.build b/drivers/net/intel/ice/meson.build
index cbdf38c1c4..a33ec59272 100644
--- a/drivers/net/intel/ice/meson.build
+++ b/drivers/net/intel/ice/meson.build
@@ -34,34 +34,13 @@ endif
 
 if arch_subdir == 'x86'
     sources += files('ice_rxtx_vec_sse.c')
+    sources_avx2 = files('ice_rxtx_vec_avx2.c')
+    sources_avx512 = files('ice_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    ice_avx2_lib = static_library('ice_avx2_lib',
-            'ice_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += ice_avx2_lib.extract_objects('ice_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        ice_avx512_lib = static_library('ice_avx512_lib',
-                'ice_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev, static_rte_hash],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += ice_avx512_lib.extract_objects('ice_rxtx_vec_avx512.c')
-    endif
 endif
 
 sources += files(
diff --git a/drivers/net/intel/idpf/meson.build b/drivers/net/intel/idpf/meson.build
index 4b272d02b1..10465f0f36 100644
--- a/drivers/net/intel/idpf/meson.build
+++ b/drivers/net/intel/idpf/meson.build
@@ -19,29 +19,8 @@ sources = files(
 )
 
 if arch_subdir == 'x86' and dpdk_conf.get('RTE_IOVA_IN_MBUF') == 1
-    idpf_avx2_lib = static_library('idpf_avx2_lib',
-        'idpf_common_rxtx_avx2.c',
-        dependencies: [static_rte_ethdev, static_rte_hash],
-        include_directories: includes,
-        c_args: [cflags, cc_avx2_flags])
-    objs += idpf_avx2_lib.extract_objects('idpf_common_rxtx_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        idpf_common_avx512_lib = static_library('idpf_common_avx512_lib',
-                'idpf_common_rxtx_avx512.c',
-                dependencies: static_rte_mbuf,
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += idpf_common_avx512_lib.extract_objects('idpf_common_rxtx_avx512.c')
-    endif
+    sources_avx2 = files('idpf_common_rxtx_avx2.c')
+    sources_avx512 = files('idpf_common_rxtx_avx512.c')
 endif
 
 subdir('base')
-- 
2.43.0

[PATCH 3/3] drivers/net: build use common AVX handling

[Bruce Richardson]

Remove from remaining net drivers the special-case code to handle AVX2
or AVX512 specific files. These can be built instead using
drivers/meson.build.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/bnxt/meson.build      | 10 +---------
 drivers/net/enic/meson.build      | 10 +---------
 drivers/net/nfp/meson.build       | 10 +---------
 drivers/net/octeon_ep/meson.build | 14 ++------------
 drivers/net/virtio/meson.build    |  9 +--------
 5 files changed, 6 insertions(+), 47 deletions(-)

diff --git a/drivers/net/bnxt/meson.build b/drivers/net/bnxt/meson.build
index fd82d0c409..dcca7df916 100644
--- a/drivers/net/bnxt/meson.build
+++ b/drivers/net/bnxt/meson.build
@@ -58,15 +58,7 @@ subdir('hcapi/cfa_v3')
 
 if arch_subdir == 'x86'
     sources += files('bnxt_rxtx_vec_sse.c')
-    # build AVX2 code with instruction set explicitly enabled for runtime selection
-    bnxt_avx2_lib = static_library('bnxt_avx2_lib',
-            'bnxt_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev,
-                static_rte_bus_pci,
-                static_rte_kvargs, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-     objs += bnxt_avx2_lib.extract_objects('bnxt_rxtx_vec_avx2.c')
+    sources_avx2 = files('bnxt_rxtx_vec_avx2.c')
 elif arch_subdir == 'arm' and dpdk_conf.get('RTE_ARCH_64')
     sources += files('bnxt_rxtx_vec_neon.c')
 endif
diff --git a/drivers/net/enic/meson.build b/drivers/net/enic/meson.build
index cfe5ec170a..2b3052fae8 100644
--- a/drivers/net/enic/meson.build
+++ b/drivers/net/enic/meson.build
@@ -29,17 +29,9 @@ sources = files(
 deps += ['hash']
 includes += include_directories('base')
 
-# Build the avx2 handler for 64-bit X86 targets, even though 'machine'
-# may not. This is to support users who build for the min supported machine
-# and need to run the binary on newer CPUs too.
 if dpdk_conf.has('RTE_ARCH_X86_64')
     cflags += '-DENIC_RXTX_VEC'
-    enic_avx2_lib = static_library('enic_avx2_lib',
-            'enic_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_bus_pci],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += enic_avx2_lib.extract_objects('enic_rxtx_vec_avx2.c')
+    sources_avx2 = files('enic_rxtx_vec_avx2.c')
 endif
 
 annotate_locks = false
diff --git a/drivers/net/nfp/meson.build b/drivers/net/nfp/meson.build
index 0a12b7dce7..a98b584042 100644
--- a/drivers/net/nfp/meson.build
+++ b/drivers/net/nfp/meson.build
@@ -52,19 +52,11 @@ cflags += no_wvla_cflag
 if arch_subdir == 'x86'
         includes += include_directories('../../common/nfp')
 
-        avx2_sources = files(
+        sources_avx2 = files(
                 'nfdk/nfp_nfdk_vec_avx2_dp.c',
                 'nfp_rxtx_vec_avx2.c',
         )
 
-        nfp_avx2_lib = static_library('nfp_avx2_lib',
-                avx2_sources,
-                dependencies: [static_rte_ethdev, static_rte_bus_pci],
-                include_directories: includes,
-                c_args: [cflags, cc_avx2_flags]
-        )
-
-        objs += nfp_avx2_lib.extract_all_objects(recursive: true)
 else
         sources += files(
                 'nfp_rxtx_vec_stub.c',
diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
index 1b34db3edc..9bf4627894 100644
--- a/drivers/net/octeon_ep/meson.build
+++ b/drivers/net/octeon_ep/meson.build
@@ -15,18 +15,8 @@ sources = files(
 
 if arch_subdir == 'x86'
     sources += files('cnxk_ep_rx_sse.c')
-    if cc.get_define('__AVX2__', args: machine_args) != ''
-        cflags += ['-DCC_AVX2_SUPPORT']
-        sources += files('cnxk_ep_rx_avx.c')
-    elif cc.has_multi_arguments(cc_avx2_flags)
-        cflags += ['-DCC_AVX2_SUPPORT']
-        otx_ep_avx2_lib = static_library('otx_ep_avx2_lib',
-                        'cnxk_ep_rx_avx.c',
-                        dependencies: [static_rte_ethdev, static_rte_pci, static_rte_bus_pci],
-                        include_directories: includes,
-                        c_args: [cflags, cc_avx2_flags])
-        objs += otx_ep_avx2_lib.extract_objects('cnxk_ep_rx_avx.c')
-    endif
+    cflags += ['-DCC_AVX2_SUPPORT']
+    sources_avx2 = files('cnxk_ep_rx_avx.c')
 endif
 
 if arch_subdir == 'arm'
diff --git a/drivers/net/virtio/meson.build b/drivers/net/virtio/meson.build
index c1c4a85bea..01bfb3c47d 100644
--- a/drivers/net/virtio/meson.build
+++ b/drivers/net/virtio/meson.build
@@ -27,15 +27,8 @@ cflags += no_wvla_cflag
 
 if arch_subdir == 'x86'
     if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
         cflags += ['-DVIRTIO_RXTX_PACKED_VEC']
-        virtio_avx512_lib = static_library('virtio_avx512_lib',
-                'virtio_rxtx_packed.c',
-                dependencies: [static_rte_ethdev,
-                    static_rte_kvargs, static_rte_bus_pci],
-                include_directories: includes,
-                c_args: cflags + cc_avx512_flags)
-        objs += virtio_avx512_lib.extract_objects('virtio_rxtx_packed.c')
+        sources_avx512 = files('virtio_rxtx_packed.c')
         if (toolchain == 'gcc' and cc.version().version_compare('>=8.3.0'))
             cflags += '-DVIRTIO_GCC_UNROLL_PRAGMA'
         elif (toolchain == 'clang' and cc.version().version_compare('>=3.7.0'))
-- 
2.43.0

Re: [PATCH 3/3] drivers/net: build use common AVX handling

[David Marchand]

On Fri, Mar 14, 2025 at 6:24 PM Bruce Richardson
<bruce.richardson@intel.com> wrote:
> diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
> index 1b34db3edc..9bf4627894 100644
> --- a/drivers/net/octeon_ep/meson.build
> +++ b/drivers/net/octeon_ep/meson.build
> @@ -15,18 +15,8 @@ sources = files(
>
>  if arch_subdir == 'x86'
>      sources += files('cnxk_ep_rx_sse.c')
> -    if cc.get_define('__AVX2__', args: machine_args) != ''
> -        cflags += ['-DCC_AVX2_SUPPORT']
> -        sources += files('cnxk_ep_rx_avx.c')
> -    elif cc.has_multi_arguments(cc_avx2_flags)
> -        cflags += ['-DCC_AVX2_SUPPORT']
> -        otx_ep_avx2_lib = static_library('otx_ep_avx2_lib',
> -                        'cnxk_ep_rx_avx.c',
> -                        dependencies: [static_rte_ethdev, static_rte_pci, static_rte_bus_pci],
> -                        include_directories: includes,
> -                        c_args: [cflags, cc_avx2_flags])
> -        objs += otx_ep_avx2_lib.extract_objects('cnxk_ep_rx_avx.c')
> -    endif
> +    cflags += ['-DCC_AVX2_SUPPORT']

Could it be set by drivers/meson.build?
Similarly to AVX512.


