Remove RTE_MACHINE_CPUFLAG_ macros
Subsystems (2)
cover Remove RTE_MACHINE_CPUFLAG_ macros
Remove RTE_MACHINE_CPUFLAG_ macros from the build. Deprecation notice sent, pasted here for reference:
build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example, RTE_MACHINE_CPUFLAG_SSE3 duplicates the compiler-provided macro __SSE3__.
Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference to RTE_MACHINE_CPUFLAG
app/test-pmd/macswap.c | 2 +- app/test/test_memcpy_perf.c | 8 ++++---- config/arm/meson.build | 6 ------ config/ppc/meson.build | 2 -- config/x86/meson.build | 2 -- doc/guides/prog_guide/writing_efficient_code.rst | 2 +- drivers/net/enic/Makefile | 2 +- drivers/net/enic/meson.build | 2 +- drivers/net/i40e/Makefile | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/Makefile | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/Makefile | 2 +- drivers/net/ice/meson.build | 2 +- drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 16 ++++++++-------- examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_acl/Makefile | 2 +- lib/librte_acl/meson.build | 2 +- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/Makefile | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 4 ++-- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 4 ++-- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- mk/rte.cpuflags.mk | 1 - 39 files changed, 58 insertions(+), 69 deletions(-)
Signature · 1 line
2.17.1
PATCH 1/4 x86: change cpuflag macros to compiler macros
Replace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros.
--- app/test/test_memcpy_perf.c | 8 ++++---- config/x86/meson.build | 2 -- drivers/net/enic/Makefile | 2 +- drivers/net/enic/meson.build | 2 +- drivers/net/i40e/Makefile | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/Makefile | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/Makefile | 2 +- drivers/net/ice/meson.build | 2 +- examples/l3fwd/l3fwd_em.c | 4 ++-- lib/librte_acl/Makefile | 2 +- lib/librte_acl/meson.build | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- mk/rte.cpuflags.mk | 1 - 21 files changed, 31 insertions(+), 34 deletions(-)
app/test/test_memcpy_perf.c +4 -4
@@ -51,13 +51,13 @@ static size_t buf_sizes[TEST_VALUE_RANGE];#define TEST_BATCH_SIZE 100/* Data is aligned on this many bytes (power of 2) */ -#ifdef RTE_MACHINE_CPUFLAG_AVX512F +#ifdef __AVX512F__#define ALIGNMENT_UNIT 64 -#elif definedRTE_MACHINE_CPUFLAG_AVX2 +#elif defined __AVX2__#define ALIGNMENT_UNIT 32 -#else /* RTE_MACHINE_CPUFLAG */ +#else #define ALIGNMENT_UNIT 16 -#endif /* RTE_MACHINE_CPUFLAG */ +#endif /* * Pointers used in performance tests. The two large buffers are for uncached
config/x86/meson.build +0 -2
@@ -18,7 +18,6 @@ endif base_flags = ['SSE', 'SSE2', 'SSE3','SSSE3', 'SSE4_1', 'SSE4_2'] foreach f:base_flags - dpdk_conf.set('RTE_MACHINE_CPUFLAG_' + f, 1) compile_time_cpuflags += ['RTE_CPUFLAG_' + f] endforeach
@@ -32,7 +31,6 @@ foreach f:optional_flags elif f == 'RDRND' f = 'RDRAND' endif - dpdk_conf.set('RTE_MACHINE_CPUFLAG_' + f, 1) compile_time_cpuflags += ['RTE_CPUFLAG_' + f] endif endforeach
drivers/net/enic/Makefile +1 -1
@@ -45,7 +45,7 @@ ifeq ($(CONFIG_RTE_ARCH_X86_64),y) # 'default' machine (corei7 which has no avx2) and run the binary on # newer CPUs that have avx2. # This part is verbatim from i40e makefile. -ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2) +ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),) CC_AVX2_SUPPORT=1 else CC_AVX2_SUPPORT=\
drivers/net/enic/meson.build +1 -1
@@ -19,7 +19,7 @@ deps += ['hash'] includes += include_directories('base') # The current implementation assumes 64-bit pointers -if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') and dpdk_conf.get('RTE_ARCH_64') +if cc.get_define('__AVX2__', args: machine_args) != '' and dpdk_conf.get('RTE_ARCH_64') sources += files('enic_rxtx_vec_avx2.c') # Build the avx2 handler if the compiler supports it, even though 'machine' # does not. This is to support users who build for the min supported machine
drivers/net/i40e/Makefile +1 -1
@@ -85,7 +85,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_I40E_PMD) += i40e_tm.c SRCS-$(CONFIG_RTE_LIBRTE_I40E_PMD) += i40e_vf_representor.c ifeq ($(CONFIG_RTE_LIBRTE_I40E_INC_VECTOR),y) -ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2) +ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),) CC_AVX2_SUPPORT=1 else CC_AVX2_SUPPORT=\
drivers/net/i40e/meson.build +1 -1
@@ -31,7 +31,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' cflags += ['-DCC_AVX2_SUPPORT'] sources += files('i40e_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2')
drivers/net/iavf/Makefile +1 -1
@@ -31,7 +31,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_IAVF_PMD) += iavf_rxtx_vec_sse.c endif ifeq ($(CONFIG_RTE_LIBRTE_IAVF_PMD), y) - ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2) + ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),) CC_AVX2_SUPPORT=1 else CC_AVX2_SUPPORT=\
drivers/net/iavf/meson.build +1 -1
@@ -21,7 +21,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' cflags += ['-DCC_AVX2_SUPPORT'] sources += files('iavf_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2')
drivers/net/ice/Makefile +1 -1
@@ -66,7 +66,7 @@ endif SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_switch_filter.c SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_fdir_filter.c SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_hash.c -ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2) +ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),) CC_AVX2_SUPPORT=1 else CC_AVX2_SUPPORT=\
drivers/net/ice/meson.build +1 -1
@@ -22,7 +22,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' sources += files('ice_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2') ice_avx2_lib = static_library('ice_avx2_lib',
examples/l3fwd/l3fwd_em.c +2 -2
@@ -215,7 +215,7 @@ static rte_xmm_t mask0;static rte_xmm_t mask1;static rte_xmm_t mask2; -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
@@ -231,7 +231,7 @@ em_mask_key(void *key, xmm_t mask) return vandq_s32(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC) +#elif defined(__ALTIVEC__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
lib/librte_acl/Makefile +1 -1
@@ -38,7 +38,7 @@ endif # #check if flag for AVX2 is already on, if not set it up manually -ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2) +ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),) CC_AVX2_SUPPORT=1 else CC_AVX2_SUPPORT=\
lib/librte_acl/meson.build +1 -1
@@ -15,7 +15,7 @@ if dpdk_conf.has('RTE_ARCH_X86') # in former case, just add avx2 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. - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' sources += files('acl_run_avx2.c') cflags += '-DCC_AVX2_SUPPORT' elif cc.has_argument('-mavx2')
lib/librte_eal/common/rte_random.c +2 -2
@@ -2,7 +2,7 @@* Copyright(c) 2019 Ericsson AB */ -#ifdef RTE_MACHINE_CPUFLAG_RDSEED +#ifdef __RDSEED__#include <x86intrin.h> #endif #include <stdlib.h>
@@ -188,7 +188,7 @@ __rte_random_initial_seed(void)if (ge_rc ==0 )return ge_seed;#endif -#ifdef RTE_MACHINE_CPUFLAG_RDSEED +#ifdef __RDSEED__unsigned int rdseed_low;unsigned int rdseed_high;
lib/librte_eal/x86/include/rte_memcpy.h +4 -4
@@ -45,7 +45,7 @@ extern "C" { static __rte_always_inline void *rte_memcpy (void *dst,const void *src,size_t n); -#ifdef RTE_MACHINE_CPUFLAG_AVX512F +#ifdef __AVX512F__#define ALIGNMENT_MASK 0x3F
@@ -286,7 +286,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_128_BACK63; } -#elif defined RTE_MACHINE_CPUFLAG_AVX2 +#elif defined __AVX2__#define ALIGNMENT_MASK 0x1F
@@ -479,7 +479,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_128_BACK31; } -#else /* RTE_MACHINE_CPUFLAG */ +#else /* __AVX512F__ */ #define ALIGNMENT_MASK 0x0F
@@ -803,7 +803,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_64_BACK15; } -#endif /* RTE_MACHINE_CPUFLAG */ +#endif /* __AVX512F__ */ static __rte_always_inline void *rte_memcpy_aligned (void *dst,const void *src,size_t n)
lib/librte_efd/rte_efd_x86.h +1 -1
@@ -19,7 +19,7 @@ efd_lookup_internal_avx2(const efd_hashfunc_t *group_hash_idx,const efd_lookuptbl_t *group_lookup_table, const uint32_t hash_val_a, const uint32_t hash_val_b) { -#ifdef RTE_MACHINE_CPUFLAG_AVX2 +#ifdef __AVX2__ efd_value_t value = 0;uint32_t i =0 ;__m256i vhash_val_a =_mm256_set1_epi32 (hash_val_a);
lib/librte_hash/rte_cuckoo_hash.c +1 -1
@@ -1691,7 +1691,7 @@ compare_signatures(uint32_t *prim_hash_matches, uint32_t *sec_hash_matches,/* For match mask the first bit of every two bits indicates the match */ switch (sig_cmp_fn) { -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)case RTE_HASH_COMPARE_SSE :/* Compare all signatures in the bucket */ *prim_hash_matches = _mm_movemask_epi8(_mm_cmpeq_epi16(
lib/librte_member/rte_member_ht.c +5 -5
@@ -176,7 +176,7 @@ rte_member_lookup_ht(const struct rte_member_setsum *ss,get_buckets_index (ss, key, &prim_bucket, &sec_bucket, &tmp_sig);switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (search_bucket_single_avx(prim_bucket, tmp_sig, buckets, set_id) ||
@@ -216,7 +216,7 @@ rte_member_lookup_bulk_ht(const struct rte_member_setsum *ss,for (i =0 ; i < num_keys; i++) {switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (search_bucket_single_avx(prim_buckets[i], tmp_sig[i], buckets, &set_id[i]) ||
@@ -253,7 +253,7 @@ rte_member_lookup_multi_ht(const struct rte_member_setsum *ss,get_buckets_index (ss, key, &prim_bucket, &sec_bucket, &tmp_sig);switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 :search_bucket_multi_avx (prim_bucket, tmp_sig, buckets, &num_matches, match_per_key, set_id);
@@ -296,7 +296,7 @@ rte_member_lookup_multi_bulk_ht(const struct rte_member_setsum *ss, match_cnt_tmp =0 ;switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : search_bucket_multi_avx(prim_buckets[i], tmp_sig[i], buckets, &match_cnt_tmp, match_per_key,
@@ -357,7 +357,7 @@ try_update(struct member_ht_bucket *buckets, uint32_t prim, uint32_t sec, enum rte_member_sig_compare_function cmp_fn) {switch (cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (update_entry_search_avx(prim, sig, buckets, set_id) || update_entry_search_avx(sec, sig, buckets,
lib/librte_member/rte_member_x86.h +1 -1
@@ -11,7 +11,7 @@ extern "C" {#include <x86intrin.h> -#if defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(__AVX2__) static inline int update_entry_search_avx(uint32_t bucket_id, member_sig_t tmp_sig,
lib/librte_net/rte_net_crc.c +1 -1
@@ -10,7 +10,7 @@#include <rte_common.h> #include <rte_net_crc.h> -#if defined(RTE_ARCH_X86_64 ) && defined(RTE_MACHINE_CPUFLAG_PCLMULQDQ ) +#if defined(RTE_ARCH_X86_64 ) && defined(__PCLMUL__)#define X86_64_SSE42_PCLMULQDQ 1#elif defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_PMULL )#define ARM64_NEON_PMULL 1
mk/rte.cpuflags.mk +0 -1
@@ -113,7 +113,6 @@ CPUFLAGS += SHA1 CPUFLAGS += SHA2 endif -MACHINE_CFLAGS += $(addprefix -DRTE_MACHINE_CPUFLAG_,$(CPUFLAGS)) # To strip whitespace comma:= ,
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v1 1/4] x86: change cpuflag macros to compiler macros
Quoted 3 lines · Radu Nicolau
Replace use of
RTE_MACHINE_CPUFLAGmacros with regular compiler macros.
