/* { 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" /* ** st2q_bf16_base: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_bf16_base, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0, z0), svst2q (p0, x0, z0)) /* ** st2q_bf16_index: ** add (x[0-9]), x0, x1, lsl #?1 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[\1\] ** ret */ TEST_STORE (st2q_bf16_index, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + x1, z0), svst2q (p0, x0 + x1, z0)) /* ** st2q_bf16_index2: ** add (x[0-9]), x0, x1, lsl #?2 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[\1\] ** ret */ TEST_STORE (st2q_bf16_index2, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + x1 * 2, z0), svst2q (p0, x0 + x1 * 2, z0)) /* ** st2q_bf16_index4: ** add (x[0-9]), x0, x1, lsl #?3 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[\1\] ** ret */ TEST_STORE (st2q_bf16_index4, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + x1 * 4, z0), svst2q (p0, x0 + x1 * 4, z0)) /* ** st2q_bf16_index8: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, x1, lsl #?4\] ** ret */ TEST_STORE (st2q_bf16_index8, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + x1 * 8, z0), svst2q (p0, x0 + x1 * 8, z0)) /* Moving the constant into a register would also be OK. */ /* ** st2q_bf16_1: ** incb x0 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_bf16_1, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + svcnth (), z0), svst2q (p0, x0 + svcnth (), z0)) /* ** st2q_bf16_2: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #2, mul vl\] ** ret */ TEST_STORE (st2q_bf16_2, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + svcnth () * 2, z0), svst2q (p0, x0 + svcnth () * 2, z0)) /* ** st2q_bf16_14: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #14, mul vl\] ** ret */ TEST_STORE (st2q_bf16_14, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + svcnth () * 14, z0), svst2q (p0, x0 + svcnth () * 14, z0)) /* Moving the constant into a register would also be OK. */ /* ** st2q_bf16_16: ** incb x0, all, mul #16 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_bf16_16, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 + svcnth () * 16, z0), svst2q (p0, x0 + svcnth () * 16, z0)) /* Moving the constant into a register would also be OK. */ /* ** st2q_bf16_m1: ** decb x0 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_bf16_m1, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 - svcnth (), z0), svst2q (p0, x0 - svcnth (), z0)) /* ** st2q_bf16_m2: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #-2, mul vl\] ** ret */ TEST_STORE (st2q_bf16_m2, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 - svcnth () * 2, z0), svst2q (p0, x0 - svcnth () * 2, z0)) /* ** st2q_bf16_m16: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #-16, mul vl\] ** ret */ TEST_STORE (st2q_bf16_m16, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 - svcnth () * 16, z0), svst2q (p0, x0 - svcnth () * 16, z0)) /* ** st2q_bf16_m18: ** addvl (x[0-9]+), x0, #-18 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[\1\] ** ret */ TEST_STORE (st2q_bf16_m18, svbfloat16x2_t, bfloat16_t, svst2q_bf16 (p0, x0 - svcnth () * 18, z0), svst2q (p0, x0 - svcnth () * 18, z0)) /* ** st2q_vnum_bf16_0: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_vnum_bf16_0, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, 0, z0), svst2q_vnum (p0, x0, 0, z0)) /* Moving the constant into a register would also be OK. */ /* ** st2q_vnum_bf16_1: ** incb x0 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_vnum_bf16_1, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, 1, z0), svst2q_vnum (p0, x0, 1, z0)) /* ** st2q_vnum_bf16_2: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #2, mul vl\] ** ret */ TEST_STORE (st2q_vnum_bf16_2, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, 2, z0), svst2q_vnum (p0, x0, 2, z0)) /* ** st2q_vnum_bf16_14: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #14, mul vl\] ** ret */ TEST_STORE (st2q_vnum_bf16_14, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, 14, z0), svst2q_vnum (p0, x0, 14, z0)) /* Moving the constant into a register would also be OK. */ /* ** st2q_vnum_bf16_16: ** incb x0, all, mul #16 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_vnum_bf16_16, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, 16, z0), svst2q_vnum (p0, x0, 16, z0)) /* Moving the constant into a register would also be OK. */ /* ** st2q_vnum_bf16_m1: ** decb x0 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0\] ** ret */ TEST_STORE (st2q_vnum_bf16_m1, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, -1, z0), svst2q_vnum (p0, x0, -1, z0)) /* ** st2q_vnum_bf16_m2: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #-2, mul vl\] ** ret */ TEST_STORE (st2q_vnum_bf16_m2, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, -2, z0), svst2q_vnum (p0, x0, -2, z0)) /* ** st2q_vnum_bf16_m16: ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, #-16, mul vl\] ** ret */ TEST_STORE (st2q_vnum_bf16_m16, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, -16, z0), svst2q_vnum (p0, x0, -16, z0)) /* ** st2q_vnum_bf16_m18: ** addvl (x[0-9]+), x0, #-18 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[\1\] ** ret */ TEST_STORE (st2q_vnum_bf16_m18, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, -18, z0), svst2q_vnum (p0, x0, -18, z0)) /* ** st2q_vnum_bf16_x1: ** cntb (x[0-9]+) ** ( ** madd (x[0-9]+), (?:x1, \1|\1, x1), x0 ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[\2\] ** | ** mul (x[0-9]+), (?:x1, \1|\1, x1) ** st2q {z0\.q(?: - |, )z1\.q}, p0, \[x0, \3\] ** ) ** ret */ TEST_STORE (st2q_vnum_bf16_x1, svbfloat16x2_t, bfloat16_t, svst2q_vnum_bf16 (p0, x0, x1, z0), svst2q_vnum (p0, x0, x1, z0))