> +    sources_avx2 = files('cnxk_ep_rx_avx.c')
>  endif
>
>  if arch_subdir == 'arm'

-- 
David Marchand

Re: [PATCH 3/3] drivers/net: build use common AVX handling

[Bruce Richardson]

On Fri, Mar 14, 2025 at 06:31:08PM +0100, David Marchand wrote:
> On Fri, Mar 14, 2025 at 6:24 PM Bruce Richardson
> <bruce.richardson@intel.com> wrote:
> > diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
> > index 1b34db3edc..9bf4627894 100644
> > --- a/drivers/net/octeon_ep/meson.build
> > +++ b/drivers/net/octeon_ep/meson.build
> > @@ -15,18 +15,8 @@ sources = files(
> >
> >  if arch_subdir == 'x86'
> >      sources += files('cnxk_ep_rx_sse.c')
> > -    if cc.get_define('__AVX2__', args: machine_args) != ''
> > -        cflags += ['-DCC_AVX2_SUPPORT']
> > -        sources += files('cnxk_ep_rx_avx.c')
> > -    elif cc.has_multi_arguments(cc_avx2_flags)
> > -        cflags += ['-DCC_AVX2_SUPPORT']
> > -        otx_ep_avx2_lib = static_library('otx_ep_avx2_lib',
> > -                        'cnxk_ep_rx_avx.c',
> > -                        dependencies: [static_rte_ethdev, static_rte_pci, static_rte_bus_pci],
> > -                        include_directories: includes,
> > -                        c_args: [cflags, cc_avx2_flags])
> > -        objs += otx_ep_avx2_lib.extract_objects('cnxk_ep_rx_avx.c')
> > -    endif
> > +    cflags += ['-DCC_AVX2_SUPPORT']
> 
> Could it be set by drivers/meson.build?
> Similarly to AVX512.
> 
Yes, but really I'd rather see it removed. It's now the same as the defines
for x86, since we always build AVX2 support when building for x86, because
all compilers we support can generate AVX2 code.

/Bruce

[PATCH v2 0/4] remove driver-specific logic for AVX builds

[Bruce Richardson]

A number of drivers had special optimized AVX2 and AVX512 code paths
for performance reasons, and these tended to have copy-pasted logic
to build those files. Centralise that logic in the main
drivers/meson.build file to avoid duplication.

v2: add patch 4 to remove use of unnecessary CC_AVX2_SUPPORT flag

Bruce Richardson (4):
  build: add generalized AVX handling for drivers
  net/intel: use common AVX build code
  drivers/net: build use common AVX handling
  drivers/net: remove AVX2 build-time define

 drivers/meson.build                   | 30 +++++++++++++++++++++++++++
 drivers/net/bnxt/bnxt_ethdev.c        |  2 --
 drivers/net/bnxt/meson.build          | 10 +--------
 drivers/net/enic/meson.build          | 10 +--------
 drivers/net/intel/i40e/meson.build    | 26 ++---------------------
 drivers/net/intel/iavf/meson.build    | 25 ++--------------------
 drivers/net/intel/ice/meson.build     | 25 ++--------------------
 drivers/net/intel/idpf/meson.build    | 25 ++--------------------
 drivers/net/nfp/meson.build           | 10 +--------
 drivers/net/octeon_ep/meson.build     | 13 +-----------
 drivers/net/octeon_ep/otx_ep_ethdev.c |  4 ----
 drivers/net/virtio/meson.build        |  9 +-------
 12 files changed, 43 insertions(+), 146 deletions(-)

--
2.43.0

[PATCH v2 1/4] build: add generalized AVX handling for drivers

[Bruce Richardson]

Add support to the top-level driver build file for AVX2 and AVX512
specific sources. This should simplify driver builds by avoiding the
need to constantly reimplement the same build logic

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/meson.build | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/meson.build b/drivers/meson.build
index 05391a575d..c42c7764bf 100644
--- a/drivers/meson.build
+++ b/drivers/meson.build
@@ -126,6 +126,8 @@ foreach subpath:subdirs
         name = drv
         annotate_locks = true
         sources = []
+        sources_avx2 = []
+        sources_avx512 = []
         headers = []
         driver_sdk_headers = [] # public headers included by drivers
         objs = []
@@ -235,6 +237,34 @@ foreach subpath:subdirs
             dpdk_includes += include_directories(drv_path)
         endif
 
+        # handle avx2 and avx512 source files
+        if arch_subdir == 'x86'
+            if sources_avx2.length() > 0
+                avx2_lib = static_library(lib_name + '_avx2_lib',
+                        sources_avx2,
+                        dependencies: static_deps,
+                        include_directories: includes,
+                        c_args: [cflags, cc_avx2_flags])
+                objs += avx2_lib.extract_objects(sources_avx2)
+            endif
+            if sources_avx512.length() > 0
+                cflags += '-DCC_AVX512_SUPPORT'
+                avx512_args = [cflags, cc_avx512_flags]
+                if cc.has_argument('-march=skylake-avx512')
+                    avx512_args += '-march=skylake-avx512'
+                    if cc.has_argument('-Wno-overriding-option')
+                        avx512_args += '-Wno-overriding-option'
+                    endif
+                endif
+                avx512_lib = static_library(lib_name + '_avx512_lib',
+                        sources_avx512,
+                        dependencies: static_deps,
+                        include_directories: includes,
+                        c_args: avx512_args)
+                objs += avx512_lib.extract_objects(sources_avx512)
+            endif
+        endif
+
         # generate pmdinfo sources by building a temporary
         # lib and then running pmdinfogen on the contents of
         # that lib. The final lib reuses the object files and
-- 
2.43.0

[PATCH v2 2/4] net/intel: use common AVX build code

[Bruce Richardson]

Remove driver-specific build instructions for the AVX2 and AVX-512 code,
and rely instead on the generic driver build file.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/intel/i40e/meson.build | 26 ++------------------------
 drivers/net/intel/iavf/meson.build | 25 ++-----------------------
 drivers/net/intel/ice/meson.build  | 25 ++-----------------------
 drivers/net/intel/idpf/meson.build | 25 ++-----------------------
 4 files changed, 8 insertions(+), 93 deletions(-)

diff --git a/drivers/net/intel/i40e/meson.build b/drivers/net/intel/i40e/meson.build
index 2973ed1a01..25a3d72714 100644
--- a/drivers/net/intel/i40e/meson.build
+++ b/drivers/net/intel/i40e/meson.build
@@ -40,35 +40,13 @@ includes += include_directories('base')
 
 if arch_subdir == 'x86'
     sources += files('i40e_rxtx_vec_sse.c')
+    sources_avx2 = files('i40e_rxtx_vec_avx2.c')
+    sources_avx512 = files('i40e_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    i40e_avx2_lib = static_library('i40e_avx2_lib',
-            'i40e_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_kvargs, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += i40e_avx2_lib.extract_objects('i40e_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        i40e_avx512_lib = static_library('i40e_avx512_lib',
-                'i40e_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev,
-                    static_rte_kvargs, static_rte_hash],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += i40e_avx512_lib.extract_objects('i40e_rxtx_vec_avx512.c')
-    endif
 elif arch_subdir == 'ppc'
        sources += files('i40e_rxtx_vec_altivec.c')
 elif arch_subdir == 'arm'
diff --git a/drivers/net/intel/iavf/meson.build b/drivers/net/intel/iavf/meson.build
index f7eac7c57a..fc3e4d2f28 100644
--- a/drivers/net/intel/iavf/meson.build
+++ b/drivers/net/intel/iavf/meson.build
@@ -28,34 +28,13 @@ includes += include_directories('base')
 
 if arch_subdir == 'x86'
     sources += files('iavf_rxtx_vec_sse.c')
+    sources_avx2 = files('iavf_rxtx_vec_avx2.c')
+    sources_avx512 = files('iavf_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    iavf_avx2_lib = static_library('iavf_avx2_lib',
-            'iavf_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += iavf_avx2_lib.extract_objects('iavf_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        iavf_avx512_lib = static_library('iavf_avx512_lib',
-                'iavf_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += iavf_avx512_lib.extract_objects('iavf_rxtx_vec_avx512.c')
-    endif
 elif arch_subdir == 'arm'
     sources += files('iavf_rxtx_vec_neon.c')
 endif
diff --git a/drivers/net/intel/ice/meson.build b/drivers/net/intel/ice/meson.build
index cbdf38c1c4..a33ec59272 100644
--- a/drivers/net/intel/ice/meson.build
+++ b/drivers/net/intel/ice/meson.build
@@ -34,34 +34,13 @@ endif
 
 if arch_subdir == 'x86'
     sources += files('ice_rxtx_vec_sse.c')
+    sources_avx2 = files('ice_rxtx_vec_avx2.c')
+    sources_avx512 = files('ice_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    ice_avx2_lib = static_library('ice_avx2_lib',
-            'ice_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += ice_avx2_lib.extract_objects('ice_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        ice_avx512_lib = static_library('ice_avx512_lib',
-                'ice_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev, static_rte_hash],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += ice_avx512_lib.extract_objects('ice_rxtx_vec_avx512.c')
-    endif
 endif
 
 sources += files(
diff --git a/drivers/net/intel/idpf/meson.build b/drivers/net/intel/idpf/meson.build
index 4b272d02b1..10465f0f36 100644
--- a/drivers/net/intel/idpf/meson.build
+++ b/drivers/net/intel/idpf/meson.build
@@ -19,29 +19,8 @@ sources = files(
 )
 
 if arch_subdir == 'x86' and dpdk_conf.get('RTE_IOVA_IN_MBUF') == 1
-    idpf_avx2_lib = static_library('idpf_avx2_lib',
-        'idpf_common_rxtx_avx2.c',
-        dependencies: [static_rte_ethdev, static_rte_hash],
-        include_directories: includes,
-        c_args: [cflags, cc_avx2_flags])
-    objs += idpf_avx2_lib.extract_objects('idpf_common_rxtx_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        idpf_common_avx512_lib = static_library('idpf_common_avx512_lib',
-                'idpf_common_rxtx_avx512.c',
-                dependencies: static_rte_mbuf,
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += idpf_common_avx512_lib.extract_objects('idpf_common_rxtx_avx512.c')
-    endif
+    sources_avx2 = files('idpf_common_rxtx_avx2.c')
+    sources_avx512 = files('idpf_common_rxtx_avx512.c')
 endif
 
 subdir('base')
-- 
2.43.0

[PATCH v2 3/4] drivers/net: build use common AVX handling

[Bruce Richardson]