I think it's worth noting in the commit log that the set of macros provided by the compilers are more complete than those provided by DPDK, and by not having our own it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
Quoted 26 lines
Signed-off-by Sean Morrissey <[email protected]>Signed-off-by Radu Nicolau <[email protected]>--- app/test/test_memcpy_perf.c | 8 ++++---- config/x86/meson.build | 2 -- drivers/net/enic/Makefile | 2 +- drivers/net/enic/meson.build | 2 +- drivers/net/i40e/Makefile | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/Makefile | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/Makefile | 2 +- drivers/net/ice/meson.build | 2 +- examples/l3fwd/l3fwd_em.c | 4 ++-- lib/librte_acl/Makefile | 2 +- lib/librte_acl/meson.build | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- mk/rte.cpuflags.mk | 1 - 21 files changed, 31 insertions(+), 34 deletions(-)
<snip>
examples/l3fwd/l3fwd_em.c em_mask_key(void *key, xmm_t mask) · line 234
Quoted 6 lines
examples/l3fwd/l3fwd_em.c +1
-1@@ -231,7 +231,7 @@ em_mask_key(void *key, xmm_t mask) return vandq_s32(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC) +#elif defined(__ALTIVEC__)
Not an x86 flag. Belongs in patch 3.
Quoted 21 lines
examples/l3fwd/l3fwd_em.c +0
-0static inline xmm_t em_mask_key (void *key,xmm_t mask) {lib/librte_acl/Makefile +1
-1@@ -38,7 +38,7 @@ endif # #check if flag for AVX2 is already on, if not set it up manually -ifeq ($(findstring RTE_MACHINE_CPUFLAG_AVX2,$(CFLAGS)),RTE_MACHINE_CPUFLAG_AVX2) +ifneq ($(filter $(AUTO_CPUFLAGS),__AVX2__),) CC_AVX2_SUPPORT=1 else CC_AVX2_SUPPORT=\lib/librte_acl/meson.build +0
-0
lib/librte_acl/meson.build if dpdk_conf.has('RTE_ARCH_X86') · line 18
Quoted 5 lines
lib/librte_acl/meson.build +1
-1@@ -15,7 +15,7 @@ if dpdk_conf.has('RTE_ARCH_X86') # in former case, just add avx2 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. - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != ''
Since this is used in a number of places, we probably should just get the result in a variable in config/x86/meson.build.
Quoted 3 lines
lib/librte_acl/meson.build +0
-0sources += files('acl_run_avx2.c') cflags += '-DCC_AVX2_SUPPORT' elif cc.has_argument('-mavx2')
<snip>
v2 cover Remove RTE_MACHINE_CPUFLAG_ macros
In reply to Radu NicolauRemove RTE_MACHINE_CPUFLAG_ macros from the build. Deprecation notice sent, pasted here for reference:
* build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example, RTE_MACHINE_CPUFLAG_SSE3 duplicates the compiler-provided macro __SSE3__.
Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference to RTE_MACHINE_CPUFLAG
app/test-pmd/macswap.c | 2 +- app/test/test_memcpy_perf.c | 8 ++++---- config/arm/meson.build | 6 ------ config/ppc/meson.build | 2 -- config/x86/meson.build | 2 -- doc/guides/prog_guide/writing_efficient_code.rst | 2 -- doc/guides/rel_notes/release_20_11.rst | 1 + drivers/net/enic/meson.build | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/meson.build | 2 +- drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 16 ++++++++-------- examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_acl/meson.build | 2 +- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 4 ++-- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 4 ++-- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- 33 files changed, 52 insertions(+), 63 deletions(-)
Signature · 1 line
2.17.1
PATCH v2 1/4 x86: change cpuflag macros to compiler macros
Replace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
--- app/test/test_memcpy_perf.c | 8 ++++---- config/x86/meson.build | 2 -- drivers/net/enic/meson.build | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/meson.build | 2 +- examples/l3fwd/l3fwd_em.c | 4 ++-- lib/librte_acl/meson.build | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- 15 files changed, 26 insertions(+), 28 deletions(-)
app/test/test_memcpy_perf.c +4 -4
@@ -51,13 +51,13 @@ static size_t buf_sizes[TEST_VALUE_RANGE];#define TEST_BATCH_SIZE 100/* Data is aligned on this many bytes (power of 2) */ -#ifdef RTE_MACHINE_CPUFLAG_AVX512F +#ifdef __AVX512F__#define ALIGNMENT_UNIT 64 -#elif definedRTE_MACHINE_CPUFLAG_AVX2 +#elif defined __AVX2__#define ALIGNMENT_UNIT 32 -#else /* RTE_MACHINE_CPUFLAG */ +#else #define ALIGNMENT_UNIT 16 -#endif /* RTE_MACHINE_CPUFLAG */ +#endif /* * Pointers used in performance tests. The two large buffers are for uncached
config/x86/meson.build +0 -2
@@ -18,7 +18,6 @@ endif base_flags = ['SSE', 'SSE2', 'SSE3','SSSE3', 'SSE4_1', 'SSE4_2'] foreach f:base_flags - dpdk_conf.set('RTE_MACHINE_CPUFLAG_' + f, 1) compile_time_cpuflags += ['RTE_CPUFLAG_' + f] endforeach
@@ -32,7 +31,6 @@ foreach f:optional_flags elif f == 'RDRND' f = 'RDRAND' endif - dpdk_conf.set('RTE_MACHINE_CPUFLAG_' + f, 1) compile_time_cpuflags += ['RTE_CPUFLAG_' + f] endif endforeach
drivers/net/enic/meson.build +1 -1
@@ -20,7 +20,7 @@ deps += ['hash'] includes += include_directories('base') # The current implementation assumes 64-bit pointers -if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') and dpdk_conf.get('RTE_ARCH_64') +if cc.get_define('__AVX2__', args: machine_args) != '' and dpdk_conf.get('RTE_ARCH_64') sources += files('enic_rxtx_vec_avx2.c') # Build the avx2 handler if the compiler supports it, even though 'machine' # does not. This is to support users who build for the min supported machine
drivers/net/i40e/meson.build +1 -1
@@ -31,7 +31,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' cflags += ['-DCC_AVX2_SUPPORT'] sources += files('i40e_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2')
drivers/net/iavf/meson.build +1 -1
@@ -21,7 +21,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' cflags += ['-DCC_AVX2_SUPPORT'] sources += files('iavf_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2')
drivers/net/ice/meson.build +1 -1
@@ -22,7 +22,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' sources += files('ice_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2') ice_avx2_lib = static_library('ice_avx2_lib',
examples/l3fwd/l3fwd_em.c +2 -2
@@ -215,7 +215,7 @@ static rte_xmm_t mask0;static rte_xmm_t mask1;static rte_xmm_t mask2; -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
@@ -231,7 +231,7 @@ em_mask_key(void *key, xmm_t mask) return vandq_s32(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC) +#elif defined(__ALTIVEC__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
lib/librte_acl/meson.build +1 -1
@@ -15,7 +15,7 @@ if dpdk_conf.has('RTE_ARCH_X86') # in former case, just add avx2 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. - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' sources += files('acl_run_avx2.c') cflags += '-DCC_AVX2_SUPPORT' elif cc.has_argument('-mavx2')
lib/librte_eal/common/rte_random.c +2 -2
@@ -2,7 +2,7 @@* Copyright(c) 2019 Ericsson AB */ -#ifdef RTE_MACHINE_CPUFLAG_RDSEED +#ifdef __RDSEED__#include <x86intrin.h> #endif #include <stdlib.h>
@@ -188,7 +188,7 @@ __rte_random_initial_seed(void)if (ge_rc ==0 )return ge_seed;#endif -#ifdef RTE_MACHINE_CPUFLAG_RDSEED +#ifdef __RDSEED__unsigned int rdseed_low;unsigned int rdseed_high;
lib/librte_eal/x86/include/rte_memcpy.h +4 -4
@@ -45,7 +45,7 @@ extern "C" { static __rte_always_inline void *rte_memcpy (void *dst,const void *src,size_t n); -#ifdef RTE_MACHINE_CPUFLAG_AVX512F +#ifdef __AVX512F__#define ALIGNMENT_MASK 0x3F
