/* { dg-do assemble { target aarch64_asm_sve2p1_ok } } */ /* { dg-do compile { target { ! aarch64_asm_sve2p1_ok } } } */ /* { dg-skip-if "" { *-*-* } { "-DSTREAMING_COMPATIBLE" } { "" } } */ /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */ #include "test_sve_acle.h" #pragma GCC target "+sve2p1" /* ** ld1udq_f64_base: ** ld1d {z0\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld1udq_f64_base, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0), z0 = svld1udq (p0, x0)) /* ** ld1udq_f64_index: ** ld1d {z0\.q}, p0/z, \[x0, x1, lsl 3\] ** ret */ TEST_LOAD (ld1udq_f64_index, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 + x1), z0 = svld1udq (p0, x0 + x1)) /* ** ld1udq_f64_1: ** ld1d {z0\.q}, p0/z, \[x0, #2, mul vl\] ** ret */ TEST_LOAD (ld1udq_f64_1, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 + svcntd ()), z0 = svld1udq (p0, x0 + svcntd ())) /* ** ld1udq_f64_7: ** ld1d {z0\.q}, p0/z, \[x0, #6, mul vl\] ** ret */ TEST_LOAD (ld1udq_f64_7, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 + svcntd () * 3), z0 = svld1udq (p0, x0 + svcntd () * 3)) /* Moving the constant into a register would also be OK. */ /* ** ld1udq_f64_8: ** incb x0, all, mul #4 ** ld1d {z0\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld1udq_f64_8, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 + svcntd () * 4), z0 = svld1udq (p0, x0 + svcntd () * 4)) /* ** ld1udq_f64_m2: ** ld1d {z0\.q}, p0/z, \[x0, #-2, mul vl\] ** ret */ TEST_LOAD (ld1udq_f64_m2, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 - svcntd ()), z0 = svld1udq (p0, x0 - svcntd ())) /* ** ld1udq_f64_m8: ** ld1d {z0\.q}, p0/z, \[x0, #-8, mul vl\] ** ret */ TEST_LOAD (ld1udq_f64_m8, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 - svcntd () * 4), z0 = svld1udq (p0, x0 - svcntd () * 4)) /* Moving the constant into a register would also be OK. */ /* ** ld1udq_f64_m10: ** decb x0, all, mul #5 ** ld1d {z0\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld1udq_f64_m10, svfloat64_t, float64_t, z0 = svld1udq_f64 (p0, x0 - svcntd () * 5), z0 = svld1udq (p0, x0 - svcntd () * 5)) /* ** ld1udq_vnum_f64_0: ** ld1d {z0\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_0, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, 0), z0 = svld1udq_vnum (p0, x0, 0)) /* ** ld1udq_vnum_f64_1: ** ld1d {z0\.q}, p0/z, \[x0, #1, mul vl\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_1, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, 1), z0 = svld1udq_vnum (p0, x0, 1)) /* ** ld1udq_vnum_f64_7: ** ld1d {z0\.q}, p0/z, \[x0, #7, mul vl\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_7, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, 7), z0 = svld1udq_vnum (p0, x0, 7)) /* Moving the constant into a register would also be OK. */ /* ** ld1udq_vnum_f64_8: ** incb x0, all, mul #4 ** ld1d {z0\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_8, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, 8), z0 = svld1udq_vnum (p0, x0, 8)) /* ** ld1udq_vnum_f64_m1: ** ld1d {z0\.q}, p0/z, \[x0, #-1, mul vl\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_m1, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, -1), z0 = svld1udq_vnum (p0, x0, -1)) /* ** ld1udq_vnum_f64_m8: ** ld1d {z0\.q}, p0/z, \[x0, #-8, mul vl\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_m8, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, -8), z0 = svld1udq_vnum (p0, x0, -8)) /* Moving the constant into a register would also be OK. */ /* ** ld1udq_vnum_f64_m9: ** dech x0, all, mul #9 ** ld1d {z0\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_m9, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, -9), z0 = svld1udq_vnum (p0, x0, -9)) /* Using MUL to calculate an index would also be OK. */ /* ** ld1udq_vnum_f64_x1: ** cnth (x[0-9]+) ** madd (x[0-9]+), (x1, \1|\1, x1), x0 ** ld1d {z0\.q}, p0/z, \[\2\] ** ret */ TEST_LOAD (ld1udq_vnum_f64_x1, svfloat64_t, float64_t, z0 = svld1udq_vnum_f64 (p0, x0, x1), z0 = svld1udq_vnum (p0, x0, x1))