Remove from remaining net drivers the special-case code to handle AVX2
or AVX512 specific files. These can be built instead using
drivers/meson.build.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/bnxt/meson.build      | 10 +---------
 drivers/net/enic/meson.build      | 10 +---------
 drivers/net/nfp/meson.build       | 10 +---------
 drivers/net/octeon_ep/meson.build | 14 ++------------
 drivers/net/virtio/meson.build    |  9 +--------
 5 files changed, 6 insertions(+), 47 deletions(-)

diff --git a/drivers/net/bnxt/meson.build b/drivers/net/bnxt/meson.build
index fd82d0c409..dcca7df916 100644
--- a/drivers/net/bnxt/meson.build
+++ b/drivers/net/bnxt/meson.build
@@ -58,15 +58,7 @@ subdir('hcapi/cfa_v3')
 
 if arch_subdir == 'x86'
     sources += files('bnxt_rxtx_vec_sse.c')
-    # build AVX2 code with instruction set explicitly enabled for runtime selection
-    bnxt_avx2_lib = static_library('bnxt_avx2_lib',
-            'bnxt_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev,
-                static_rte_bus_pci,
-                static_rte_kvargs, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-     objs += bnxt_avx2_lib.extract_objects('bnxt_rxtx_vec_avx2.c')
+    sources_avx2 = files('bnxt_rxtx_vec_avx2.c')
 elif arch_subdir == 'arm' and dpdk_conf.get('RTE_ARCH_64')
     sources += files('bnxt_rxtx_vec_neon.c')
 endif
diff --git a/drivers/net/enic/meson.build b/drivers/net/enic/meson.build
index cfe5ec170a..2b3052fae8 100644
--- a/drivers/net/enic/meson.build
+++ b/drivers/net/enic/meson.build
@@ -29,17 +29,9 @@ sources = files(
 deps += ['hash']
 includes += include_directories('base')
 
-# Build the avx2 handler for 64-bit X86 targets, even though 'machine'
-# may not. This is to support users who build for the min supported machine
-# and need to run the binary on newer CPUs too.
 if dpdk_conf.has('RTE_ARCH_X86_64')
     cflags += '-DENIC_RXTX_VEC'
-    enic_avx2_lib = static_library('enic_avx2_lib',
-            'enic_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_bus_pci],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += enic_avx2_lib.extract_objects('enic_rxtx_vec_avx2.c')
+    sources_avx2 = files('enic_rxtx_vec_avx2.c')
 endif
 
 annotate_locks = false
diff --git a/drivers/net/nfp/meson.build b/drivers/net/nfp/meson.build
index 0a12b7dce7..a98b584042 100644
--- a/drivers/net/nfp/meson.build
+++ b/drivers/net/nfp/meson.build
@@ -52,19 +52,11 @@ cflags += no_wvla_cflag
 if arch_subdir == 'x86'
         includes += include_directories('../../common/nfp')
 
-        avx2_sources = files(
+        sources_avx2 = files(
                 'nfdk/nfp_nfdk_vec_avx2_dp.c',
                 'nfp_rxtx_vec_avx2.c',
         )
 
-        nfp_avx2_lib = static_library('nfp_avx2_lib',
-                avx2_sources,
-                dependencies: [static_rte_ethdev, static_rte_bus_pci],
-                include_directories: includes,
-                c_args: [cflags, cc_avx2_flags]
-        )
-
-        objs += nfp_avx2_lib.extract_all_objects(recursive: true)
 else
         sources += files(
                 'nfp_rxtx_vec_stub.c',
diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
index 1b34db3edc..9bf4627894 100644
--- a/drivers/net/octeon_ep/meson.build
+++ b/drivers/net/octeon_ep/meson.build
@@ -15,18 +15,8 @@ sources = files(
 
 if arch_subdir == 'x86'
     sources += files('cnxk_ep_rx_sse.c')
-    if cc.get_define('__AVX2__', args: machine_args) != ''
-        cflags += ['-DCC_AVX2_SUPPORT']
-        sources += files('cnxk_ep_rx_avx.c')
-    elif cc.has_multi_arguments(cc_avx2_flags)
-        cflags += ['-DCC_AVX2_SUPPORT']
-        otx_ep_avx2_lib = static_library('otx_ep_avx2_lib',
-                        'cnxk_ep_rx_avx.c',
-                        dependencies: [static_rte_ethdev, static_rte_pci, static_rte_bus_pci],
-                        include_directories: includes,
-                        c_args: [cflags, cc_avx2_flags])
-        objs += otx_ep_avx2_lib.extract_objects('cnxk_ep_rx_avx.c')
-    endif
+    cflags += ['-DCC_AVX2_SUPPORT']
+    sources_avx2 = files('cnxk_ep_rx_avx.c')
 endif
 
 if arch_subdir == 'arm'
diff --git a/drivers/net/virtio/meson.build b/drivers/net/virtio/meson.build
index c1c4a85bea..01bfb3c47d 100644
--- a/drivers/net/virtio/meson.build
+++ b/drivers/net/virtio/meson.build
@@ -27,15 +27,8 @@ cflags += no_wvla_cflag
 
 if arch_subdir == 'x86'
     if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
         cflags += ['-DVIRTIO_RXTX_PACKED_VEC']
-        virtio_avx512_lib = static_library('virtio_avx512_lib',
-                'virtio_rxtx_packed.c',
-                dependencies: [static_rte_ethdev,
-                    static_rte_kvargs, static_rte_bus_pci],
-                include_directories: includes,
-                c_args: cflags + cc_avx512_flags)
-        objs += virtio_avx512_lib.extract_objects('virtio_rxtx_packed.c')
+        sources_avx512 = files('virtio_rxtx_packed.c')
         if (toolchain == 'gcc' and cc.version().version_compare('>=8.3.0'))
             cflags += '-DVIRTIO_GCC_UNROLL_PRAGMA'
         elif (toolchain == 'clang' and cc.version().version_compare('>=3.7.0'))
-- 
2.43.0

[PATCH v2 4/4] drivers/net: remove AVX2 build-time define

[Bruce Richardson]

Since all supported compilers can generate AVX2 code, we will always
enable the build of the AVX2 files on x86. This means that
CC_AVX2_SUPPORT is always true on x86, so it can be removed and a
regular "#ifdef RTE_ARCH_x86" used in its place.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/bnxt/bnxt_ethdev.c        | 2 --
 drivers/net/octeon_ep/meson.build     | 1 -
 drivers/net/octeon_ep/otx_ep_ethdev.c | 4 ----
 3 files changed, 7 deletions(-)

diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
index 144d4377bd..a318604c20 100644
--- a/drivers/net/bnxt/bnxt_ethdev.c
+++ b/drivers/net/bnxt/bnxt_ethdev.c
@@ -3238,8 +3238,6 @@ static const struct {
 #if defined(RTE_ARCH_X86)
 	{bnxt_crx_pkts_vec,		"Vector SSE"},
 	{bnxt_recv_pkts_vec,		"Vector SSE"},
-#endif
-#if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
 	{bnxt_crx_pkts_vec_avx2,	"Vector AVX2"},
 	{bnxt_recv_pkts_vec_avx2,	"Vector AVX2"},
 #endif
diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
index 9bf4627894..a4a7663d1d 100644
--- a/drivers/net/octeon_ep/meson.build
+++ b/drivers/net/octeon_ep/meson.build
@@ -15,7 +15,6 @@ sources = files(
 
 if arch_subdir == 'x86'
     sources += files('cnxk_ep_rx_sse.c')
-    cflags += ['-DCC_AVX2_SUPPORT']
     sources_avx2 = files('cnxk_ep_rx_avx.c')
 endif
 
diff --git a/drivers/net/octeon_ep/otx_ep_ethdev.c b/drivers/net/octeon_ep/otx_ep_ethdev.c
index 8b14734b0c..10f2f8a2e0 100644
--- a/drivers/net/octeon_ep/otx_ep_ethdev.c
+++ b/drivers/net/octeon_ep/otx_ep_ethdev.c
@@ -91,11 +91,9 @@ otx_ep_set_rx_func(struct rte_eth_dev *eth_dev)
 		eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts;
 #ifdef RTE_ARCH_X86
 		eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_sse;
-#ifdef CC_AVX2_SUPPORT
 		if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
 		    rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1)
 			eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_avx;
-#endif
 #elif defined(RTE_ARCH_ARM64)
 		eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_neon;
 #endif
@@ -105,11 +103,9 @@ otx_ep_set_rx_func(struct rte_eth_dev *eth_dev)
 		eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts;
 #ifdef RTE_ARCH_X86
 		eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_sse;
-#ifdef CC_AVX2_SUPPORT
 		if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
 		    rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1)
 			eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_avx;
-#endif
 #elif defined(RTE_ARCH_ARM64)
 		eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_neon;
 #endif
-- 
2.43.0

Re: [PATCH v2 4/4] drivers/net: remove AVX2 build-time define

[David Marchand]

On Fri, Mar 14, 2025 at 6:44 PM Bruce Richardson
<bruce.richardson@intel.com> wrote:
>
> Since all supported compilers can generate AVX2 code, we will always
> enable the build of the AVX2 files on x86. This means that
> CC_AVX2_SUPPORT is always true on x86, so it can be removed and a
> regular "#ifdef RTE_ARCH_x86" used in its place.
>
> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
> ---
>  drivers/net/bnxt/bnxt_ethdev.c        | 2 --
>  drivers/net/octeon_ep/meson.build     | 1 -
>  drivers/net/octeon_ep/otx_ep_ethdev.c | 4 ----
>  3 files changed, 7 deletions(-)
>
> diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
> index 144d4377bd..a318604c20 100644
> --- a/drivers/net/bnxt/bnxt_ethdev.c
> +++ b/drivers/net/bnxt/bnxt_ethdev.c
> @@ -3238,8 +3238,6 @@ static const struct {
>  #if defined(RTE_ARCH_X86)
>         {bnxt_crx_pkts_vec,             "Vector SSE"},
>         {bnxt_recv_pkts_vec,            "Vector SSE"},
> -#endif
> -#if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)

I was wondering why this was not coming along a meson.build.
On the principle, this
Fixes: d58c6c077cc2 ("net/bnxt: add AVX2 Rx for compressed CQE")


>         {bnxt_crx_pkts_vec_avx2,        "Vector AVX2"},
>         {bnxt_recv_pkts_vec_avx2,       "Vector AVX2"},
>  #endif


-- 
David Marchand

Re: [PATCH v2 0/4] remove driver-specific logic for AVX builds

[David Marchand]

On Fri, Mar 14, 2025 at 6:44 PM Bruce Richardson
<bruce.richardson@intel.com> wrote:
>
> A number of drivers had special optimized AVX2 and AVX512 code paths
> for performance reasons, and these tended to have copy-pasted logic
> to build those files. Centralise that logic in the main
> drivers/meson.build file to avoid duplication.
>
> v2: add patch 4 to remove use of unnecessary CC_AVX2_SUPPORT flag
>
> Bruce Richardson (4):
>   build: add generalized AVX handling for drivers
>   net/intel: use common AVX build code
>   drivers/net: build use common AVX handling
>   drivers/net: remove AVX2 build-time define
>
>  drivers/meson.build                   | 30 +++++++++++++++++++++++++++
>  drivers/net/bnxt/bnxt_ethdev.c        |  2 --
>  drivers/net/bnxt/meson.build          | 10 +--------
>  drivers/net/enic/meson.build          | 10 +--------
>  drivers/net/intel/i40e/meson.build    | 26 ++---------------------
>  drivers/net/intel/iavf/meson.build    | 25 ++--------------------
>  drivers/net/intel/ice/meson.build     | 25 ++--------------------
>  drivers/net/intel/idpf/meson.build    | 25 ++--------------------
>  drivers/net/nfp/meson.build           | 10 +--------
>  drivers/net/octeon_ep/meson.build     | 13 +-----------
>  drivers/net/octeon_ep/otx_ep_ethdev.c |  4 ----
>  drivers/net/virtio/meson.build        |  9 +-------
>  12 files changed, 43 insertions(+), 146 deletions(-)

event/dlb2 can be converted too, or is there something special about
this driver?


