| Message ID | 20221204015123.362726-2-richard.henderson@linaro.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Use a more portable way to enable target specific functions | expand |
On Sat, Dec 03, 2022 at 07:51:22PM -0600, Richard Henderson wrote: > Use the attribute, which is supported by clang, instead of > the #pragma, which is not supported and, for some reason, > also not detected by the meson probe, so we fail by -Werror. > > Signed-off-by: Richard Henderson <richard.henderson@linaro.org> > --- > meson.build | 8 ++------ > util/bufferiszero.c | 41 ++++++----------------------------------- > 2 files changed, 8 insertions(+), 41 deletions(-) > > diff --git a/util/bufferiszero.c b/util/bufferiszero.c > index ec3cd4ca15..1790ded7d4 100644 > --- a/util/bufferiszero.c > +++ b/util/bufferiszero.c > @@ -64,18 +64,11 @@ buffer_zero_int(const void *buf, size_t len) > } > > #if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) || defined(__SSE2__) > -/* Do not use push_options pragmas unnecessarily, because clang > - * does not support them. > - */ > -#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) > -#pragma GCC push_options > -#pragma GCC target("sse2") > -#endif > -#include <emmintrin.h> So the old code included emmintrin.h, and possibly either immintrin.h / simmintrin.h, but the new code only includes immintrin.h. I'm not saying that's wrong, I'm just wondering why it is changing, as it feels possibly tangential to the pragma -> attribute conversion. Could you mention this in the commit message, or split it to a separate cleanup patch if its a functionally unrelated change. > +#include <immintrin.h> > > /* Note that each of these vectorized functions require len >= 64. */ > > -static bool > +static bool __attribute__((target("sse2"))) > buffer_zero_sse2(const void *buf, size_t len) > { > __m128i t = _mm_loadu_si128(buf); > @@ -104,20 +97,9 @@ buffer_zero_sse2(const void *buf, size_t len) > > return _mm_movemask_epi8(_mm_cmpeq_epi8(t, zero)) == 0xFFFF; > } > -#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) > -#pragma GCC pop_options > -#endif > > #ifdef CONFIG_AVX2_OPT > -/* Note that due to restrictions/bugs wrt __builtin functions in gcc <= 4.8, > - * the includes have to be within the corresponding push_options region, and > - * therefore the regions themselves have to be ordered with increasing ISA. > - */ > -#pragma GCC push_options > -#pragma GCC target("sse4") > -#include <smmintrin.h> > - > -static bool > +static bool __attribute__((target("sse4"))) > buffer_zero_sse4(const void *buf, size_t len) > { > __m128i t = _mm_loadu_si128(buf); > @@ -145,12 +127,7 @@ buffer_zero_sse4(const void *buf, size_t len) > return _mm_testz_si128(t, t); > } > > -#pragma GCC pop_options > -#pragma GCC push_options > -#pragma GCC target("avx2") > -#include <immintrin.h> > - > -static bool > +static bool __attribute__((target("avx2"))) > buffer_zero_avx2(const void *buf, size_t len) > { > /* Begin with an unaligned head of 32 bytes. */ > @@ -176,15 +153,10 @@ buffer_zero_avx2(const void *buf, size_t len) > > return _mm256_testz_si256(t, t); > } > -#pragma GCC pop_options > #endif /* CONFIG_AVX2_OPT */ > > #ifdef CONFIG_AVX512F_OPT > -#pragma GCC push_options > -#pragma GCC target("avx512f") > -#include <immintrin.h> > - > -static bool > +static bool __attribute__((target("avx512f"))) > buffer_zero_avx512(const void *buf, size_t len) > { > /* Begin with an unaligned head of 64 bytes. */ > @@ -210,8 +182,7 @@ buffer_zero_avx512(const void *buf, size_t len) > return !_mm512_test_epi64_mask(t, t); > > } > -#pragma GCC pop_options > -#endif > +#endif /* CONFIG_AVX512F_OPT */ > > > /* Note that for test_buffer_is_zero_next_accel, the most preferred > -- > 2.34.1 > With regards, Daniel
On 12/5/22 05:17, Daniel P. Berrangé wrote: > On Sat, Dec 03, 2022 at 07:51:22PM -0600, Richard Henderson wrote: >> Use the attribute, which is supported by clang, instead of >> the #pragma, which is not supported and, for some reason, >> also not detected by the meson probe, so we fail by -Werror. >> >> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> >> --- >> meson.build | 8 ++------ >> util/bufferiszero.c | 41 ++++++----------------------------------- >> 2 files changed, 8 insertions(+), 41 deletions(-) >> > > >> diff --git a/util/bufferiszero.c b/util/bufferiszero.c >> index ec3cd4ca15..1790ded7d4 100644 >> --- a/util/bufferiszero.c >> +++ b/util/bufferiszero.c >> @@ -64,18 +64,11 @@ buffer_zero_int(const void *buf, size_t len) >> } >> >> #if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) || defined(__SSE2__) >> -/* Do not use push_options pragmas unnecessarily, because clang >> - * does not support them. >> - */ >> -#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) >> -#pragma GCC push_options >> -#pragma GCC target("sse2") >> -#endif >> -#include <emmintrin.h> > > So the old code included emmintrin.h, and possibly either > immintrin.h / simmintrin.h, but the new code only > includes immintrin.h. > > I'm not saying that's wrong, I'm just wondering why it is > changing, as it feels possibly tangential to the pragma > -> attribute conversion. Could you mention this in the > commit message, or split it to a separate cleanup patch > if its a functionally unrelated change. Adding Include only <immintrin.h> as that is the outermost "official" header for these intrinsics -- emmintrin.h and smmintrin.