@@ -286,7 +286,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_128_BACK63; } -#elif defined RTE_MACHINE_CPUFLAG_AVX2 +#elif defined __AVX2__#define ALIGNMENT_MASK 0x1F
@@ -479,7 +479,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_128_BACK31; } -#else /* RTE_MACHINE_CPUFLAG */ +#else /* __AVX512F__ */ #define ALIGNMENT_MASK 0x0F
@@ -803,7 +803,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_64_BACK15; } -#endif /* RTE_MACHINE_CPUFLAG */ +#endif /* __AVX512F__ */ static __rte_always_inline void *rte_memcpy_aligned (void *dst,const void *src,size_t n)
lib/librte_efd/rte_efd_x86.h +1 -1
@@ -19,7 +19,7 @@ efd_lookup_internal_avx2(const efd_hashfunc_t *group_hash_idx,const efd_lookuptbl_t *group_lookup_table, const uint32_t hash_val_a, const uint32_t hash_val_b) { -#ifdef RTE_MACHINE_CPUFLAG_AVX2 +#ifdef __AVX2__ efd_value_t value = 0;uint32_t i =0 ;__m256i vhash_val_a =_mm256_set1_epi32 (hash_val_a);
lib/librte_hash/rte_cuckoo_hash.c +1 -1
@@ -1691,7 +1691,7 @@ compare_signatures(uint32_t *prim_hash_matches, uint32_t *sec_hash_matches,/* For match mask the first bit of every two bits indicates the match */ switch (sig_cmp_fn) { -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)case RTE_HASH_COMPARE_SSE :/* Compare all signatures in the bucket */ *prim_hash_matches = _mm_movemask_epi8(_mm_cmpeq_epi16(
lib/librte_member/rte_member_ht.c +5 -5
@@ -176,7 +176,7 @@ rte_member_lookup_ht(const struct rte_member_setsum *ss,get_buckets_index (ss, key, &prim_bucket, &sec_bucket, &tmp_sig);switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (search_bucket_single_avx(prim_bucket, tmp_sig, buckets, set_id) ||
@@ -216,7 +216,7 @@ rte_member_lookup_bulk_ht(const struct rte_member_setsum *ss,for (i =0 ; i < num_keys; i++) {switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (search_bucket_single_avx(prim_buckets[i], tmp_sig[i], buckets, &set_id[i]) ||
@@ -253,7 +253,7 @@ rte_member_lookup_multi_ht(const struct rte_member_setsum *ss,get_buckets_index (ss, key, &prim_bucket, &sec_bucket, &tmp_sig);switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 :search_bucket_multi_avx (prim_bucket, tmp_sig, buckets, &num_matches, match_per_key, set_id);
@@ -296,7 +296,7 @@ rte_member_lookup_multi_bulk_ht(const struct rte_member_setsum *ss, match_cnt_tmp =0 ;switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : search_bucket_multi_avx(prim_buckets[i], tmp_sig[i], buckets, &match_cnt_tmp, match_per_key,
@@ -357,7 +357,7 @@ try_update(struct member_ht_bucket *buckets, uint32_t prim, uint32_t sec, enum rte_member_sig_compare_function cmp_fn) {switch (cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (update_entry_search_avx(prim, sig, buckets, set_id) || update_entry_search_avx(sec, sig, buckets,
lib/librte_member/rte_member_x86.h +1 -1
@@ -11,7 +11,7 @@ extern "C" {#include <x86intrin.h> -#if defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(__AVX2__) static inline int update_entry_search_avx(uint32_t bucket_id, member_sig_t tmp_sig,
lib/librte_net/rte_net_crc.c +1 -1
@@ -10,7 +10,7 @@#include <rte_common.h> #include <rte_net_crc.h> -#if defined(RTE_ARCH_X86_64 ) && defined(RTE_MACHINE_CPUFLAG_PCLMULQDQ ) +#if defined(RTE_ARCH_X86_64 ) && defined(__PCLMUL__)#define X86_64_SSE42_PCLMULQDQ 1#elif defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_PMULL )#define ARM64_NEON_PMULL 1
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v2 1/4] x86: change cpuflag macros to compiler macros
Quoted 14 lines · Radu Nicolau
examples/l3fwd/l3fwd_em.c +1
-1@@ -215,7 +215,7 @@ static rte_xmm_t mask0;static rte_xmm_t mask1;static rte_xmm_t mask2; -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
examples/l3fwd/l3fwd_em.c em_mask_key(void *key, xmm_t mask) · line 234
Quoted 5 lines
examples/l3fwd/l3fwd_em.c +1
-1@@ -231,7 +231,7 @@ em_mask_key(void *key, xmm_t mask) return vandq_s32(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC) +#elif defined(__ALTIVEC__)
It should be in patch 3.
Quoted 3 lines
examples/l3fwd/l3fwd_em.c +0
-0static inline xmm_t em_mask_key (void *key,xmm_t mask) {
Signature · 1 line
David Marchand
PATCH v2 2/4 arm: change cpuflag macros to compiler macros
In reply to Radu NicolauReplace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
--- app/test-pmd/macswap.c | 2 +- config/arm/meson.build | 6 ------ drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 12 ++++++------ examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- 18 files changed, 25 insertions(+), 31 deletions(-)
app/test-pmd/macswap.c +1 -1
@@ -39,7 +39,7 @@#include "testpmd.h" #if defined(RTE_ARCH_X86 )#include "macswap_sse.h" -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM__NEON)#include "macswap_neon.h" #else #include "macswap.h"
config/arm/meson.build +0 -6
@@ -208,20 +208,14 @@ message(machine_args) if (cc.get_define('__ARM_NEON', args: machine_args) != '' or cc.get_define('__aarch64__', args: machine_args) != '') - dpdk_conf.set('RTE_MACHINE_CPUFLAG_NEON', 1) compile_time_cpuflags += ['RTE_CPUFLAG_NEON'] endif if cc.get_define('__ARM_FEATURE_CRC32', args: machine_args) != '' - dpdk_conf.set('RTE_MACHINE_CPUFLAG_CRC32', 1) compile_time_cpuflags += ['RTE_CPUFLAG_CRC32'] endif if cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '' - dpdk_conf.set('RTE_MACHINE_CPUFLAG_AES', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_PMULL', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_SHA1', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_SHA2', 1) compile_time_cpuflags += ['RTE_CPUFLAG_AES', 'RTE_CPUFLAG_PMULL', 'RTE_CPUFLAG_SHA1', 'RTE_CPUFLAG_SHA2'] endif
drivers/net/ixgbe/ixgbe_ethdev.c +1 -1
@@ -3960,7 +3960,7 @@ ixgbe_dev_supported_ptypes_get(struct rte_eth_dev *dev) dev->rx_pkt_burst == ixgbe_recv_pkts_bulk_alloc)return ptypes; -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_NEON)if (dev->rx_pkt_burst == ixgbe_recv_pkts_vec || dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec)return ptypes;
examples/l3fwd/l3fwd.h +1 -1
@@ -12,7 +12,7 @@#define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1 -#if !defined(NO_HASH_MULTI_LOOKUP ) && defined(RTE_MACHINE_CPUFLAG_NEON ) +#if !defined(NO_HASH_MULTI_LOOKUP ) && defined(__ARM_NEON)#define NO_HASH_MULTI_LOOKUP 1#endif
examples/l3fwd/l3fwd_em.c +6 -6
@@ -28,7 +28,7 @@#include "l3fwd.h" #include "l3fwd_event.h" -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32)#define EM_HASH_CRC 1#endif
@@ -223,7 +223,7 @@ em_mask_key(void *key, xmm_t mask) return _mm_and_si128(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_NEON) +#elif defined(__ARM_NEON)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
@@ -303,7 +303,7 @@ em_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)return (ret <0 ) ? portid : ipv6_l3fwd_out_if[ret]; } -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEON#if defined(NO_HASH_MULTI_LOOKUP )#include "l3fwd_em_sequential.h" #else
@@ -685,7 +685,7 @@ em_main_loop(__rte_unused void *dummy)if (nb_rx ==0 )continue ; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEONl3fwd_em_send_packets (nb_rx, pkts_burst, portid, qconf);#else
@@ -723,7 +723,7 @@ em_event_loop_single(struct l3fwd_event_resources *evt_rsrc,struct rte_mbuf *mbuf = ev.mbuf; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEON mbuf->port =em_get_dst_port (lconf, mbuf, mbuf->port);process_packet (mbuf, &mbuf->port);#else