As part of the discussion on the base drivers, and seeing this series,
it would be cool if we had a generic framework to specify a set of
cflags for a set of sources.
But at least this series lgtm and is a first step.


-- 
David Marchand

Re: [PATCH v2 0/4] remove driver-specific logic for AVX builds

[Bruce Richardson]

On Mon, Mar 17, 2025 at 10:50:24AM +0100, David Marchand wrote:
> On Fri, Mar 14, 2025 at 6:44 PM Bruce Richardson
> <bruce.richardson@intel.com> wrote:
> >
> > A number of drivers had special optimized AVX2 and AVX512 code paths
> > for performance reasons, and these tended to have copy-pasted logic
> > to build those files. Centralise that logic in the main
> > drivers/meson.build file to avoid duplication.
> >
> > v2: add patch 4 to remove use of unnecessary CC_AVX2_SUPPORT flag
> >
> > Bruce Richardson (4):
> >   build: add generalized AVX handling for drivers
> >   net/intel: use common AVX build code
> >   drivers/net: build use common AVX handling
> >   drivers/net: remove AVX2 build-time define
> >
> >  drivers/meson.build                   | 30 +++++++++++++++++++++++++++
> >  drivers/net/bnxt/bnxt_ethdev.c        |  2 --
> >  drivers/net/bnxt/meson.build          | 10 +--------
> >  drivers/net/enic/meson.build          | 10 +--------
> >  drivers/net/intel/i40e/meson.build    | 26 ++---------------------
> >  drivers/net/intel/iavf/meson.build    | 25 ++--------------------
> >  drivers/net/intel/ice/meson.build     | 25 ++--------------------
> >  drivers/net/intel/idpf/meson.build    | 25 ++--------------------
> >  drivers/net/nfp/meson.build           | 10 +--------
> >  drivers/net/octeon_ep/meson.build     | 13 +-----------
> >  drivers/net/octeon_ep/otx_ep_ethdev.c |  4 ----
> >  drivers/net/virtio/meson.build        |  9 +-------
> >  12 files changed, 43 insertions(+), 146 deletions(-)
> 
> event/dlb2 can be converted too, or is there something special about
> this driver?
> 

Ideally, yes, but it's also a bit special in that it only builds *either*
the SSE code path or the AVX-512 one, and has duplicate symbols/functions
in the two files. This is a bit strange, and probably not what we want, so
I may need to do some work on it before converting it to this scheme.

> 
> As part of the discussion on the base drivers, and seeing this series,
> it would be cool if we had a generic framework to specify a set of
> cflags for a set of sources.

I'd rather not have that. Meson doesn't support having per-c-file flags,
and I don't think we should resort to massive amounts of hackery to try and
support this. I think having the same set of cflags across a whole
component is a good principle, one which we should only violate when we
have to e.g. for base dirs, and for ISA-specific files.

> But at least this series lgtm and is a first step.
> 
Thanks.
/Bruce

[PATCH v3 00/11] remove component-specific logic for AVX builds

[Bruce Richardson]

A number of libs and drivers had special optimized AVX2 and AVX512 code
paths for performance reasons, and these tended to have copy-pasted
logic to build those files. Centralise that logic in the main
drivers/ and lib/ meson.build files to avoid duplication.

v3: add patch for event/dlb2 AVX512 handling.
    add common code for libraries as well as drivers.
v2: add patch 4 to remove use of unnecessary CC_AVX2_SUPPORT flag


Bruce Richardson (11):
  build: add generalized AVX handling for drivers
  net/intel: use common AVX build code
  drivers/net: build use common AVX handling
  drivers/net: remove AVX2 build-time define
  event/dlb2: build using common AVX handling
  build: add generalized AVX handling for libs
  acl: use common AVX build handling
  fib: use common AVX build handling
  net: simplify build-time logic for x86
  net: use common AVX512 build code
  member: use common AVX512 build support

 drivers/event/dlb2/dlb2_sse.c         |  4 ++
 drivers/event/dlb2/meson.build        | 16 +-------
 drivers/meson.build                   | 30 ++++++++++++++
 drivers/net/bnxt/bnxt_ethdev.c        |  2 -
 drivers/net/bnxt/meson.build          | 10 +----
 drivers/net/enic/meson.build          | 10 +----
 drivers/net/intel/i40e/meson.build    | 26 +-----------
 drivers/net/intel/iavf/meson.build    | 25 +-----------
 drivers/net/intel/ice/meson.build     | 25 +-----------
 drivers/net/intel/idpf/meson.build    | 25 +-----------
 drivers/net/nfp/meson.build           | 10 +----
 drivers/net/octeon_ep/meson.build     | 13 +-----
 drivers/net/octeon_ep/otx_ep_ethdev.c |  4 --
 drivers/net/virtio/meson.build        |  9 +----
 lib/acl/meson.build                   | 54 ++-----------------------
 lib/fib/dir24_8.c                     |  6 +--
 lib/fib/meson.build                   | 18 +--------
 lib/fib/trie.c                        |  6 +--
 lib/member/meson.build                | 46 ++++-----------------
 lib/meson.build                       | 34 +++++++++++++++-
 lib/net/meson.build                   | 58 +++------------------------
 lib/net/rte_net_crc.c                 | 16 ++++----
 22 files changed, 114 insertions(+), 333 deletions(-)

--
2.43.0

[PATCH v3 01/11] build: add generalized AVX handling for drivers

[Bruce Richardson]

Add support to the top-level driver build file for AVX2 and AVX512
specific sources. This should simplify driver builds by avoiding the
need to constantly reimplement the same build logic

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/meson.build | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

diff --git a/drivers/meson.build b/drivers/meson.build
index 05391a575d..45082f9338 100644
--- a/drivers/meson.build
+++ b/drivers/meson.build
@@ -126,6 +126,8 @@ foreach subpath:subdirs
         name = drv
         annotate_locks = true
         sources = []
+        sources_avx2 = []
+        sources_avx512 = []
         headers = []
         driver_sdk_headers = [] # public headers included by drivers
         objs = []
@@ -235,6 +237,34 @@ foreach subpath:subdirs
             dpdk_includes += include_directories(drv_path)
         endif
 
+        # handle avx2 and avx512 source files
+        if arch_subdir == 'x86'
+            if sources_avx2.length() > 0
+                avx2_lib = static_library(lib_name + '_avx2_lib',
+                        sources_avx2,
+                        dependencies: static_deps,
+                        include_directories: includes,
+                        c_args: [cflags, cc_avx2_flags])
+                objs += avx2_lib.extract_objects(sources_avx2)
+            endif
+            if sources_avx512.length() > 0 and cc_has_avx512
+                cflags += '-DCC_AVX512_SUPPORT'
+                avx512_args = [cflags, cc_avx512_flags]
+                if cc.has_argument('-march=skylake-avx512')
+                    avx512_args += '-march=skylake-avx512'
+                    if cc.has_argument('-Wno-overriding-option')
+                        avx512_args += '-Wno-overriding-option'
+                    endif
+                endif
+                avx512_lib = static_library(lib_name + '_avx512_lib',
+                        sources_avx512,
+                        dependencies: static_deps,
+                        include_directories: includes,
+                        c_args: avx512_args)
+                objs += avx512_lib.extract_objects(sources_avx512)
+            endif
+        endif
+
         # generate pmdinfo sources by building a temporary
         # lib and then running pmdinfogen on the contents of
         # that lib. The final lib reuses the object files and
-- 
2.43.0

[PATCH v3 02/11] net/intel: use common AVX build code

[Bruce Richardson]

Remove driver-specific build instructions for the AVX2 and AVX-512 code,
and rely instead on the generic driver build file.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/intel/i40e/meson.build | 26 ++------------------------
 drivers/net/intel/iavf/meson.build | 25 ++-----------------------
 drivers/net/intel/ice/meson.build  | 25 ++-----------------------
 drivers/net/intel/idpf/meson.build | 25 ++-----------------------
 4 files changed, 8 insertions(+), 93 deletions(-)

diff --git a/drivers/net/intel/i40e/meson.build b/drivers/net/intel/i40e/meson.build
index 2973ed1a01..25a3d72714 100644
--- a/drivers/net/intel/i40e/meson.build
+++ b/drivers/net/intel/i40e/meson.build
@@ -40,35 +40,13 @@ includes += include_directories('base')
 
 if arch_subdir == 'x86'
     sources += files('i40e_rxtx_vec_sse.c')
+    sources_avx2 = files('i40e_rxtx_vec_avx2.c')
+    sources_avx512 = files('i40e_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    i40e_avx2_lib = static_library('i40e_avx2_lib',
-            'i40e_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_kvargs, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += i40e_avx2_lib.extract_objects('i40e_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        i40e_avx512_lib = static_library('i40e_avx512_lib',
-                'i40e_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev,
-                    static_rte_kvargs, static_rte_hash],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += i40e_avx512_lib.extract_objects('i40e_rxtx_vec_avx512.c')
-    endif
 elif arch_subdir == 'ppc'
        sources += files('i40e_rxtx_vec_altivec.c')
 elif arch_subdir == 'arm'
diff --git a/drivers/net/intel/iavf/meson.build b/drivers/net/intel/iavf/meson.build
index f7eac7c57a..fc3e4d2f28 100644
--- a/drivers/net/intel/iavf/meson.build
+++ b/drivers/net/intel/iavf/meson.build
@@ -28,34 +28,13 @@ includes += include_directories('base')
 
 if arch_subdir == 'x86'
     sources += files('iavf_rxtx_vec_sse.c')
+    sources_avx2 = files('iavf_rxtx_vec_avx2.c')
+    sources_avx512 = files('iavf_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    iavf_avx2_lib = static_library('iavf_avx2_lib',
-            'iavf_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += iavf_avx2_lib.extract_objects('iavf_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        iavf_avx512_lib = static_library('iavf_avx512_lib',
-                'iavf_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += iavf_avx512_lib.extract_objects('iavf_rxtx_vec_avx512.c')
-    endif
 elif arch_subdir == 'arm'
     sources += files('iavf_rxtx_vec_neon.c')
 endif
diff --git a/drivers/net/intel/ice/meson.build b/drivers/net/intel/ice/meson.build
index cbdf38c1c4..a33ec59272 100644
--- a/drivers/net/intel/ice/meson.build
+++ b/drivers/net/intel/ice/meson.build
@@ -34,34 +34,13 @@ endif
 
 if arch_subdir == 'x86'
     sources += files('ice_rxtx_vec_sse.c')
+    sources_avx2 = files('ice_rxtx_vec_avx2.c')
+    sources_avx512 = files('ice_rxtx_vec_avx512.c')
 
     if is_windows and cc.get_id() != 'clang'
         cflags += ['-fno-asynchronous-unwind-tables']
     endif
 