> are older SSE2 and SSE4 specific headers, while the immintrin.h includes all of the Intel intrinsics. r~
diff --git a/meson.build b/meson.build index 5c6b5a1c75..11b873f911 100644 --- a/meson.build +++ b/meson.build @@ -2324,11 +2324,9 @@ config_host_data.set('CONFIG_CPUID_H', have_cpuid_h) config_host_data.set('CONFIG_AVX2_OPT', get_option('avx2') \ .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot enable AVX2') \ .require(cc.links(''' - #pragma GCC push_options - #pragma GCC target("avx2") #include <cpuid.h> #include <immintrin.h> - static int bar(void *a) { + static int __attribute__((target("avx2"))) bar(void *a) { __m256i x = *(__m256i *)a; return _mm256_testz_si256(x, x); } @@ -2338,11 +2336,9 @@ config_host_data.set('CONFIG_AVX2_OPT', get_option('avx2') \ config_host_data.set('CONFIG_AVX512F_OPT', get_option('avx512f') \ .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot enable AVX512F') \ .require(cc.links(''' - #pragma GCC push_options - #pragma GCC target("avx512f") #include <cpuid.h> #include <immintrin.h> - static int bar(void *a) { + static int __attribute__((target("avx512f"))) bar(void *a) { __m512i x = *(__m512i *)a; return _mm512_test_epi64_mask(x, x); } diff --git a/util/bufferiszero.c b/util/bufferiszero.c index ec3cd4ca15..1790ded7d4 100644 --- a/util/bufferiszero.c +++ b/util/bufferiszero.c @@ -64,18 +64,11 @@ buffer_zero_int(const void *buf, size_t len) } #if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) || defined(__SSE2__) -/* Do not use push_options pragmas unnecessarily, because clang - * does not support them. - */ -#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) -#pragma GCC push_options -#pragma GCC target("sse2") -#endif -#include <emmintrin.h> +#include <immintrin.h> /* Note that each of these vectorized functions require len >= 64. */ -static bool +static bool __attribute__((target("sse2"))) buffer_zero_sse2(const void *buf, size_t len) { __m128i t = _mm_loadu_si128(buf); @@ -104,20 +97,9 @@ buffer_zero_sse2(const void *buf, size_t len) return _mm_movemask_epi8(_mm_cmpeq_epi8(t, zero)) == 0xFFFF; } -#if defined(CONFIG_AVX512F_OPT) || defined(CONFIG_AVX2_OPT) -#pragma GCC pop_options -#endif #ifdef CONFIG_AVX2_OPT -/* Note that due to restrictions/bugs wrt __builtin functions in gcc <= 4.8, - * the includes have to be within the corresponding push_options region, and - * therefore the regions themselves have to be ordered with increasing ISA. - */ -#pragma GCC push_options -#pragma GCC target("sse4") -#include <smmintrin.h> - -static bool +static bool __attribute__((target("sse4"))) buffer_zero_sse4(const void *buf, size_t len) { __m128i t = _mm_loadu_si128(buf); @@ -145,12 +127,7 @@ buffer_zero_sse4(const void *buf, size_t len) return _mm_testz_si128(t, t); } -#pragma GCC pop_options -#pragma GCC push_options -#pragma GCC target("avx2") -#include <immintrin.h> - -static bool +static bool __attribute__((target("avx2"))) buffer_zero_avx2(const void *buf, size_t len) { /* Begin with an unaligned head of 32 bytes. */ @@ -176,15 +153,10 @@ buffer_zero_avx2(const void *buf, size_t len) return _mm256_testz_si256(t, t); } -#pragma GCC pop_options #endif /* CONFIG_AVX2_OPT */ #ifdef CONFIG_AVX512F_OPT -#pragma GCC push_options -#pragma GCC target("avx512f") -#include <immintrin.h> - -static bool +static bool __attribute__((target("avx512f"))) buffer_zero_avx512(const void *buf, size_t len) { /* Begin with an unaligned head of 64 bytes. */ @@ -210,8 +182,7 @@ buffer_zero_avx512(const void *buf, size_t len) return !_mm512_test_epi64_mask(t, t); } -#pragma GCC pop_options -#endif +#endif /* CONFIG_AVX512F_OPT */ /* Note that for test_buffer_is_zero_next_accel, the most preferred
Use the attribute, which is supported by clang, instead of the #pragma, which is not supported and, for some reason, also not detected by the meson probe, so we fail by -Werror. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- meson.build | 8 ++------ util/bufferiszero.c | 41 ++++++----------------------------------- 2 files changed, 8 insertions(+), 41 deletions(-)