@@ -784,7 +784,7 @@ em_event_loop_burst(struct l3fwd_event_resources *evt_rsrc,continue ; } -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEONl3fwd_em_process_events (nb_deq, (struct rte_event **)&events, lconf);#else
examples/l3fwd/l3fwd_em_hlm.h +1 -1
@@ -9,7 +9,7 @@#if definedRTE_ARCH_X86 #include "l3fwd_sse.h" #include "l3fwd_em_hlm_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_neon.h" #include "l3fwd_em_hlm_neon.h" #endif
examples/l3fwd/l3fwd_em_sequential.h +1 -1
@@ -16,7 +16,7 @@#if definedRTE_ARCH_X86 #include "l3fwd_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_neon.h" #endif
examples/l3fwd/l3fwd_lpm.c +3 -3
@@ -163,7 +163,7 @@ lpm_get_dst_port_with_ipv4(const struct lcore_conf *qconf, struct rte_mbuf *pkt,#if defined(RTE_ARCH_X86 )#include "l3fwd_lpm_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_lpm_neon.h" #elif defined(RTE_ARCH_PPC_64 )#include "l3fwd_lpm_altivec.h"
@@ -240,7 +240,7 @@ lpm_main_loop(__rte_unused void *dummy)if (nb_rx ==0 )continue ; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON \ +#if definedRTE_ARCH_X86 || defined __ARM_NEON \ || definedRTE_ARCH_PPC_64 l3fwd_lpm_send_packets (nb_rx, pkts_burst, portid, qconf);
@@ -259,7 +259,7 @@ lpm_process_event_pkt(const struct lcore_conf *lconf, struct rte_mbuf *mbuf) { mbuf->port =lpm_get_dst_port (lconf, mbuf, mbuf->port); -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON \ +#if definedRTE_ARCH_X86 || defined __ARM_NEON \ || definedRTE_ARCH_PPC_64 process_packet (mbuf, &mbuf->port);#else
lib/librte_eal/arm/include/rte_memcpy_32.h +1 -1
@@ -16,7 +16,7 @@ extern "C" {#ifdef RTE_ARCH_ARM_NEON_MEMCPY -#ifndef RTE_MACHINE_CPUFLAG_NEON +#ifndef __ARM_NEON#error "Cannot optimize memcpy by NEON as the CPU seems to not support this"#endif
lib/librte_hash/rte_cuckoo_hash.c +1 -1
@@ -1704,7 +1704,7 @@ compare_signatures(uint32_t *prim_hash_matches, uint32_t *sec_hash_matches, (__m128i const *)sec_bkt->sig_current), _mm_set1_epi16(sig)));break ; -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)case RTE_HASH_COMPARE_NEON : {uint16x8_t vmat, vsig, x;int16x8_t shift = {-15 ,-13 ,-11 ,-9 ,-7 ,-5 ,-3 ,-1 };
lib/librte_hash/rte_hash_crc.h +1 -1
@@ -424,7 +424,7 @@ crc32c_sse42_u64(uint64_t data, uint64_t init_val)static uint8_t crc32_alg =CRC32_SW ; -#if defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)#include "rte_crc_arm64.h" #else
lib/librte_hash/rte_thash.h +2 -2
@@ -28,7 +28,7 @@ extern "C" {#include <rte_ip.h> #include <rte_common.h> -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_NEON)#include <rte_vect.h> #endif
@@ -149,7 +149,7 @@ rte_thash_load_v6_addrs(const struct rte_ipv6_hdr *orig, ipv6 =_mm_loadu_si128 ((const __m128i *)orig->dst_addr); *(__m128i *)targ->v6.dst_addr =_mm_shuffle_epi8 (ipv6, rte_thash_ipv6_bswap_mask); -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)uint8x16_t ipv6 =vld1q_u8 ((uint8_t const *)orig->src_addr);vst1q_u8 ((uint8_t *)targ->v6.src_addr,vrev32q_u8 (ipv6)); ipv6 =vld1q_u8 ((uint8_t const *)orig->dst_addr);
lib/librte_member/rte_member.h +1 -1
@@ -68,7 +68,7 @@ typedef uint16_t member_set_t;#define RTE_MEMBER_NAMESIZE 32/** @internal Hash function used by membership library. */ -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32)#include <rte_hash_crc.h> #define MEMBER_HASH_FUNC rte_hash_crc#else
lib/librte_net/rte_net_crc.c +1 -1
@@ -12,7 +12,7 @@#if defined(RTE_ARCH_X86_64 ) && defined(__PCLMUL__)#define X86_64_SSE42_PCLMULQDQ 1 -#elif defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_PMULL ) +#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRYPTO)#define ARM64_NEON_PMULL 1#endif
lib/librte_node/ip4_lookup.c +1 -1
@@ -30,7 +30,7 @@ struct ip4_lookup_node_main {static struct ip4_lookup_node_main ip4_lookup_nm; -#if defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(__ARM_NEON)#include "ip4_lookup_neon.h" #elif defined(RTE_ARCH_X86 )#include "ip4_lookup_sse.h"
lib/librte_sched/rte_sched.c +1 -1
@@ -29,7 +29,7 @@#ifdef RTE_ARCH_X86 #define SCHED_VECTOR_SSE4 -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)#define SCHED_VECTOR_NEON #endif
lib/librte_table/rte_lru_arm64.h +1 -1
@@ -13,7 +13,7 @@ extern "C" {#include <rte_vect.h> #ifndef RTE_TABLE_HASH_LRU_STRATEGY -#ifdef RTE_MACHINE_CPUFLAG_NEON +#ifdef __ARM_NEON#define RTE_TABLE_HASH_LRU_STRATEGY 3#else /* if no NEON, use simple scalar version */ #define RTE_TABLE_HASH_LRU_STRATEGY 1
lib/librte_table/rte_table_hash_func.h +1 -1
@@ -41,7 +41,7 @@ rte_crc32_u64(uint64_t crc, uint64_t v) return _mm_crc32_u64(crc, v); } -#elif defined(RTE_ARCH_ARM64) && defined(RTE_MACHINE_CPUFLAG_CRC32) +#elif defined(RTE_ARCH_ARM64) && defined(__ARM_FEATURE_CRC32)#include "rte_table_hash_func_arm64.h" #else
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v2 2/4] arm: change cpuflag macros to compiler macros
app/test-pmd/macswap.c line 42
Quoted 10 lines · Radu Nicolau
app/test-pmd/macswap.c +1
-1@@ -39,7 +39,7 @@#include "testpmd.h" #if defined(RTE_ARCH_X86 )#include "macswap_sse.h" -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM__NEON)
Typo for __ARM_NEON?
Quoted 3 lines
app/test-pmd/macswap.c +0
-0#include "macswap_neon.h" #else #include "macswap.h"
Signature · 1 line
David Marchand
PATCH v2 3/4 ppc: change cpuflag macros to compiler macros
In reply to Radu NicolauReplace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
--- config/ppc/meson.build | 2 -- 1 file changed, 2 deletions(-)
config/ppc/meson.build +0 -2
@@ -21,5 +21,3 @@ endif dpdk_conf.set('RTE_MAX_LCORE', 1536) dpdk_conf.set('RTE_MAX_NUMA_NODES', 32) dpdk_conf.set('RTE_CACHE_LINE_SIZE', 128) -dpdk_conf.set('RTE_MACHINE_CPUFLAG_ALTIVEC', 1) -dpdk_conf.set('RTE_MACHINE_CPUFLAG_VSX', 1)
Signature · 1 line
2.17.1
PATCH v2 4/4 doc: remove reference to RTE_MACHINE_CPUFLAG
In reply to Radu NicolauRTE_MACHINE_CPUFLAG macros are replaced with predefined compiler defines.
--- doc/guides/prog_guide/writing_efficient_code.rst | 2 -- doc/guides/rel_notes/release_20_11.rst | 1 + 2 files changed, 1 insertion(+), 2 deletions(-)
doc/guides/prog_guide/writing_efficient_code.rst +0 -2
@@ -273,5 +273,3 @@ main() function and checks if the current machine is suitable for running the bi Along with compiler optimizations, a set of preprocessor defines are automatically added to the build process (regardless of the compiler version). These defines correspond to the instruction sets that the target CPU should be able to support. -For example, a binary compiled for any SSE4.2-capable processor will have RTE_MACHINE_CPUFLAG_SSE4_2 defined, -thus enabling compile-time code path selection for different platforms.
doc/guides/rel_notes/release_20_11.rst +1 -0
@@ -91,6 +91,7 @@ Removed Items Also, make sure to start the actual text at the margin. ======================================================= +* Removed ``RTE_MACHINE_CPUFLAG`` macros. API Changes -----------
Signature · 1 line
2.17.1
v3 cover Remove RTE_MACHINE_CPUFLAG_ macros
In reply to Radu NicolauRemove RTE_MACHINE_CPUFLAG_ macros from the build. Deprecation notice sent, pasted here for reference:
* build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example, RTE_MACHINE_CPUFLAG_SSE3 duplicates the compiler-provided macro __SSE3__.
Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference to RTE_MACHINE_CPUFLAG
app/test-pmd/macswap.c | 2 +- app/test/test_memcpy_perf.c | 8 ++++---- config/arm/meson.build | 6 ------ config/ppc/meson.build | 2 -- config/x86/meson.build | 2 -- doc/guides/prog_guide/writing_efficient_code.rst | 2 -- doc/guides/rel_notes/release_20_11.rst | 1 + drivers/net/enic/meson.build | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/meson.build | 2 +- drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 16 ++++++++-------- examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_acl/meson.build | 2 +- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 4 ++-- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 4 ++-- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- 33 files changed, 52 insertions(+), 63 deletions(-)
Signature · 1 line
2.17.1
PATCH v3 1/4 x86: change cpuflag macros to compiler macros
Replace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
--- app/test/test_memcpy_perf.c | 8 ++++---- config/x86/meson.build | 2 -- drivers/net/enic/meson.build | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/meson.build | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/librte_acl/meson.build | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- 15 files changed, 25 insertions(+), 27 deletions(-)
app/test/test_memcpy_perf.c +4 -4
@@ -51,13 +51,13 @@ static size_t buf_sizes[TEST_VALUE_RANGE];#define TEST_BATCH_SIZE 100/* Data is aligned on this many bytes (power of 2) */ -#ifdef RTE_MACHINE_CPUFLAG_AVX512F +#ifdef __AVX512F__#define ALIGNMENT_UNIT 64 -#elif definedRTE_MACHINE_CPUFLAG_AVX2 +#elif defined __AVX2__#define ALIGNMENT_UNIT 32 -#else /* RTE_MACHINE_CPUFLAG */ +#else #define ALIGNMENT_UNIT 16 -#endif /* RTE_MACHINE_CPUFLAG */ +#endif /* * Pointers used in performance tests. The two large buffers are for uncached
config/x86/meson.build +0 -2
@@ -18,7 +18,6 @@ endif base_flags = ['SSE', 'SSE2', 'SSE3','SSSE3', 'SSE4_1', 'SSE4_2'] foreach f:base_flags - dpdk_conf.set('RTE_MACHINE_CPUFLAG_' + f, 1) compile_time_cpuflags += ['RTE_CPUFLAG_' + f] endforeach
@@ -32,7 +31,6 @@ foreach f:optional_flags elif f == 'RDRND' f = 'RDRAND' endif - dpdk_conf.set('RTE_MACHINE_CPUFLAG_' + f, 1) compile_time_cpuflags += ['RTE_CPUFLAG_' + f] endif endforeach
drivers/net/enic/meson.build +1 -1
@@ -20,7 +20,7 @@ deps += ['hash'] includes += include_directories('base') # The current implementation assumes 64-bit pointers -if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') and dpdk_conf.get('RTE_ARCH_64') +if cc.get_define('__AVX2__', args: machine_args) != '' and dpdk_conf.get('RTE_ARCH_64') sources += files('enic_rxtx_vec_avx2.c') # Build the avx2 handler if the compiler supports it, even though 'machine' # does not. This is to support users who build for the min supported machine
drivers/net/i40e/meson.build +1 -1
@@ -31,7 +31,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' cflags += ['-DCC_AVX2_SUPPORT'] sources += files('i40e_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2')
drivers/net/iavf/meson.build +1 -1
@@ -21,7 +21,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' cflags += ['-DCC_AVX2_SUPPORT'] sources += files('iavf_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2')
drivers/net/ice/meson.build +1 -1
@@ -22,7 +22,7 @@ if arch_subdir == 'x86' # compile AVX2 version if either: # a. we have AVX supported in minimum instruction set baseline # b. it's not minimum instruction set, but supported by compiler - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' sources += files('ice_rxtx_vec_avx2.c') elif cc.has_argument('-mavx2') ice_avx2_lib = static_library('ice_avx2_lib',
examples/l3fwd/l3fwd_em.c +1 -1
@@ -215,7 +215,7 @@ static rte_xmm_t mask0;static rte_xmm_t mask1;static rte_xmm_t mask2; -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
lib/librte_acl/meson.build +1 -1
@@ -15,7 +15,7 @@ if dpdk_conf.has('RTE_ARCH_X86') # in former case, just add avx2 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. - if dpdk_conf.has('RTE_MACHINE_CPUFLAG_AVX2') + if cc.get_define('__AVX2__', args: machine_args) != '' sources += files('acl_run_avx2.c') cflags += '-DCC_AVX2_SUPPORT' elif cc.has_argument('-mavx2')
lib/librte_eal/common/rte_random.c +2 -2
@@ -2,7 +2,7 @@* Copyright(c) 2019 Ericsson AB */ -#ifdef RTE_MACHINE_CPUFLAG_RDSEED +#ifdef __RDSEED__#include <x86intrin.h> #endif #include <stdlib.h>
@@ -188,7 +188,7 @@ __rte_random_initial_seed(void)if (ge_rc ==0 )return ge_seed;#endif -#ifdef RTE_MACHINE_CPUFLAG_RDSEED +#ifdef __RDSEED__unsigned int rdseed_low;unsigned int rdseed_high;
lib/librte_eal/x86/include/rte_memcpy.h +4 -4
@@ -45,7 +45,7 @@ extern "C" { static __rte_always_inline void *rte_memcpy (void *dst,const void *src,size_t n); -#ifdef RTE_MACHINE_CPUFLAG_AVX512F +#ifdef __AVX512F__#define ALIGNMENT_MASK 0x3F
@@ -286,7 +286,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_128_BACK63; } -#elif defined RTE_MACHINE_CPUFLAG_AVX2 +#elif defined __AVX2__#define ALIGNMENT_MASK 0x1F
@@ -479,7 +479,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_128_BACK31; } -#else /* RTE_MACHINE_CPUFLAG */ +#else /* __AVX512F__ */ #define ALIGNMENT_MASK 0x0F
@@ -803,7 +803,7 @@ rte_memcpy_generic(void *dst, const void *src, size_t n) goto COPY_BLOCK_64_BACK15; } -#endif /* RTE_MACHINE_CPUFLAG */ +#endif /* __AVX512F__ */ static __rte_always_inline void *rte_memcpy_aligned (void *dst,const void *src,size_t n)
lib/librte_efd/rte_efd_x86.h +1 -1
@@ -19,7 +19,7 @@ efd_lookup_internal_avx2(const efd_hashfunc_t *group_hash_idx,const efd_lookuptbl_t *group_lookup_table, const uint32_t hash_val_a, const uint32_t hash_val_b) { -#ifdef RTE_MACHINE_CPUFLAG_AVX2 +#ifdef __AVX2__ efd_value_t value = 0;uint32_t i =0 ;__m256i vhash_val_a =_mm256_set1_epi32 (hash_val_a);
lib/librte_hash/rte_cuckoo_hash.c +1 -1
@@ -1691,7 +1691,7 @@ compare_signatures(uint32_t *prim_hash_matches, uint32_t *sec_hash_matches,/* For match mask the first bit of every two bits indicates the match */ switch (sig_cmp_fn) { -#if defined(RTE_MACHINE_CPUFLAG_SSE2 ) +#if defined(__SSE2__)case RTE_HASH_COMPARE_SSE :/* Compare all signatures in the bucket */ *prim_hash_matches = _mm_movemask_epi8(_mm_cmpeq_epi16(
lib/librte_member/rte_member_ht.c +5 -5
@@ -176,7 +176,7 @@ rte_member_lookup_ht(const struct rte_member_setsum *ss,get_buckets_index (ss, key, &prim_bucket, &sec_bucket, &tmp_sig);switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (search_bucket_single_avx(prim_bucket, tmp_sig, buckets, set_id) ||
@@ -216,7 +216,7 @@ rte_member_lookup_bulk_ht(const struct rte_member_setsum *ss,for (i =0 ; i < num_keys; i++) {switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (search_bucket_single_avx(prim_buckets[i], tmp_sig[i], buckets, &set_id[i]) ||
@@ -253,7 +253,7 @@ rte_member_lookup_multi_ht(const struct rte_member_setsum *ss,get_buckets_index (ss, key, &prim_bucket, &sec_bucket, &tmp_sig);switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 :search_bucket_multi_avx (prim_bucket, tmp_sig, buckets, &num_matches, match_per_key, set_id);
@@ -296,7 +296,7 @@ rte_member_lookup_multi_bulk_ht(const struct rte_member_setsum *ss, match_cnt_tmp =0 ;switch (ss->sig_cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : search_bucket_multi_avx(prim_buckets[i], tmp_sig[i], buckets, &match_cnt_tmp, match_per_key,
@@ -357,7 +357,7 @@ try_update(struct member_ht_bucket *buckets, uint32_t prim, uint32_t sec, enum rte_member_sig_compare_function cmp_fn) {switch (cmp_fn) { -#if defined(RTE_ARCH_X86 ) && defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(RTE_ARCH_X86 ) && defined(__AVX2__)case RTE_MEMBER_COMPARE_AVX2 : if (update_entry_search_avx(prim, sig, buckets, set_id) || update_entry_search_avx(sec, sig, buckets,
lib/librte_member/rte_member_x86.h +1 -1
@@ -11,7 +11,7 @@ extern "C" {#include <x86intrin.h> -#if defined(RTE_MACHINE_CPUFLAG_AVX2 ) +#if defined(__AVX2__) static inline int update_entry_search_avx(uint32_t bucket_id, member_sig_t tmp_sig,
lib/librte_net/rte_net_crc.c +1 -1
@@ -10,7 +10,7 @@#include <rte_common.h> #include <rte_net_crc.h> -#if defined(RTE_ARCH_X86_64 ) && defined(RTE_MACHINE_CPUFLAG_PCLMULQDQ ) +#if defined(RTE_ARCH_X86_64 ) && defined(__PCLMUL__)#define X86_64_SSE42_PCLMULQDQ 1#elif defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_PMULL )#define ARM64_NEON_PMULL 1
Signature · 1 line
2.17.1
PATCH v3 2/4 arm: change cpuflag macros to compiler macros
In reply to Radu NicolauReplace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
--- app/test-pmd/macswap.c | 2 +- config/arm/meson.build | 6 ------ drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 12 ++++++------ examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- 18 files changed, 25 insertions(+), 31 deletions(-)
app/test-pmd/macswap.c +1 -1