-    ice_avx2_lib = static_library('ice_avx2_lib',
-            'ice_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += ice_avx2_lib.extract_objects('ice_rxtx_vec_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        ice_avx512_lib = static_library('ice_avx512_lib',
-                'ice_rxtx_vec_avx512.c',
-                dependencies: [static_rte_ethdev, static_rte_hash],
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += ice_avx512_lib.extract_objects('ice_rxtx_vec_avx512.c')
-    endif
 endif
 
 sources += files(
diff --git a/drivers/net/intel/idpf/meson.build b/drivers/net/intel/idpf/meson.build
index 4b272d02b1..10465f0f36 100644
--- a/drivers/net/intel/idpf/meson.build
+++ b/drivers/net/intel/idpf/meson.build
@@ -19,29 +19,8 @@ sources = files(
 )
 
 if arch_subdir == 'x86' and dpdk_conf.get('RTE_IOVA_IN_MBUF') == 1
-    idpf_avx2_lib = static_library('idpf_avx2_lib',
-        'idpf_common_rxtx_avx2.c',
-        dependencies: [static_rte_ethdev, static_rte_hash],
-        include_directories: includes,
-        c_args: [cflags, cc_avx2_flags])
-    objs += idpf_avx2_lib.extract_objects('idpf_common_rxtx_avx2.c')
-
-    if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
-        avx512_args = cflags + cc_avx512_flags
-        if cc.has_argument('-march=skylake-avx512')
-            avx512_args += '-march=skylake-avx512'
-            if cc.has_argument('-Wno-overriding-option')
-                avx512_args += '-Wno-overriding-option'
-            endif
-        endif
-        idpf_common_avx512_lib = static_library('idpf_common_avx512_lib',
-                'idpf_common_rxtx_avx512.c',
-                dependencies: static_rte_mbuf,
-                include_directories: includes,
-                c_args: avx512_args)
-        objs += idpf_common_avx512_lib.extract_objects('idpf_common_rxtx_avx512.c')
-    endif
+    sources_avx2 = files('idpf_common_rxtx_avx2.c')
+    sources_avx512 = files('idpf_common_rxtx_avx512.c')
 endif
 
 subdir('base')
-- 
2.43.0

[PATCH v3 03/11] drivers/net: build use common AVX handling

[Bruce Richardson]

Remove from remaining net drivers the special-case code to handle AVX2
or AVX512 specific files. These can be built instead using
drivers/meson.build.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/bnxt/meson.build      | 10 +---------
 drivers/net/enic/meson.build      | 10 +---------
 drivers/net/nfp/meson.build       | 10 +---------
 drivers/net/octeon_ep/meson.build | 14 ++------------
 drivers/net/virtio/meson.build    |  9 +--------
 5 files changed, 6 insertions(+), 47 deletions(-)

diff --git a/drivers/net/bnxt/meson.build b/drivers/net/bnxt/meson.build
index fd82d0c409..dcca7df916 100644
--- a/drivers/net/bnxt/meson.build
+++ b/drivers/net/bnxt/meson.build
@@ -58,15 +58,7 @@ subdir('hcapi/cfa_v3')
 
 if arch_subdir == 'x86'
     sources += files('bnxt_rxtx_vec_sse.c')
-    # build AVX2 code with instruction set explicitly enabled for runtime selection
-    bnxt_avx2_lib = static_library('bnxt_avx2_lib',
-            'bnxt_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev,
-                static_rte_bus_pci,
-                static_rte_kvargs, static_rte_hash],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-     objs += bnxt_avx2_lib.extract_objects('bnxt_rxtx_vec_avx2.c')
+    sources_avx2 = files('bnxt_rxtx_vec_avx2.c')
 elif arch_subdir == 'arm' and dpdk_conf.get('RTE_ARCH_64')
     sources += files('bnxt_rxtx_vec_neon.c')
 endif
diff --git a/drivers/net/enic/meson.build b/drivers/net/enic/meson.build
index cfe5ec170a..2b3052fae8 100644
--- a/drivers/net/enic/meson.build
+++ b/drivers/net/enic/meson.build
@@ -29,17 +29,9 @@ sources = files(
 deps += ['hash']
 includes += include_directories('base')
 
-# Build the avx2 handler for 64-bit X86 targets, even though 'machine'
-# may not. This is to support users who build for the min supported machine
-# and need to run the binary on newer CPUs too.
 if dpdk_conf.has('RTE_ARCH_X86_64')
     cflags += '-DENIC_RXTX_VEC'
-    enic_avx2_lib = static_library('enic_avx2_lib',
-            'enic_rxtx_vec_avx2.c',
-            dependencies: [static_rte_ethdev, static_rte_bus_pci],
-            include_directories: includes,
-            c_args: [cflags, cc_avx2_flags])
-    objs += enic_avx2_lib.extract_objects('enic_rxtx_vec_avx2.c')
+    sources_avx2 = files('enic_rxtx_vec_avx2.c')
 endif
 
 annotate_locks = false
diff --git a/drivers/net/nfp/meson.build b/drivers/net/nfp/meson.build
index 0a12b7dce7..a98b584042 100644
--- a/drivers/net/nfp/meson.build
+++ b/drivers/net/nfp/meson.build
@@ -52,19 +52,11 @@ cflags += no_wvla_cflag
 if arch_subdir == 'x86'
         includes += include_directories('../../common/nfp')
 
-        avx2_sources = files(
+        sources_avx2 = files(
                 'nfdk/nfp_nfdk_vec_avx2_dp.c',
                 'nfp_rxtx_vec_avx2.c',
         )
 
-        nfp_avx2_lib = static_library('nfp_avx2_lib',
-                avx2_sources,
-                dependencies: [static_rte_ethdev, static_rte_bus_pci],
-                include_directories: includes,
-                c_args: [cflags, cc_avx2_flags]
-        )
-
-        objs += nfp_avx2_lib.extract_all_objects(recursive: true)
 else
         sources += files(
                 'nfp_rxtx_vec_stub.c',
diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
index 1b34db3edc..9bf4627894 100644
--- a/drivers/net/octeon_ep/meson.build
+++ b/drivers/net/octeon_ep/meson.build
@@ -15,18 +15,8 @@ sources = files(
 
 if arch_subdir == 'x86'
     sources += files('cnxk_ep_rx_sse.c')
-    if cc.get_define('__AVX2__', args: machine_args) != ''
-        cflags += ['-DCC_AVX2_SUPPORT']
-        sources += files('cnxk_ep_rx_avx.c')
-    elif cc.has_multi_arguments(cc_avx2_flags)
-        cflags += ['-DCC_AVX2_SUPPORT']
-        otx_ep_avx2_lib = static_library('otx_ep_avx2_lib',
-                        'cnxk_ep_rx_avx.c',
-                        dependencies: [static_rte_ethdev, static_rte_pci, static_rte_bus_pci],
-                        include_directories: includes,
-                        c_args: [cflags, cc_avx2_flags])
-        objs += otx_ep_avx2_lib.extract_objects('cnxk_ep_rx_avx.c')
-    endif
+    cflags += ['-DCC_AVX2_SUPPORT']
+    sources_avx2 = files('cnxk_ep_rx_avx.c')
 endif
 
 if arch_subdir == 'arm'
diff --git a/drivers/net/virtio/meson.build b/drivers/net/virtio/meson.build
index c1c4a85bea..01bfb3c47d 100644
--- a/drivers/net/virtio/meson.build
+++ b/drivers/net/virtio/meson.build
@@ -27,15 +27,8 @@ cflags += no_wvla_cflag
 
 if arch_subdir == 'x86'
     if cc_has_avx512
-        cflags += ['-DCC_AVX512_SUPPORT']
         cflags += ['-DVIRTIO_RXTX_PACKED_VEC']
-        virtio_avx512_lib = static_library('virtio_avx512_lib',
-                'virtio_rxtx_packed.c',
-                dependencies: [static_rte_ethdev,
-                    static_rte_kvargs, static_rte_bus_pci],
-                include_directories: includes,
-                c_args: cflags + cc_avx512_flags)
-        objs += virtio_avx512_lib.extract_objects('virtio_rxtx_packed.c')
+        sources_avx512 = files('virtio_rxtx_packed.c')
         if (toolchain == 'gcc' and cc.version().version_compare('>=8.3.0'))
             cflags += '-DVIRTIO_GCC_UNROLL_PRAGMA'
         elif (toolchain == 'clang' and cc.version().version_compare('>=3.7.0'))
-- 
2.43.0

[PATCH v3 04/11] drivers/net: remove AVX2 build-time define

[Bruce Richardson]

