Use _mm_set1_epi{32,64x} to init mask in x86-64 [cz]asum

for skylake kernels. This is the same method as used in [sd]asum.
_mm_set1_epi64x was commented out for zasum, but has the advantage
of avoiding possible undefined behaviour (using an uninitialized
variable), optimized out by NVHPC and icx. The new code works
fine with those compilers.

For GCC 12.3 the generated code is identical; no matter what method
you use, the compiler optimizes the code into a compile-time
constant, there is no performance benefit using mm_cmpeq_epi8
since the corresponding instruction (VPCMPEQB) isn't actually
generated!
This commit is contained in:
Bart Oldeman 2023-11-19 21:21:23 +00:00
parent 22aa401656
commit c34e2cf380
2 changed files with 8 additions and 13 deletions

View File

@ -2,10 +2,9 @@
#ifdef __NVCOMPILER #ifdef __NVCOMPILER
#define NVCOMPVERS ( __NVCOMPILER_MAJOR__ * 100 + __NVCOMPILER_MINOR__ ) #define NVCOMPVERS ( __NVCOMPILER_MAJOR__ * 100 + __NVCOMPILER_MINOR__ )
#endif #endif
#if ((( defined(__GNUC__) && __GNUC__ > 6 && defined(__AVX512CD__)) || (defined(__clang__) && (__clang_major__ >= 9 &&__clang_major__ !=17)) || ( defined(__NVCOMPILER) && NVCOMPVERS >= 2309))) #if ((( defined(__GNUC__) && __GNUC__ > 6 && defined(__AVX512CD__)) || (defined(__clang__) && __clang_major__ >= 9)) || (defined(__NVCOMPILER) && NVCOMPVERS >= 2203))
#if (!(defined(__NVCOMPILER) )) #if (!(defined(__NVCOMPILER) && NVCOMPVERS < 2203))
//&& NVCOMPVERS < 2309))
#define HAVE_CASUM_KERNEL 1 #define HAVE_CASUM_KERNEL 1
@ -21,15 +20,14 @@ static FLOAT casum_kernel(BLASLONG n, FLOAT *x)
if (n2 < 64) { if (n2 < 64) {
__m128 accum_10, accum_11, accum_12, accum_13; __m128 accum_10, accum_11, accum_12, accum_13;
__m128 abs_mask1 = abs_mask1; __m128 abs_mask1;
accum_10 = _mm_setzero_ps(); accum_10 = _mm_setzero_ps();
accum_11 = _mm_setzero_ps(); accum_11 = _mm_setzero_ps();
accum_12 = _mm_setzero_ps(); accum_12 = _mm_setzero_ps();
accum_13 = _mm_setzero_ps(); accum_13 = _mm_setzero_ps();
abs_mask1 = (__m128)_mm_cmpeq_epi8((__m128i) abs_mask1, (__m128i) abs_mask1); abs_mask1 = (__m128)_mm_set1_epi32(0x7fffffff);
abs_mask1 = (__m128)_mm_srli_epi32((__m128i) abs_mask1, 1);
_mm_prefetch(&x1[0], _MM_HINT_T0); _mm_prefetch(&x1[0], _MM_HINT_T0);

View File

@ -2,10 +2,9 @@
#ifdef __NVCOMPILER #ifdef __NVCOMPILER
#define NVCOMPVERS ( __NVCOMPILER_MAJOR__ * 100 + __NVCOMPILER_MINOR__ ) #define NVCOMPVERS ( __NVCOMPILER_MAJOR__ * 100 + __NVCOMPILER_MINOR__ )
#endif #endif
#if ((( defined(__GNUC__) && __GNUC__ > 6 && defined(__AVX512CD__)) || (defined(__clang__) && ( __clang_major__ >= 9 && __clang_major__ != 17)) || (defined(__NVCOMPILER) && NVCOMPVERS >= 2309))) #if ((( defined(__GNUC__) && __GNUC__ > 6 && defined(__AVX512CD__)) || (defined(__clang__) && __clang_major__ >= 9)) || (defined(__NVCOMPILER) && NVCOMPVERS >= 2203))
#if (!(defined(__NVCOMPILER) )) #if (!(defined(__NVCOMPILER) && NVCOMPVERS < 2203))
//&& NVCOMPVERS < 2309))
#define HAVE_ZASUM_KERNEL 1 #define HAVE_ZASUM_KERNEL 1
@ -22,16 +21,14 @@ static FLOAT zasum_kernel(BLASLONG n, FLOAT *x)
if (n2 < 32) { if (n2 < 32) {
__m128d accum_10, accum_11, accum_12, accum_13; __m128d accum_10, accum_11, accum_12, accum_13;
__m128d abs_mask1 = abs_mask1; __m128d abs_mask1;
accum_10 = _mm_setzero_pd(); accum_10 = _mm_setzero_pd();
accum_11 = _mm_setzero_pd(); accum_11 = _mm_setzero_pd();
accum_12 = _mm_setzero_pd(); accum_12 = _mm_setzero_pd();
accum_13 = _mm_setzero_pd(); accum_13 = _mm_setzero_pd();
// abs_mask1 = (__m128d)_mm_set1_epi64x(0x7fffffffffffffff); abs_mask1 = (__m128d)_mm_set1_epi64x(0x7fffffffffffffff);
abs_mask1 = (__m128d)_mm_cmpeq_epi8((__m128i) abs_mask1, (__m128i) abs_mask1);
abs_mask1 = (__m128d)_mm_srli_epi64((__m128i) abs_mask1, 1);
_mm_prefetch(&x1[0], _MM_HINT_T0); _mm_prefetch(&x1[0], _MM_HINT_T0);
if (n2 >= 16){ if (n2 >= 16){