@@ -39,7 +39,7 @@#include "testpmd.h" #if defined(RTE_ARCH_X86 )#include "macswap_sse.h" -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)#include "macswap_neon.h" #else #include "macswap.h"
config/arm/meson.build +0 -6
@@ -208,20 +208,14 @@ message(machine_args) if (cc.get_define('__ARM_NEON', args: machine_args) != '' or cc.get_define('__aarch64__', args: machine_args) != '') - dpdk_conf.set('RTE_MACHINE_CPUFLAG_NEON', 1) compile_time_cpuflags += ['RTE_CPUFLAG_NEON'] endif if cc.get_define('__ARM_FEATURE_CRC32', args: machine_args) != '' - dpdk_conf.set('RTE_MACHINE_CPUFLAG_CRC32', 1) compile_time_cpuflags += ['RTE_CPUFLAG_CRC32'] endif if cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '' - dpdk_conf.set('RTE_MACHINE_CPUFLAG_AES', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_PMULL', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_SHA1', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_SHA2', 1) compile_time_cpuflags += ['RTE_CPUFLAG_AES', 'RTE_CPUFLAG_PMULL', 'RTE_CPUFLAG_SHA1', 'RTE_CPUFLAG_SHA2'] endif
drivers/net/ixgbe/ixgbe_ethdev.c +1 -1
@@ -3960,7 +3960,7 @@ ixgbe_dev_supported_ptypes_get(struct rte_eth_dev *dev) dev->rx_pkt_burst == ixgbe_recv_pkts_bulk_alloc)return ptypes; -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_NEON)if (dev->rx_pkt_burst == ixgbe_recv_pkts_vec || dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec)return ptypes;
examples/l3fwd/l3fwd.h +1 -1
@@ -12,7 +12,7 @@#define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1 -#if !defined(NO_HASH_MULTI_LOOKUP ) && defined(RTE_MACHINE_CPUFLAG_NEON ) +#if !defined(NO_HASH_MULTI_LOOKUP ) && defined(__ARM_NEON)#define NO_HASH_MULTI_LOOKUP 1#endif
examples/l3fwd/l3fwd_em.c +6 -6
@@ -28,7 +28,7 @@#include "l3fwd.h" #include "l3fwd_event.h" -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32)#define EM_HASH_CRC 1#endif
@@ -223,7 +223,7 @@ em_mask_key(void *key, xmm_t mask) return _mm_and_si128(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_NEON) +#elif defined(__ARM_NEON)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
@@ -303,7 +303,7 @@ em_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)return (ret <0 ) ? portid : ipv6_l3fwd_out_if[ret]; } -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEON#if defined(NO_HASH_MULTI_LOOKUP )#include "l3fwd_em_sequential.h" #else
@@ -685,7 +685,7 @@ em_main_loop(__rte_unused void *dummy)if (nb_rx ==0 )continue ; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEONl3fwd_em_send_packets (nb_rx, pkts_burst, portid, qconf);#else
@@ -723,7 +723,7 @@ em_event_loop_single(struct l3fwd_event_resources *evt_rsrc,struct rte_mbuf *mbuf = ev.mbuf; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEON mbuf->port =em_get_dst_port (lconf, mbuf, mbuf->port);process_packet (mbuf, &mbuf->port);#else
@@ -784,7 +784,7 @@ em_event_loop_burst(struct l3fwd_event_resources *evt_rsrc,continue ; } -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEONl3fwd_em_process_events (nb_deq, (struct rte_event **)&events, lconf);#else
examples/l3fwd/l3fwd_em_hlm.h +1 -1
@@ -9,7 +9,7 @@#if definedRTE_ARCH_X86 #include "l3fwd_sse.h" #include "l3fwd_em_hlm_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_neon.h" #include "l3fwd_em_hlm_neon.h" #endif
examples/l3fwd/l3fwd_em_sequential.h +1 -1
@@ -16,7 +16,7 @@#if definedRTE_ARCH_X86 #include "l3fwd_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_neon.h" #endif
examples/l3fwd/l3fwd_lpm.c +3 -3
@@ -163,7 +163,7 @@ lpm_get_dst_port_with_ipv4(const struct lcore_conf *qconf, struct rte_mbuf *pkt,#if defined(RTE_ARCH_X86 )#include "l3fwd_lpm_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_lpm_neon.h" #elif defined(RTE_ARCH_PPC_64 )#include "l3fwd_lpm_altivec.h"
@@ -240,7 +240,7 @@ lpm_main_loop(__rte_unused void *dummy)if (nb_rx ==0 )continue ; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON \ +#if definedRTE_ARCH_X86 || defined __ARM_NEON \ || definedRTE_ARCH_PPC_64 l3fwd_lpm_send_packets (nb_rx, pkts_burst, portid, qconf);
@@ -259,7 +259,7 @@ lpm_process_event_pkt(const struct lcore_conf *lconf, struct rte_mbuf *mbuf) { mbuf->port =lpm_get_dst_port (lconf, mbuf, mbuf->port); -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON \ +#if definedRTE_ARCH_X86 || defined __ARM_NEON \ || definedRTE_ARCH_PPC_64 process_packet (mbuf, &mbuf->port);#else
lib/librte_eal/arm/include/rte_memcpy_32.h +1 -1
@@ -16,7 +16,7 @@ extern "C" {#ifdef RTE_ARCH_ARM_NEON_MEMCPY -#ifndef RTE_MACHINE_CPUFLAG_NEON +#ifndef __ARM_NEON#error "Cannot optimize memcpy by NEON as the CPU seems to not support this"#endif
lib/librte_hash/rte_cuckoo_hash.c +1 -1
@@ -1704,7 +1704,7 @@ compare_signatures(uint32_t *prim_hash_matches, uint32_t *sec_hash_matches, (__m128i const *)sec_bkt->sig_current), _mm_set1_epi16(sig)));break ; -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)case RTE_HASH_COMPARE_NEON : {uint16x8_t vmat, vsig, x;int16x8_t shift = {-15 ,-13 ,-11 ,-9 ,-7 ,-5 ,-3 ,-1 };
lib/librte_hash/rte_hash_crc.h +1 -1
@@ -424,7 +424,7 @@ crc32c_sse42_u64(uint64_t data, uint64_t init_val)static uint8_t crc32_alg =CRC32_SW ; -#if defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)#include "rte_crc_arm64.h" #else
lib/librte_hash/rte_thash.h +2 -2
@@ -28,7 +28,7 @@ extern "C" {#include <rte_ip.h> #include <rte_common.h> -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_NEON)#include <rte_vect.h> #endif
@@ -149,7 +149,7 @@ rte_thash_load_v6_addrs(const struct rte_ipv6_hdr *orig, ipv6 =_mm_loadu_si128 ((const __m128i *)orig->dst_addr); *(__m128i *)targ->v6.dst_addr =_mm_shuffle_epi8 (ipv6, rte_thash_ipv6_bswap_mask); -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)uint8x16_t ipv6 =vld1q_u8 ((uint8_t const *)orig->src_addr);vst1q_u8 ((uint8_t *)targ->v6.src_addr,vrev32q_u8 (ipv6)); ipv6 =vld1q_u8 ((uint8_t const *)orig->dst_addr);
lib/librte_member/rte_member.h +1 -1
@@ -68,7 +68,7 @@ typedef uint16_t member_set_t;#define RTE_MEMBER_NAMESIZE 32/** @internal Hash function used by membership library. */ -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32)#include <rte_hash_crc.h> #define MEMBER_HASH_FUNC rte_hash_crc#else
lib/librte_net/rte_net_crc.c +1 -1
@@ -12,7 +12,7 @@#if defined(RTE_ARCH_X86_64 ) && defined(__PCLMUL__)#define X86_64_SSE42_PCLMULQDQ 1 -#elif defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_PMULL ) +#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRYPTO)#define ARM64_NEON_PMULL 1#endif
lib/librte_node/ip4_lookup.c +1 -1
@@ -30,7 +30,7 @@ struct ip4_lookup_node_main {static struct ip4_lookup_node_main ip4_lookup_nm; -#if defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(__ARM_NEON)#include "ip4_lookup_neon.h" #elif defined(RTE_ARCH_X86 )#include "ip4_lookup_sse.h"
lib/librte_sched/rte_sched.c +1 -1
@@ -29,7 +29,7 @@#ifdef RTE_ARCH_X86 #define SCHED_VECTOR_SSE4 -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)#define SCHED_VECTOR_NEON #endif
lib/librte_table/rte_lru_arm64.h +1 -1
@@ -13,7 +13,7 @@ extern "C" {#include <rte_vect.h> #ifndef RTE_TABLE_HASH_LRU_STRATEGY -#ifdef RTE_MACHINE_CPUFLAG_NEON +#ifdef __ARM_NEON#define RTE_TABLE_HASH_LRU_STRATEGY 3#else /* if no NEON, use simple scalar version */ #define RTE_TABLE_HASH_LRU_STRATEGY 1
lib/librte_table/rte_table_hash_func.h +1 -1
@@ -41,7 +41,7 @@ rte_crc32_u64(uint64_t crc, uint64_t v) return _mm_crc32_u64(crc, v); } -#elif defined(RTE_ARCH_ARM64) && defined(RTE_MACHINE_CPUFLAG_CRC32) +#elif defined(RTE_ARCH_ARM64) && defined(__ARM_FEATURE_CRC32)#include "rte_table_hash_func_arm64.h" #else
Signature · 1 line
2.17.1
PATCH v3 3/4 ppc: change cpuflag macros to compiler macros
In reply to Radu NicolauReplace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.
--- config/ppc/meson.build | 2 -- examples/l3fwd/l3fwd_em.c | 2 +- 2 files changed, 1 insertion(+), 3 deletions(-)
config/ppc/meson.build +0 -2
@@ -21,5 +21,3 @@ endif dpdk_conf.set('RTE_MAX_LCORE', 1536) dpdk_conf.set('RTE_MAX_NUMA_NODES', 32) dpdk_conf.set('RTE_CACHE_LINE_SIZE', 128) -dpdk_conf.set('RTE_MACHINE_CPUFLAG_ALTIVEC', 1) -dpdk_conf.set('RTE_MACHINE_CPUFLAG_VSX', 1)
examples/l3fwd/l3fwd_em.c +1 -1
@@ -231,7 +231,7 @@ em_mask_key(void *key, xmm_t mask) return vandq_s32(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_ALTIVEC) +#elif defined(__ALTIVEC__)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v3 3/4] ppc: change cpuflag macros to compiler macros
Quoted 8 lines · Radu Nicolau
Replace use of
RTE_MACHINE_CPUFLAGmacros with regular compiler macros, which are more complete than those provided by DPDK, and as such it allows new instruction sets to be leveraged without having to do extra work to set them up in DPDK.Signed-off-by Sean Morrissey <[email protected]>Signed-off-by Radu Nicolau <[email protected]>---
PATCH v3 4/4 doc: remove reference to RTE_MACHINE_CPUFLAG
In reply to Radu NicolauRTE_MACHINE_CPUFLAG macros are replaced with predefined compiler defines.