Since all supported compilers can generate AVX2 code, we will always
enable the build of the AVX2 files on x86. This means that
CC_AVX2_SUPPORT is always true on x86, so it can be removed and a
regular "#ifdef RTE_ARCH_x86" used in its place.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/bnxt/bnxt_ethdev.c        | 2 --
 drivers/net/octeon_ep/meson.build     | 1 -
 drivers/net/octeon_ep/otx_ep_ethdev.c | 4 ----
 3 files changed, 7 deletions(-)

diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
index 144d4377bd..a318604c20 100644
--- a/drivers/net/bnxt/bnxt_ethdev.c
+++ b/drivers/net/bnxt/bnxt_ethdev.c
@@ -3238,8 +3238,6 @@ static const struct {
 #if defined(RTE_ARCH_X86)
 	{bnxt_crx_pkts_vec,		"Vector SSE"},
 	{bnxt_recv_pkts_vec,		"Vector SSE"},
-#endif
-#if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
 	{bnxt_crx_pkts_vec_avx2,	"Vector AVX2"},
 	{bnxt_recv_pkts_vec_avx2,	"Vector AVX2"},
 #endif
diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
index 9bf4627894..a4a7663d1d 100644
--- a/drivers/net/octeon_ep/meson.build
+++ b/drivers/net/octeon_ep/meson.build
@@ -15,7 +15,6 @@ sources = files(
 
 if arch_subdir == 'x86'
     sources += files('cnxk_ep_rx_sse.c')
-    cflags += ['-DCC_AVX2_SUPPORT']
     sources_avx2 = files('cnxk_ep_rx_avx.c')
 endif
 
diff --git a/drivers/net/octeon_ep/otx_ep_ethdev.c b/drivers/net/octeon_ep/otx_ep_ethdev.c
index 8b14734b0c..10f2f8a2e0 100644
--- a/drivers/net/octeon_ep/otx_ep_ethdev.c
+++ b/drivers/net/octeon_ep/otx_ep_ethdev.c
@@ -91,11 +91,9 @@ otx_ep_set_rx_func(struct rte_eth_dev *eth_dev)
 		eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts;
 #ifdef RTE_ARCH_X86
 		eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_sse;
-#ifdef CC_AVX2_SUPPORT
 		if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
 		    rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1)
 			eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_avx;
-#endif
 #elif defined(RTE_ARCH_ARM64)
 		eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_neon;
 #endif
@@ -105,11 +103,9 @@ otx_ep_set_rx_func(struct rte_eth_dev *eth_dev)
 		eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts;
 #ifdef RTE_ARCH_X86
 		eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_sse;
-#ifdef CC_AVX2_SUPPORT
 		if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
 		    rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1)
 			eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_avx;
-#endif
 #elif defined(RTE_ARCH_ARM64)
 		eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_neon;
 #endif
-- 
2.43.0

[PATCH v3 05/11] event/dlb2: build using common AVX handling

[Bruce Richardson]

remove special-case handling for AVX512, and rely on mechanisms in
the drivers meson.build file.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/event/dlb2/dlb2_sse.c  |  4 ++++
 drivers/event/dlb2/meson.build | 16 ++--------------
 2 files changed, 6 insertions(+), 14 deletions(-)

diff --git a/drivers/event/dlb2/dlb2_sse.c b/drivers/event/dlb2/dlb2_sse.c
index f2e1f9fb7e..06474d61dd 100644
--- a/drivers/event/dlb2/dlb2_sse.c
+++ b/drivers/event/dlb2/dlb2_sse.c
@@ -5,6 +5,8 @@
 #include <stdint.h>
 #include <stdbool.h>
 
+#ifndef CC_AVX512_SUPPORT
+
 #include "dlb2_priv.h"
 #include "dlb2_iface.h"
 #include "dlb2_inline_fns.h"
@@ -226,3 +228,5 @@ dlb2_event_build_hcws(struct dlb2_port *qm_port,
 		break;
 	}
 }
+
+#endif /* no CC_AVX512_SUPPORT */
diff --git a/drivers/event/dlb2/meson.build b/drivers/event/dlb2/meson.build
index c024edb311..13d0fa544e 100644
--- a/drivers/event/dlb2/meson.build
+++ b/drivers/event/dlb2/meson.build
@@ -20,22 +20,10 @@ sources = files(
         'pf/base/dlb2_resource.c',
         'rte_pmd_dlb2.c',
         'dlb2_selftest.c',
+        'dlb2_sse.c',
 )
 
-if target_has_avx512
-    cflags += '-DCC_AVX512_SUPPORT'
-    sources += files('dlb2_avx512.c')
-
-elif cc_has_avx512
-    cflags += '-DCC_AVX512_SUPPORT'
-    avx512_tmplib = static_library('avx512_tmp',
-                               'dlb2_avx512.c',
-                               dependencies: [static_rte_eal, static_rte_eventdev],
-                               c_args: cflags + cc_avx512_flags)
-    objs += avx512_tmplib.extract_objects('dlb2_avx512.c')
-else
-    sources += files('dlb2_sse.c')
-endif
+sources_avx512 += files('dlb2_avx512.c')
 
 headers = files('rte_pmd_dlb2.h')
 
-- 
2.43.0

[PATCH v3 06/11] build: add generalized AVX handling for libs

[Bruce Richardson]

Add support to the top-level lib build file for AVX2 and AVX512
specific sources. This should simplify library builds by avoiding the
need to constantly reimplement the same build logic

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/meson.build | 34 +++++++++++++++++++++++++++++++++-
 1 file changed, 33 insertions(+), 1 deletion(-)

diff --git a/lib/meson.build b/lib/meson.build
index ce92cb5537..30ac84263c 100644
--- a/lib/meson.build
+++ b/lib/meson.build
@@ -122,6 +122,9 @@ foreach l:libraries
     use_function_versioning = false
     annotate_locks = true
     sources = []
+    sources_avx2 = []
+    sources_avx512 = []
+    cflags_avx512 = [] # extra cflags for the avx512 code, e.g. extra avx512 feature flags
     headers = []
     indirect_headers = [] # public headers not directly included by apps
     driver_sdk_headers = [] # public headers included by drivers
@@ -242,7 +245,36 @@ foreach l:libraries
         cflags += '-Wthread-safety'
     endif
 
-    # first build static lib
+    # handle avx2 and avx512 source files
+    if arch_subdir == 'x86'
+        if sources_avx2.length() > 0
+            avx2_lib = static_library(libname + '_avx2_lib',
+                    sources_avx2,
+                    dependencies: static_deps,
+                    include_directories: includes,
+                    c_args: [cflags, cc_avx2_flags])
+            objs += avx2_lib.extract_objects(sources_avx2)
+        endif
+        if sources_avx512.length() > 0 and cc_has_avx512
+            cflags += '-DCC_AVX512_SUPPORT'
+            avx512_args = [cflags, cflags_avx512, cc_avx512_flags]
+            if cc.has_argument('-march=skylake-avx512')
+                avx512_args += '-march=skylake-avx512'
+                if cc.has_argument('-Wno-overriding-option')
+                    avx512_args += '-Wno-overriding-option'
+                endif
+            endif
+            avx512_lib = static_library(libname + '_avx512_lib',
+                    sources_avx512,
+                    dependencies: static_deps,
+                    include_directories: includes,
+                    c_args: avx512_args)
+            objs += avx512_lib.extract_objects(sources_avx512)
+        endif
+    endif
+
+
+    # build static lib
     static_lib = static_library(libname,
             sources,
             objects: objs,
-- 
2.43.0

[PATCH v3 07/11] acl: use common AVX build handling

[Bruce Richardson]

remove custom logic for building AVX2 and AVX-512 files.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/acl/meson.build | 54 ++++-----------------------------------------
 1 file changed, 4 insertions(+), 50 deletions(-)

diff --git a/lib/acl/meson.build b/lib/acl/meson.build
index a80c172812..87e9f25f8e 100644
--- a/lib/acl/meson.build
+++ b/lib/acl/meson.build
@@ -15,57 +15,11 @@ headers = files('rte_acl.h', 'rte_acl_osdep.h')
 
 if dpdk_conf.has('RTE_ARCH_X86')
     sources += files('acl_run_sse.c')
-
-    avx2_tmplib = static_library('avx2_tmp',
-            'acl_run_avx2.c',
-            dependencies: static_rte_eal,
-            c_args: [cflags, cc_avx2_flags])
-    objs += avx2_tmplib.extract_objects('acl_run_avx2.c')
-
-    # compile AVX512 version if:
-    # we are building 64-bit binary AND binutils can generate proper code
-
-    if dpdk_conf.has('RTE_ARCH_X86_64') and binutils_ok
-
-        # compile AVX512 version if either:
-        # a. we have AVX512 supported in minimum instruction set
-        #    baseline
-        # b. it's not minimum instruction set, but supported by
-        #    compiler
-        #
-        # in former case, just add avx512 C file to files list
-        # in latter case, compile c file to static lib, using correct
-        # compiler flags, and then have the .o file from static lib
-        # linked into main lib.
-
-        # check if all required flags already enabled (variant a).
-        acl_avx512_flags = ['__AVX512F__', '__AVX512VL__',
-            '__AVX512CD__', '__AVX512BW__']
-
-        acl_avx512_on = true
-        foreach f:acl_avx512_flags
-
-            if cc.get_define(f, args: machine_args) == ''
-                acl_avx512_on = false
-            endif
-        endforeach
-
-        if acl_avx512_on == true
-
-            sources += files('acl_run_avx512.c')
-            cflags += '-DCC_AVX512_SUPPORT'
-
-        elif cc_has_avx512
-            avx512_tmplib = static_library('avx512_tmp',
-                'acl_run_avx512.c',
-                dependencies: static_rte_eal,
-                c_args: cflags + cc_avx512_flags)
-            objs += avx512_tmplib.extract_objects(
-                    'acl_run_avx512.c')
-            cflags += '-DCC_AVX512_SUPPORT'
-        endif
+    sources_avx2 += files('acl_run_avx2.c')
+    # AVX512 is only supported on 64-bit builds
+    if dpdk_conf.has('RTE_ARCH_X86_64')
+        sources_avx512 += files('acl_run_avx512.c')
     endif
-
 elif dpdk_conf.has('RTE_ARCH_ARM')
     cflags += '-flax-vector-conversions'
     sources += files('acl_run_neon.c')
-- 
2.43.0

[PATCH v3 08/11] fib: use common AVX build handling

[Bruce Richardson]

Remove custom logic for building AVX2 and AVX-512 files. Within the C
code this requires some renaming of build macros to use the standard
defines.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/fib/dir24_8.c   |  6 +++---
 lib/fib/meson.build | 18 +-----------------
 lib/fib/trie.c      |  6 +++---
 3 files changed, 7 insertions(+), 23 deletions(-)

diff --git a/lib/fib/dir24_8.c b/lib/fib/dir24_8.c
index c48d962d41..2ba7e93511 100644
--- a/lib/fib/dir24_8.c
+++ b/lib/fib/dir24_8.c
@@ -16,11 +16,11 @@
 #include "dir24_8.h"
 #include "fib_log.h"
 
