1 /* { dg-do compile } */ 2 /* { dg-options "-O2 -ftree-vectorize -fno-trapping-math" } */ 3 4 #include <stdint.h> 5 6 #define DEF_LOOP(FLOAT_TYPE, INT_TYPE) \ 7 void __attribute__ ((noipa)) \ 8 test_##INT_TYPE (FLOAT_TYPE *__restrict r, \ 9 INT_TYPE *__restrict a, \ 10 INT_TYPE *__restrict pred, int n) \ 11 { \ 12 for (int i = 0; i < n; ++i) \ 13 r[i] = pred[i] ? (FLOAT_TYPE) a[i] : 0.0; \ 14 } 15 16 #define TEST_ALL(T) \ 17 T (_Float16, int16_t) \ 18 T (_Float16, uint16_t) \ 19 T (float, int32_t) \ 20 T (float, uint32_t) \ 21 T (double, int64_t) \ 22 T (double, uint64_t) 23 24 TEST_ALL (DEF_LOOP) 25 26 /* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.h, p[0-7]/m,} 1 } } */ 27 /* { dg-final { scan-assembler-times {\tucvtf\tz[0-9]+\.h, p[0-7]/m,} 1 } } */ 28 /* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.s, p[0-7]/m,} 1 } } */ 29 /* { dg-final { scan-assembler-times {\tucvtf\tz[0-9]+\.s, p[0-7]/m,} 1 } } */ 30 /* { dg-final { scan-assembler-times {\tscvtf\tz[0-9]+\.d, p[0-7]/m,} 1 } } */ 31 /* { dg-final { scan-assembler-times {\tucvtf\tz[0-9]+\.d, p[0-7]/m,} 1 } } */ 32 33 /* Really we should be able to use MOVPRFX /z here, but at the moment 34 we're relying on combine to merge a SEL and an arithmetic operation, 35 and the SEL doesn't allow the "false" value to be zero when the "true" 36 value is a register. */ 37 /* { dg-final { scan-assembler-times {\tmovprfx\tz[0-9]+, z[0-9]+\n} 6 } } */ 38 39 /* { dg-final { scan-assembler-not {\tmov\tz[^\n]*z} } } */ 40 /* { dg-final { scan-assembler-not {\tsel\t} } } */ 41