--- doc/guides/prog_guide/writing_efficient_code.rst | 2 -- doc/guides/rel_notes/release_20_11.rst | 1 + 2 files changed, 1 insertion(+), 2 deletions(-)
doc/guides/prog_guide/writing_efficient_code.rst +0 -2
@@ -273,5 +273,3 @@ main() function and checks if the current machine is suitable for running the bi Along with compiler optimizations, a set of preprocessor defines are automatically added to the build process (regardless of the compiler version). These defines correspond to the instruction sets that the target CPU should be able to support. -For example, a binary compiled for any SSE4.2-capable processor will have RTE_MACHINE_CPUFLAG_SSE4_2 defined, -thus enabling compile-time code path selection for different platforms.
doc/guides/rel_notes/release_20_11.rst +1 -0
@@ -91,6 +91,7 @@ Removed Items Also, make sure to start the actual text at the margin. ======================================================= +* Removed ``RTE_MACHINE_CPUFLAG`` macros. API Changes -----------
Signature · 1 line
2.17.1
Quoted 42 lines · Radu Nicolau
Remove
RTE_MACHINE_CPUFLAG_macros from the build. Deprecation notice sent, pasted here for reference:* build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example,
RTE_MACHINE_CPUFLAG_SSE3duplicates the compiler-provided macro__SSE3__.Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference toRTE_MACHINE_CPUFLAGapp/test-pmd/macswap.c | 2 +- app/test/test_memcpy_perf.c | 8 ++++---- config/arm/meson.build | 6 ------ config/ppc/meson.build | 2 -- config/x86/meson.build | 2 -- doc/guides/prog_guide/writing_efficient_code.rst | 2 -- doc/guides/rel_notes/release_20_11.rst | 1 + drivers/net/enic/meson.build | 2 +- drivers/net/i40e/meson.build | 2 +- drivers/net/iavf/meson.build | 2 +- drivers/net/ice/meson.build | 2 +- drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 16 ++++++++-------- examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_acl/meson.build | 2 +- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_eal/common/rte_random.c | 4 ++-- lib/librte_eal/x86/include/rte_memcpy.h | 8 ++++---- lib/librte_efd/rte_efd_x86.h | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 4 ++-- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +-
Quoted 8 lines
lib/librte_member/rte_member_ht.c | 10 +++++----- lib/librte_member/rte_member_x86.h | 2 +- lib/librte_net/rte_net_crc.c | 4 ++-- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- 33 files changed, 52 insertions(+), 63 deletions(-)
Rewrote the release note update based on the deprecation notice. Dropped the notice.
Applied, thanks.
Signature · 1 line
David Marchand
PATCH 2/4 arm: change cpuflag macros to compiler macros
In reply to Radu NicolauReplace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros.
--- app/test-pmd/macswap.c | 2 +- config/arm/meson.build | 6 ------ drivers/net/ixgbe/ixgbe_ethdev.c | 2 +- examples/l3fwd/l3fwd.h | 2 +- examples/l3fwd/l3fwd_em.c | 12 ++++++------ examples/l3fwd/l3fwd_em_hlm.h | 2 +- examples/l3fwd/l3fwd_em_sequential.h | 2 +- examples/l3fwd/l3fwd_lpm.c | 6 +++--- lib/librte_eal/arm/include/rte_memcpy_32.h | 2 +- lib/librte_hash/Makefile | 2 +- lib/librte_hash/rte_cuckoo_hash.c | 2 +- lib/librte_hash/rte_hash_crc.h | 2 +- lib/librte_hash/rte_thash.h | 4 ++-- lib/librte_member/rte_member.h | 2 +- lib/librte_net/rte_net_crc.c | 2 +- lib/librte_node/ip4_lookup.c | 2 +- lib/librte_sched/rte_sched.c | 2 +- lib/librte_table/rte_lru_arm64.h | 2 +- lib/librte_table/rte_table_hash_func.h | 2 +- 19 files changed, 26 insertions(+), 32 deletions(-)
app/test-pmd/macswap.c +1 -1
@@ -39,7 +39,7 @@#include "testpmd.h" #if defined(RTE_ARCH_X86 )#include "macswap_sse.h" -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM__NEON)#include "macswap_neon.h" #else #include "macswap.h"
config/arm/meson.build +0 -6
@@ -208,20 +208,14 @@ message(machine_args) if (cc.get_define('__ARM_NEON', args: machine_args) != '' or cc.get_define('__aarch64__', args: machine_args) != '') - dpdk_conf.set('RTE_MACHINE_CPUFLAG_NEON', 1) compile_time_cpuflags += ['RTE_CPUFLAG_NEON'] endif if cc.get_define('__ARM_FEATURE_CRC32', args: machine_args) != '' - dpdk_conf.set('RTE_MACHINE_CPUFLAG_CRC32', 1) compile_time_cpuflags += ['RTE_CPUFLAG_CRC32'] endif if cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '' - dpdk_conf.set('RTE_MACHINE_CPUFLAG_AES', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_PMULL', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_SHA1', 1) - dpdk_conf.set('RTE_MACHINE_CPUFLAG_SHA2', 1) compile_time_cpuflags += ['RTE_CPUFLAG_AES', 'RTE_CPUFLAG_PMULL', 'RTE_CPUFLAG_SHA1', 'RTE_CPUFLAG_SHA2'] endif
drivers/net/ixgbe/ixgbe_ethdev.c +1 -1
@@ -3960,7 +3960,7 @@ ixgbe_dev_supported_ptypes_get(struct rte_eth_dev *dev) dev->rx_pkt_burst == ixgbe_recv_pkts_bulk_alloc)return ptypes; -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_NEON)if (dev->rx_pkt_burst == ixgbe_recv_pkts_vec || dev->rx_pkt_burst == ixgbe_recv_scattered_pkts_vec)return ptypes;
examples/l3fwd/l3fwd.h +1 -1
@@ -12,7 +12,7 @@#define RTE_LOGTYPE_L3FWD RTE_LOGTYPE_USER1 -#if !defined(NO_HASH_MULTI_LOOKUP ) && defined(RTE_MACHINE_CPUFLAG_NEON ) +#if !defined(NO_HASH_MULTI_LOOKUP ) && defined(__ARM_NEON)#define NO_HASH_MULTI_LOOKUP 1#endif
examples/l3fwd/l3fwd_em.c +6 -6
@@ -28,7 +28,7 @@#include "l3fwd.h" #include "l3fwd_event.h" -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32)#define EM_HASH_CRC 1#endif
@@ -223,7 +223,7 @@ em_mask_key(void *key, xmm_t mask) return _mm_and_si128(data, mask); } -#elif defined(RTE_MACHINE_CPUFLAG_NEON) +#elif defined(__ARM_NEON)static inline xmm_t em_mask_key (void *key,xmm_t mask) {
@@ -303,7 +303,7 @@ em_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)return (ret <0 ) ? portid : ipv6_l3fwd_out_if[ret]; } -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEON#if defined(NO_HASH_MULTI_LOOKUP )#include "l3fwd_em_sequential.h" #else
@@ -685,7 +685,7 @@ em_main_loop(__rte_unused void *dummy)if (nb_rx ==0 )continue ; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEONl3fwd_em_send_packets (nb_rx, pkts_burst, portid, qconf);#else
@@ -723,7 +723,7 @@ em_event_loop_single(struct l3fwd_event_resources *evt_rsrc,struct rte_mbuf *mbuf = ev.mbuf; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEON mbuf->port =em_get_dst_port (lconf, mbuf, mbuf->port);process_packet (mbuf, &mbuf->port);#else
@@ -784,7 +784,7 @@ em_event_loop_burst(struct l3fwd_event_resources *evt_rsrc,continue ; } -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON +#if definedRTE_ARCH_X86 || defined __ARM_NEONl3fwd_em_process_events (nb_deq, (struct rte_event **)&events, lconf);#else
examples/l3fwd/l3fwd_em_hlm.h +1 -1
@@ -9,7 +9,7 @@#if definedRTE_ARCH_X86 #include "l3fwd_sse.h" #include "l3fwd_em_hlm_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_neon.h" #include "l3fwd_em_hlm_neon.h" #endif
examples/l3fwd/l3fwd_em_sequential.h +1 -1
@@ -16,7 +16,7 @@#if definedRTE_ARCH_X86 #include "l3fwd_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_neon.h" #endif
examples/l3fwd/l3fwd_lpm.c +3 -3
@@ -163,7 +163,7 @@ lpm_get_dst_port_with_ipv4(const struct lcore_conf *qconf, struct rte_mbuf *pkt,#if defined(RTE_ARCH_X86 )#include "l3fwd_lpm_sse.h" -#elif definedRTE_MACHINE_CPUFLAG_NEON +#elif defined __ARM_NEON#include "l3fwd_lpm_neon.h" #elif defined(RTE_ARCH_PPC_64 )#include "l3fwd_lpm_altivec.h"
@@ -240,7 +240,7 @@ lpm_main_loop(__rte_unused void *dummy)if (nb_rx ==0 )continue ; -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON \ +#if definedRTE_ARCH_X86 || defined __ARM_NEON \ || definedRTE_ARCH_PPC_64 l3fwd_lpm_send_packets (nb_rx, pkts_burst, portid, qconf);
@@ -259,7 +259,7 @@ lpm_process_event_pkt(const struct lcore_conf *lconf, struct rte_mbuf *mbuf) { mbuf->port =lpm_get_dst_port (lconf, mbuf, mbuf->port); -#if definedRTE_ARCH_X86 || definedRTE_MACHINE_CPUFLAG_NEON \ +#if definedRTE_ARCH_X86 || defined __ARM_NEON \ || definedRTE_ARCH_PPC_64 process_packet (mbuf, &mbuf->port);#else
lib/librte_eal/arm/include/rte_memcpy_32.h +1 -1
@@ -16,7 +16,7 @@ extern "C" {#ifdef RTE_ARCH_ARM_NEON_MEMCPY -#ifndef RTE_MACHINE_CPUFLAG_NEON +#ifndef __ARM_NEON#error "Cannot optimize memcpy by NEON as the CPU seems to not support this"#endif
lib/librte_hash/Makefile +1 -1