-#ifdef CC_DIR24_8_AVX512_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 
 #include "dir24_8_avx512.h"
 
-#endif /* CC_DIR24_8_AVX512_SUPPORT */
+#endif /* CC_AVX512_SUPPORT */
 
 #define DIR24_8_NAMESIZE	64
 
@@ -63,7 +63,7 @@ get_scalar_fn_inlined(enum rte_fib_dir24_8_nh_sz nh_sz, bool be_addr)
 static inline rte_fib_lookup_fn_t
 get_vector_fn(enum rte_fib_dir24_8_nh_sz nh_sz, bool be_addr)
 {
-#ifdef CC_DIR24_8_AVX512_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 	if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) <= 0 ||
 			rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512DQ) <= 0 ||
 			rte_vect_get_max_simd_bitwidth() < RTE_VECT_SIMD_512)
diff --git a/lib/fib/meson.build b/lib/fib/meson.build
index 0c19cc8201..55d9ddcee9 100644
--- a/lib/fib/meson.build
+++ b/lib/fib/meson.build
@@ -15,21 +15,5 @@ deps += ['rcu']
 deps += ['net']
 
 if dpdk_conf.has('RTE_ARCH_X86_64')
-    if target_has_avx512
-        cflags += ['-DCC_DIR24_8_AVX512_SUPPORT', '-DCC_TRIE_AVX512_SUPPORT']
-        sources += files('dir24_8_avx512.c', 'trie_avx512.c')
-
-    elif cc_has_avx512
-        cflags += ['-DCC_DIR24_8_AVX512_SUPPORT', '-DCC_TRIE_AVX512_SUPPORT']
-        dir24_8_avx512_tmp = static_library('dir24_8_avx512_tmp',
-                'dir24_8_avx512.c',
-                dependencies: [static_rte_eal, static_rte_rcu],
-                c_args: cflags + cc_avx512_flags)
-        objs += dir24_8_avx512_tmp.extract_objects('dir24_8_avx512.c')
-        trie_avx512_tmp = static_library('trie_avx512_tmp',
-                'trie_avx512.c',
-                dependencies: [static_rte_eal, static_rte_rcu, static_rte_net],
-                c_args: cflags + cc_avx512_flags)
-        objs += trie_avx512_tmp.extract_objects('trie_avx512.c')
-    endif
+    sources_avx512 = files('dir24_8_avx512.c', 'trie_avx512.c')
 endif
diff --git a/lib/fib/trie.c b/lib/fib/trie.c
index 4893f6c636..6c20057ac5 100644
--- a/lib/fib/trie.c
+++ b/lib/fib/trie.c
@@ -14,11 +14,11 @@
 #include <rte_fib6.h>
 #include "trie.h"
 
-#ifdef CC_TRIE_AVX512_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 
 #include "trie_avx512.h"
 
-#endif /* CC_TRIE_AVX512_SUPPORT */
+#endif /* CC_AVX512_SUPPORT */
 
 #define TRIE_NAMESIZE		64
 
@@ -45,7 +45,7 @@ get_scalar_fn(enum rte_fib_trie_nh_sz nh_sz)
 static inline rte_fib6_lookup_fn_t
 get_vector_fn(enum rte_fib_trie_nh_sz nh_sz)
 {
-#ifdef CC_TRIE_AVX512_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 	if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) <= 0 ||
 			rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512DQ) <= 0 ||
 			rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512BW) <= 0 ||
-- 
2.43.0

[PATCH v3 09/11] net: simplify build-time logic for x86

[Bruce Richardson]

All DPDK-supported versions of clang and gcc have the "-mpclmul" and
"-maes" flags, so we never need to check for those. This allows the SSE
code path to be unconditionally built on x86.

For the AVX512 code path, simplify it by only checking for the
build-time support, and always doing a separate build with AVX512
support when that compiler support is present.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/net/meson.build   | 53 +++++--------------------------------------
 lib/net/rte_net_crc.c |  8 +++----
 2 files changed, 10 insertions(+), 51 deletions(-)

diff --git a/lib/net/meson.build b/lib/net/meson.build
index c9b34afc98..4bbbad3f42 100644
--- a/lib/net/meson.build
+++ b/lib/net/meson.build
@@ -42,57 +42,16 @@ deps += ['mbuf']
 use_function_versioning = true
 
 if dpdk_conf.has('RTE_ARCH_X86_64')
-    net_crc_sse42_cpu_support = (cc.get_define('__PCLMUL__', args: machine_args) != '')
-    net_crc_avx512_cpu_support = (
-            target_has_avx512 and
-            cc.get_define('__VPCLMULQDQ__', args: machine_args) != ''
-    )
-
-    net_crc_sse42_cc_support = (cc.has_argument('-mpclmul') and cc.has_argument('-maes'))
-    net_crc_avx512_cc_support = (cc.has_argument('-mvpclmulqdq') and cc_has_avx512)
-
-    build_static_net_crc_sse42_lib = 0
-    build_static_net_crc_avx512_lib = 0
-
-    if net_crc_sse42_cpu_support == true
-        sources += files('net_crc_sse.c')
-        cflags += ['-DCC_X86_64_SSE42_PCLMULQDQ_SUPPORT']
-        if net_crc_avx512_cpu_support == true
-            sources += files('net_crc_avx512.c')
-            cflags += ['-DCC_X86_64_AVX512_VPCLMULQDQ_SUPPORT']
-        elif net_crc_avx512_cc_support == true
-            build_static_net_crc_avx512_lib = 1
-            net_crc_avx512_lib_cflags = cc_avx512_flags + ['-mvpclmulqdq']
-            cflags += ['-DCC_X86_64_AVX512_VPCLMULQDQ_SUPPORT']
-        endif
-    elif net_crc_sse42_cc_support == true
-        build_static_net_crc_sse42_lib = 1
-        net_crc_sse42_lib_cflags = ['-mpclmul', '-maes']
-        cflags += ['-DCC_X86_64_SSE42_PCLMULQDQ_SUPPORT']
-        if net_crc_avx512_cc_support == true
-            build_static_net_crc_avx512_lib = 1
-            net_crc_avx512_lib_cflags = cc_avx512_flags + ['-mvpclmulqdq', '-mpclmul']
-            cflags += ['-DCC_X86_64_AVX512_VPCLMULQDQ_SUPPORT']
-        endif
-    endif
-
-    if build_static_net_crc_sse42_lib == 1
-        net_crc_sse42_lib = static_library(
-                'net_crc_sse42_lib',
-                'net_crc_sse.c',
-                dependencies: static_rte_eal,
-                c_args: [cflags,
-                    net_crc_sse42_lib_cflags])
-        objs += net_crc_sse42_lib.extract_objects('net_crc_sse.c')
-    endif
-
-    if build_static_net_crc_avx512_lib == 1
+    sources += files('net_crc_sse.c')
+    cflags += ['-mpclmul', '-maes']
+    if cc.has_argument('-mvpclmulqdq') and cc_has_avx512
+        net_crc_avx512_lib_cflags = cc_avx512_flags + ['-mvpclmulqdq']
+        cflags += ['-DCC_X86_64_AVX512_VPCLMULQDQ_SUPPORT']
         net_crc_avx512_lib = static_library(
                 'net_crc_avx512_lib',
                 'net_crc_avx512.c',
                 dependencies: static_rte_eal,
-                c_args: [cflags,
-                    net_crc_avx512_lib_cflags])
+                c_args: [cflags, net_crc_avx512_lib_cflags])
         objs += net_crc_avx512_lib.extract_objects('net_crc_avx512.c')
     endif
 
diff --git a/lib/net/rte_net_crc.c b/lib/net/rte_net_crc.c
index 2fb3eec231..c9773d6300 100644
--- a/lib/net/rte_net_crc.c
+++ b/lib/net/rte_net_crc.c
@@ -66,7 +66,7 @@ static const rte_net_crc_handler handlers_avx512[] = {
 	[RTE_NET_CRC32_ETH] = rte_crc32_eth_avx512_handler,
 };
 #endif
-#ifdef CC_X86_64_SSE42_PCLMULQDQ_SUPPORT
+#ifdef RTE_ARCH_X86_64
 static const rte_net_crc_handler handlers_sse42[] = {
 	[RTE_NET_CRC16_CCITT] = rte_crc16_ccitt_sse42_handler,
 	[RTE_NET_CRC32_ETH] = rte_crc32_eth_sse42_handler,
@@ -211,7 +211,7 @@ avx512_vpclmulqdq_init(void)
 static const rte_net_crc_handler *
 sse42_pclmulqdq_get_handlers(void)
 {
-#ifdef CC_X86_64_SSE42_PCLMULQDQ_SUPPORT
+#ifdef RTE_ARCH_X86_64
 	if (SSE42_PCLMULQDQ_CPU_SUPPORTED &&
 			max_simd_bitwidth >= RTE_VECT_SIMD_128)
 		return handlers_sse42;
@@ -223,7 +223,7 @@ sse42_pclmulqdq_get_handlers(void)
 static void
 sse42_pclmulqdq_init(void)
 {
-#ifdef CC_X86_64_SSE42_PCLMULQDQ_SUPPORT
+#ifdef RTE_ARCH_x86_64
 	if (SSE42_PCLMULQDQ_CPU_SUPPORTED)
 		rte_net_crc_sse42_init();
 #endif
@@ -316,7 +316,7 @@ handlers_init(enum rte_net_crc_alg alg)
 #endif
 		/* fall-through */
 	case RTE_NET_CRC_SSE42:
-#ifdef CC_X86_64_SSE42_PCLMULQDQ_SUPPORT
+#ifdef RTE_ARCH_X86_64
 		if (SSE42_PCLMULQDQ_CPU_SUPPORTED) {
 			handlers_dpdk26[alg].f[RTE_NET_CRC16_CCITT] =
 				rte_crc16_ccitt_sse42_handler;
-- 
2.43.0

[PATCH v3 10/11] net: use common AVX512 build code

[Bruce Richardson]

