/* { dg-do run { target { powerpc*-*-linux* && lp64 } } } */ /* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */ /* { dg-skip-if "" { powerpc*-*-*spe* } { "*" } { "" } } */ /* { dg-require-effective-target p8vector_hw } */ /* { dg-options "-mcpu=power8 -O2" } */ #include #include #include #ifdef DEBUG #include #define UNUSED #ifdef __LITTLE_ENDIAN__ #define HI_WORD 1 #define LO_WORD 0 #else #define HI_WORD 0 #define LO_WORD 1 #endif #else #define UNUSED __attribute__((__unused__)) #endif #ifndef S_TYPE #define S_TYPE __uint128_t #endif #ifndef V_TYPE #define V_TYPE vector S_TYPE #endif static int compare (S_TYPE, V_TYPE, const char *, const char *) __attribute__((__noinline__)); static int compare (S_TYPE scalar, V_TYPE vect, const char *nl UNUSED, const char *which UNUSED) { unsigned long scalar_lo = (unsigned long) scalar; unsigned long scalar_hi = (unsigned long) (scalar >> 64); unsigned long vect_lo; unsigned long vect_hi; vector long long tmp; int ret; __asm__ ("mfvsrd %0,%x3\n\t" "xxpermdi %x2,%x3,%x3,3\n\t" "mfvsrd %1,%x2" : "=r" (vect_hi), "=r" (vect_lo), "=wa" (tmp) : "wa" (vect)); ret = (scalar_lo != vect_lo) || (scalar_hi != vect_hi); #ifdef DEBUG printf ("%s%s: 0x%.16lx %.16lx %s 0x%.16lx %.16lx\n", nl, which, scalar_hi, scalar_lo, (ret) ? "!=" : "==", vect_hi, vect_lo); fflush (stdout); #endif return ret; } static void convert_via_mem (V_TYPE *, S_TYPE *) __attribute__((__noinline__)); static void convert_via_mem (V_TYPE *v, S_TYPE *s) { *v = (V_TYPE) { *s }; __asm__ volatile ("nop" : "+m" (*s), "+m" (*v) : : "memory"); } /* Check if vadduqm returns the same values as normal 128-bit add. */ /* Values to add together. */ const static struct { unsigned long hi_1; unsigned long lo_1; unsigned long hi_2; unsigned long lo_2; } values[] = { { 0x0000000000000000UL, 0xfffffffffffffffeUL, 0x0000000000000000UL, 0x0000000000000002UL }, { 0x0000000000000000UL, 0x0000000000000002UL, 0x0000000000000000UL, 0xfffffffffffffffeUL }, { 0xffffffffffffffffUL, 0xfffffffffffffffeUL, 0x0000000000000000UL, 0x0000000000000002UL }, { 0xfffffffffffffff2UL, 0xffffffffffffffffUL, 0x0000000000000002UL, 0x0000000000000000UL }, { 0x7fffffffffffffffUL, 0xfffffffffffffffeUL, 0x0000000000000000UL, 0x0000000000000002UL }, { 0x7ffffffffffffff2UL, 0xffffffffffffffffUL, 0x0000000000000002UL, 0x0000000000000000UL }, }; int main (void) { int reg_errors = 0; int mem_errors = 0; size_t i; const char *nl = ""; for (i = 0; i < sizeof (values) / sizeof (values[0]); i++) { S_TYPE s_reg_res, s_reg_in1, s_reg_in2, s_mem_res, s_mem_in1, s_mem_in2; V_TYPE v_reg_res, v_reg_in1, v_reg_in2, v_mem_res, v_mem_in1, v_mem_in2; s_reg_in1 = ((((S_TYPE)values[i].hi_1 << 64)) + ((S_TYPE)values[i].lo_1)); reg_errors += compare (s_reg_in1, (V_TYPE) { s_reg_in1 }, nl, "reg, in1"); s_reg_in2 = ((((S_TYPE)values[i].hi_2 << 64)) + ((S_TYPE)values[i].lo_2)); reg_errors += compare (s_reg_in2, (V_TYPE) { s_reg_in2 }, "", "reg, in2"); s_reg_res = s_reg_in1 + s_reg_in2; v_reg_in1 = (V_TYPE) { s_reg_in1 }; v_reg_in2 = (V_TYPE) { s_reg_in2 }; v_reg_res = vec_vadduqm (v_reg_in1, v_reg_in2); reg_errors += compare (s_reg_res, v_reg_res, "", "reg, res"); s_mem_in1 = s_reg_in1; convert_via_mem (&v_mem_in1, &s_mem_in1); mem_errors += compare (s_mem_in1, (V_TYPE) { s_mem_in1 }, "\n", "mem, in1"); s_mem_in2 = s_reg_in2; convert_via_mem (&v_mem_in2, &s_mem_in2); mem_errors += compare (s_mem_in2, (V_TYPE) { s_mem_in2 }, "", "mem, in2"); s_mem_res = s_mem_in1 + s_mem_in2; v_mem_res = vec_vadduqm (v_mem_in1, v_mem_in2); mem_errors += compare (s_mem_res, v_mem_res, "", "mem, res"); nl = "\n"; } #ifdef DEBUG putchar ('\n'); if (!reg_errors) fputs ("no errors found on register operations\n", stdout); else printf ("%d error%s found on register operations\n", reg_errors, (reg_errors == 1) ? "s" : ""); if (!mem_errors) fputs ("no errors found on memory operations\n", stdout); else printf ("%d error%s found on memory operations\n", mem_errors, (mem_errors == 1) ? "s" : ""); fflush (stdout); #endif if ((reg_errors + mem_errors) != 0) abort (); return 0; }