@@ -20,7 +20,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_HASH) += rte_fbk_hash.c SYMLINK-$(CONFIG_RTE_LIBRTE_HASH)-include := rte_hash.h SYMLINK-$(CONFIG_RTE_LIBRTE_HASH)-include += rte_hash_crc.h ifeq ($(CONFIG_RTE_ARCH_ARM64),y) -ifneq ($(findstring RTE_MACHINE_CPUFLAG_CRC32,$(CFLAGS)),) +ifneq ($(filter $(AUTO_CPUFLAGS),__ARM_FEATURE_CRC32),) SYMLINK-$(CONFIG_RTE_LIBRTE_HASH)-include += rte_crc_arm64.h endif endif
lib/librte_hash/rte_cuckoo_hash.c +1 -1
@@ -1704,7 +1704,7 @@ compare_signatures(uint32_t *prim_hash_matches, uint32_t *sec_hash_matches, (__m128i const *)sec_bkt->sig_current), _mm_set1_epi16(sig)));break ; -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)case RTE_HASH_COMPARE_NEON : {uint16x8_t vmat, vsig, x;int16x8_t shift = {-15 ,-13 ,-11 ,-9 ,-7 ,-5 ,-3 ,-1 };
lib/librte_hash/rte_hash_crc.h +1 -1
@@ -424,7 +424,7 @@ crc32c_sse42_u64(uint64_t data, uint64_t init_val)static uint8_t crc32_alg =CRC32_SW ; -#if defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)#include "rte_crc_arm64.h" #else
lib/librte_hash/rte_thash.h +2 -2
@@ -28,7 +28,7 @@ extern "C" {#include <rte_ip.h> #include <rte_common.h> -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_NEON)#include <rte_vect.h> #endif
@@ -149,7 +149,7 @@ rte_thash_load_v6_addrs(const struct rte_ipv6_hdr *orig, ipv6 =_mm_loadu_si128 ((const __m128i *)orig->dst_addr); *(__m128i *)targ->v6.dst_addr =_mm_shuffle_epi8 (ipv6, rte_thash_ipv6_bswap_mask); -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)uint8x16_t ipv6 =vld1q_u8 ((uint8_t const *)orig->src_addr);vst1q_u8 ((uint8_t *)targ->v6.src_addr,vrev32q_u8 (ipv6)); ipv6 =vld1q_u8 ((uint8_t const *)orig->dst_addr);
lib/librte_member/rte_member.h +1 -1
@@ -68,7 +68,7 @@ typedef uint16_t member_set_t;#define RTE_MEMBER_NAMESIZE 32/** @internal Hash function used by membership library. */ -#if defined(RTE_ARCH_X86 ) || defined(RTE_MACHINE_CPUFLAG_CRC32 ) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32)#include <rte_hash_crc.h> #define MEMBER_HASH_FUNC rte_hash_crc#else
lib/librte_net/rte_net_crc.c +1 -1
@@ -12,7 +12,7 @@#if defined(RTE_ARCH_X86_64 ) && defined(__PCLMUL__)#define X86_64_SSE42_PCLMULQDQ 1 -#elif defined(RTE_ARCH_ARM64 ) && defined(RTE_MACHINE_CPUFLAG_PMULL ) +#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRYPTO)#define ARM64_NEON_PMULL 1#endif
lib/librte_node/ip4_lookup.c +1 -1
@@ -30,7 +30,7 @@ struct ip4_lookup_node_main {static struct ip4_lookup_node_main ip4_lookup_nm; -#if defined(RTE_MACHINE_CPUFLAG_NEON ) +#if defined(__ARM_NEON)#include "ip4_lookup_neon.h" #elif defined(RTE_ARCH_X86 )#include "ip4_lookup_sse.h"
lib/librte_sched/rte_sched.c +1 -1
@@ -29,7 +29,7 @@#ifdef RTE_ARCH_X86 #define SCHED_VECTOR_SSE4 -#elif defined(RTE_MACHINE_CPUFLAG_NEON ) +#elif defined(__ARM_NEON)#define SCHED_VECTOR_NEON #endif
lib/librte_table/rte_lru_arm64.h +1 -1
@@ -13,7 +13,7 @@ extern "C" {#include <rte_vect.h> #ifndef RTE_TABLE_HASH_LRU_STRATEGY -#ifdef RTE_MACHINE_CPUFLAG_NEON +#ifdef __ARM_NEON#define RTE_TABLE_HASH_LRU_STRATEGY 3#else /* if no NEON, use simple scalar version */ #define RTE_TABLE_HASH_LRU_STRATEGY 1
lib/librte_table/rte_table_hash_func.h +1 -1
@@ -41,7 +41,7 @@ rte_crc32_u64(uint64_t crc, uint64_t v) return _mm_crc32_u64(crc, v); } -#elif defined(RTE_ARCH_ARM64) && defined(RTE_MACHINE_CPUFLAG_CRC32) +#elif defined(RTE_ARCH_ARM64) && defined(__ARM_FEATURE_CRC32)#include "rte_table_hash_func_arm64.h" #else
Signature · 1 line
2.17.1
PATCH 3/4 ppc: change cpuflag macros to compiler macros
In reply to Radu NicolauReplace use of RTE_MACHINE_CPUFLAG macros with regular compiler macros.
--- config/ppc/meson.build | 2 -- 1 file changed, 2 deletions(-)
config/ppc/meson.build +0 -2
@@ -21,5 +21,3 @@ endif dpdk_conf.set('RTE_MAX_LCORE', 1536) dpdk_conf.set('RTE_MAX_NUMA_NODES', 32) dpdk_conf.set('RTE_CACHE_LINE_SIZE', 128) -dpdk_conf.set('RTE_MACHINE_CPUFLAG_ALTIVEC', 1) -dpdk_conf.set('RTE_MACHINE_CPUFLAG_VSX', 1)
Signature · 1 line
2.17.1
PATCH 4/4 doc: remove reference to RTE_MACHINE_CPUFLAG
In reply to Radu NicolauRTE_MACHINE_CPUFLAG macros are replaced with predefined compiler defines.
--- doc/guides/prog_guide/writing_efficient_code.rst | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
doc/guides/prog_guide/writing_efficient_code.rst +1 -1
@@ -273,5 +273,5 @@ main() function and checks if the current machine is suitable for running the bi Along with compiler optimizations, a set of preprocessor defines are automatically added to the build process (regardless of the compiler version). These defines correspond to the instruction sets that the target CPU should be able to support. -For example, a binary compiled for any SSE4.2-capable processor will have RTE_MACHINE_CPUFLAG_SSE4_2 defined, +For example, a binary compiled for any SSE4.2-capable processor will have a pre-defined compiler macro, thus enabling compile-time code path selection for different platforms.
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v1 4/4] doc: remove reference to RTE_MACHINE_CPUFLAG
Quoted 15 lines · Radu Nicolau
RTE_MACHINE_CPUFLAGmacros are replaced with predefined compiler defines.Signed-off-by Sean Morrissey <[email protected]>Signed-off-by Radu Nicolau <[email protected]>--- doc/guides/prog_guide/writing_efficient_code.rst | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)doc/guides/prog_guide/writing_efficient_code.rst +0
-0
doc/guides/prog_guide/writing_efficient_code.rst main() function and checks if the current machine is suitable for running the bi · line 277
Quoted 6 lines
doc/guides/prog_guide/writing_efficient_code.rst +1
-1@@ -273,5 +273,5 @@ main() function and checks if the current machine is suitable for running the bi Along with compiler optimizations, a set of preprocessor defines are automatically added to the build process (regardless of the compiler version). These defines correspond to the instruction sets that the target CPU should be able to support. -For example, a binary compiled for any SSE4.2-capable processor will have RTE_MACHINE_CPUFLAG_SSE4_2 defined, +For example, a binary compiled for any SSE4.2-capable processor will have a pre-defined compiler macro, thus enabling compile-time code path selection for different platforms.
Personally, I'd suggest just removing the whole section rather than part of a single line.
Quoted 8 lines · Radu Nicolau
Remove
RTE_MACHINE_CPUFLAG_macros from the build. Deprecation notice sent, pasted here for reference:build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example,
RTE_MACHINE_CPUFLAG_SSE3duplicates the compiler-provided macro__SSE3__.
Quoted 5 lines
Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference toRTE_MACHINE_CPUFLAG
I'm not too familiar with clang and icc. Do all compilers use the same macro definitions for the same CPU features? I would have thought the RTE_* definitions were there because there are compiler or compiler version differences that need to be supported.
Dave
Quoted 15 lines · David Christensen
Quoted 8 lines · Radu Nicolau
Remove
RTE_MACHINE_CPUFLAG_macros from the build. Deprecation notice sent, pasted here for reference:build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example,
RTE_MACHINE_CPUFLAG_SSE3duplicates the compiler-provided macro__SSE3__.Quoted 5 lines
Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference toRTE_MACHINE_CPUFLAG
Quoted 3 lines
I'm not too familiar with clang and icc. Do all compilers use the same macro definitions for the same CPU features? I would have thought the RTE_* definitions were there because there are compiler or compiler version differences that need to be supported.
Dave
All supported compilers should have these defines, and the RTE_MACHINE_CPU macros were defined based on the compiler defined macros, as an unnecessary extra step.
Furthermore, as per Bruce's comment, new features will be easily implemented directly in the relevant sections without needing to update the build system.
Hello Radu,
Quoted 9 lines · Radu Nicolau
Remove
RTE_MACHINE_CPUFLAG_macros from the build. Deprecation notice sent, pasted here for reference:build macros: The macros defining RTE_MACHINE_CPUFLAG_* will be removed from the build. The information provided by these macros is available through standard compiler macros. For example,
RTE_MACHINE_CPUFLAG_SSE3duplicates the compiler-provided macro__SSE3__.
Quoted 5 lines
Radu Nicolau (4):
x86: change cpuflag macros to compiler macros
arm: change cpuflag macros to compiler macros
ppc: change cpuflag macros to compiler macros
doc: remove reference toRTE_MACHINE_CPUFLAG
Expecting a v2 to address Bruce comment.
Thanks.
Signature · 1 line
David Marchand