Use the common support for AVX512 code present in lib/meson.build,
rather than hard-coding it. The only complication is an extra check for
the "-mvpclmulqdq" command-line flag before adding the AVX512 sources.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/net/meson.build   | 13 ++++---------
 lib/net/rte_net_crc.c |  8 ++++----
 2 files changed, 8 insertions(+), 13 deletions(-)

diff --git a/lib/net/meson.build b/lib/net/meson.build
index 4bbbad3f42..7a6c419f40 100644
--- a/lib/net/meson.build
+++ b/lib/net/meson.build
@@ -44,15 +44,10 @@ use_function_versioning = true
 if dpdk_conf.has('RTE_ARCH_X86_64')
     sources += files('net_crc_sse.c')
     cflags += ['-mpclmul', '-maes']
-    if cc.has_argument('-mvpclmulqdq') and cc_has_avx512
-        net_crc_avx512_lib_cflags = cc_avx512_flags + ['-mvpclmulqdq']
-        cflags += ['-DCC_X86_64_AVX512_VPCLMULQDQ_SUPPORT']
-        net_crc_avx512_lib = static_library(
-                'net_crc_avx512_lib',
-                'net_crc_avx512.c',
-                dependencies: static_rte_eal,
-                c_args: [cflags, net_crc_avx512_lib_cflags])
-        objs += net_crc_avx512_lib.extract_objects('net_crc_avx512.c')
+    # only build AVX-512 support if we also have PCLMULQDQ support
+    if cc.has_argument('-mvpclmulqdq')
+        sources_avx512 += files('net_crc_avx512.c')
+        cflags_avx512 += ['-mvpclmulqdq']
     endif
 
 elif (dpdk_conf.has('RTE_ARCH_ARM64') and
diff --git a/lib/net/rte_net_crc.c b/lib/net/rte_net_crc.c
index c9773d6300..24ec267fc8 100644
--- a/lib/net/rte_net_crc.c
+++ b/lib/net/rte_net_crc.c
@@ -60,7 +60,7 @@ static const rte_net_crc_handler handlers_scalar[] = {
 	[RTE_NET_CRC16_CCITT] = rte_crc16_ccitt_handler,
 	[RTE_NET_CRC32_ETH] = rte_crc32_eth_handler,
 };
-#ifdef CC_X86_64_AVX512_VPCLMULQDQ_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 static const rte_net_crc_handler handlers_avx512[] = {
 	[RTE_NET_CRC16_CCITT] = rte_crc16_ccitt_avx512_handler,
 	[RTE_NET_CRC32_ETH] = rte_crc32_eth_avx512_handler,
@@ -185,7 +185,7 @@ rte_crc32_eth_handler(const uint8_t *data, uint32_t data_len)
 static const rte_net_crc_handler *
 avx512_vpclmulqdq_get_handlers(void)
 {
-#ifdef CC_X86_64_AVX512_VPCLMULQDQ_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 	if (AVX512_VPCLMULQDQ_CPU_SUPPORTED &&
 			max_simd_bitwidth >= RTE_VECT_SIMD_512)
 		return handlers_avx512;
@@ -197,7 +197,7 @@ avx512_vpclmulqdq_get_handlers(void)
 static void
 avx512_vpclmulqdq_init(void)
 {
-#ifdef CC_X86_64_AVX512_VPCLMULQDQ_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 	if (AVX512_VPCLMULQDQ_CPU_SUPPORTED)
 		rte_net_crc_avx512_init();
 #endif
@@ -305,7 +305,7 @@ handlers_init(enum rte_net_crc_alg alg)
 
 	switch (alg) {
 	case RTE_NET_CRC_AVX512:
-#ifdef CC_X86_64_AVX512_VPCLMULQDQ_SUPPORT
+#ifdef CC_AVX512_SUPPORT
 		if (AVX512_VPCLMULQDQ_CPU_SUPPORTED) {
 			handlers_dpdk26[alg].f[RTE_NET_CRC16_CCITT] =
 				rte_crc16_ccitt_avx512_handler;
-- 
2.43.0

[PATCH v3 11/11] member: use common AVX512 build support

[Bruce Richardson]

Use the support for building AVX512 code present in lib/meson.build
rather than reimplementing it in the library meson.build file.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 lib/member/meson.build | 46 +++++++-----------------------------------
 1 file changed, 7 insertions(+), 39 deletions(-)

diff --git a/lib/member/meson.build b/lib/member/meson.build
index 4341b424df..07f9afaed9 100644
--- a/lib/member/meson.build
+++ b/lib/member/meson.build
@@ -20,44 +20,12 @@ sources = files(
 
 deps += ['hash', 'ring']
 
-# compile AVX512 version if:
-if dpdk_conf.has('RTE_ARCH_X86_64') and binutils_ok
-    # compile AVX512 version if either:
-    # a. we have AVX512 supported in minimum instruction set
-    #    baseline
-    # b. it's not minimum instruction set, but supported by
-    #    compiler
-    #
-    # in former case, just add avx512 C file to files list
-    # in latter case, compile c file to static lib, using correct
-    # compiler flags, and then have the .o file from static lib
-    # linked into main lib.
-
-    member_avx512_args = cc_avx512_flags
-    if not is_ms_compiler
-            member_avx512_args += '-mavx512ifma'
-    endif
-
-    # check if all required flags already enabled
-    sketch_avx512_flags = ['__AVX512F__', '__AVX512DQ__', '__AVX512IFMA__']
-
-    sketch_avx512_on = true
-    foreach f:sketch_avx512_flags
-        if cc.get_define(f, args: machine_args) == ''
-            sketch_avx512_on = false
-        endif
-    endforeach
-
-    if sketch_avx512_on == true
-        cflags += ['-DCC_AVX512_SUPPORT']
-        sources += files('rte_member_sketch_avx512.c')
-    elif cc.has_multi_arguments(member_avx512_args)
-        sketch_avx512_tmp = static_library('sketch_avx512_tmp',
-            'rte_member_sketch_avx512.c',
-            include_directories: includes,
-            dependencies: [static_rte_eal, static_rte_hash],
-            c_args: cflags + member_avx512_args)
-        objs += sketch_avx512_tmp.extract_objects('rte_member_sketch_avx512.c')
-        cflags += ['-DCC_AVX512_SUPPORT']
+# compile AVX512 version if we have avx512 on MSVC or the 'ifma' flag on GCC/Clang
+if dpdk_conf.has('RTE_ARCH_X86_64')
+    if is_ms_compiler
+        sources_avx512 += files('rte_member_sketch_avx512.c')
+    elif cc.has_argument('-mavx512ifma')
+        sources_avx512 += files('rte_member_sketch_avx512.c')
+        cflags_avx512 += '-mavx512ifma'
     endif
 endif
-- 
2.43.0

Re: [PATCH v3 04/11] drivers/net: remove AVX2 build-time define

[Ajit Khaparde]

[-- Attachment #1: Type: text/plain, Size: 3143 bytes --]

On Tue, Mar 18, 2025 at 10:35 AM Bruce Richardson
<bruce.richardson@intel.com> wrote:
>
> Since all supported compilers can generate AVX2 code, we will always
> enable the build of the AVX2 files on x86. This means that
> CC_AVX2_SUPPORT is always true on x86, so it can be removed and a
> regular "#ifdef RTE_ARCH_x86" used in its place.
>
> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
For bnxt:
Acked-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
> ---
>  drivers/net/bnxt/bnxt_ethdev.c        | 2 --
>  drivers/net/octeon_ep/meson.build     | 1 -
>  drivers/net/octeon_ep/otx_ep_ethdev.c | 4 ----
>  3 files changed, 7 deletions(-)
>
> diff --git a/drivers/net/bnxt/bnxt_ethdev.c b/drivers/net/bnxt/bnxt_ethdev.c
> index 144d4377bd..a318604c20 100644
> --- a/drivers/net/bnxt/bnxt_ethdev.c
> +++ b/drivers/net/bnxt/bnxt_ethdev.c
> @@ -3238,8 +3238,6 @@ static const struct {
>  #if defined(RTE_ARCH_X86)
>         {bnxt_crx_pkts_vec,             "Vector SSE"},
>         {bnxt_recv_pkts_vec,            "Vector SSE"},
> -#endif
> -#if defined(RTE_ARCH_X86) && defined(CC_AVX2_SUPPORT)
>         {bnxt_crx_pkts_vec_avx2,        "Vector AVX2"},
>         {bnxt_recv_pkts_vec_avx2,       "Vector AVX2"},
>  #endif
> diff --git a/drivers/net/octeon_ep/meson.build b/drivers/net/octeon_ep/meson.build
> index 9bf4627894..a4a7663d1d 100644
> --- a/drivers/net/octeon_ep/meson.build
> +++ b/drivers/net/octeon_ep/meson.build
> @@ -15,7 +15,6 @@ sources = files(
>
>  if arch_subdir == 'x86'
>      sources += files('cnxk_ep_rx_sse.c')
> -    cflags += ['-DCC_AVX2_SUPPORT']
>      sources_avx2 = files('cnxk_ep_rx_avx.c')
>  endif
>
> diff --git a/drivers/net/octeon_ep/otx_ep_ethdev.c b/drivers/net/octeon_ep/otx_ep_ethdev.c
> index 8b14734b0c..10f2f8a2e0 100644
> --- a/drivers/net/octeon_ep/otx_ep_ethdev.c
> +++ b/drivers/net/octeon_ep/otx_ep_ethdev.c
> @@ -91,11 +91,9 @@ otx_ep_set_rx_func(struct rte_eth_dev *eth_dev)
>                 eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts;
>  #ifdef RTE_ARCH_X86
>                 eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_sse;
> -#ifdef CC_AVX2_SUPPORT
>                 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
>                     rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1)
>                         eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_avx;
> -#endif
>  #elif defined(RTE_ARCH_ARM64)
>                 eth_dev->rx_pkt_burst = &cnxk_ep_recv_pkts_neon;
>  #endif
> @@ -105,11 +103,9 @@ otx_ep_set_rx_func(struct rte_eth_dev *eth_dev)
>                 eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts;
>  #ifdef RTE_ARCH_X86
>                 eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_sse;
> -#ifdef CC_AVX2_SUPPORT
>                 if (rte_vect_get_max_simd_bitwidth() >= RTE_VECT_SIMD_256 &&
>                     rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1)
>                         eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_avx;
> -#endif
>  #elif defined(RTE_ARCH_ARM64)
>                 eth_dev->rx_pkt_burst = &cn9k_ep_recv_pkts_neon;
>  #endif
> --
> 2.43.0
>

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4205 bytes --]