/* { 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" /* ** ld3q_f16_base: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_f16_base, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0), z0 = svld3q (p0, x0)) /* ** ld3q_f16_index: ** add (x[0-9]), x0, x1, lsl #?1 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_f16_index, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + x1), z0 = svld3q (p0, x0 + x1)) /* ** ld3q_f16_index2: ** add (x[0-9]), x0, x1, lsl #?2 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_f16_index2, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + x1 * 2), z0 = svld3q (p0, x0 + x1 * 2)) /* ** ld3q_f16_index4: ** add (x[0-9]), x0, x1, lsl #?3 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_f16_index4, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + x1 * 4), z0 = svld3q (p0, x0 + x1 * 4)) /* ** ld3q_f16_index8: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, x1, lsl #?4\] ** ret */ TEST_LOAD (ld3q_f16_index8, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + x1 * 8), z0 = svld3q (p0, x0 + x1 * 8)) /* Moving the constant into a register would also be OK. */ /* ** ld3q_f16_1: ** incb x0 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_f16_1, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + svcnth ()), z0 = svld3q (p0, x0 + svcnth ())) /* Moving the constant into a register would also be OK. */ /* ** ld3q_f16_2: ** incb x0, all, mul #2 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_f16_2, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + svcnth () * 2), z0 = svld3q (p0, x0 + svcnth () * 2)) /* ** ld3q_f16_3: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #3, mul vl\] ** ret */ TEST_LOAD (ld3q_f16_3, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + svcnth () * 3), z0 = svld3q (p0, x0 + svcnth () * 3)) /* ** ld3q_f16_21: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #21, mul vl\] ** ret */ TEST_LOAD (ld3q_f16_21, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + svcnth () * 21), z0 = svld3q (p0, x0 + svcnth () * 21)) /* ** ld3q_f16_24: ** addvl (x[0-9]+), x0, #24 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_f16_24, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 + svcnth () * 24), z0 = svld3q (p0, x0 + svcnth () * 24)) /* Moving the constant into a register would also be OK. */ /* ** ld3q_f16_m1: ** decb x0 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_f16_m1, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 - svcnth ()), z0 = svld3q (p0, x0 - svcnth ())) /* Moving the constant into a register would also be OK. */ /* ** ld3q_f16_m2: ** decb x0, all, mul #2 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_f16_m2, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 - svcnth () * 2), z0 = svld3q (p0, x0 - svcnth () * 2)) /* ** ld3q_f16_m3: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #-3, mul vl\] ** ret */ TEST_LOAD (ld3q_f16_m3, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 - svcnth () * 3), z0 = svld3q (p0, x0 - svcnth () * 3)) /* ** ld3q_f16_m24: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #-24, mul vl\] ** ret */ TEST_LOAD (ld3q_f16_m24, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 - svcnth () * 24), z0 = svld3q (p0, x0 - svcnth () * 24)) /* ** ld3q_f16_m27: ** addvl (x[0-9]+), x0, #-27 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_f16_m27, svfloat16x3_t, float16_t, z0 = svld3q_f16 (p0, x0 - svcnth () * 27), z0 = svld3q (p0, x0 - svcnth () * 27)) /* ** ld3q_vnum_f16_0: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_vnum_f16_0, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, 0), z0 = svld3q_vnum (p0, x0, 0)) /* Moving the constant into a register would also be OK. */ /* ** ld3q_vnum_f16_1: ** incb x0 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_vnum_f16_1, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, 1), z0 = svld3q_vnum (p0, x0, 1)) /* Moving the constant into a register would also be OK. */ /* ** ld3q_vnum_f16_2: ** incb x0, all, mul #2 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_vnum_f16_2, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, 2), z0 = svld3q_vnum (p0, x0, 2)) /* ** ld3q_vnum_f16_3: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #3, mul vl\] ** ret */ TEST_LOAD (ld3q_vnum_f16_3, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, 3), z0 = svld3q_vnum (p0, x0, 3)) /* ** ld3q_vnum_f16_21: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #21, mul vl\] ** ret */ TEST_LOAD (ld3q_vnum_f16_21, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, 21), z0 = svld3q_vnum (p0, x0, 21)) /* ** ld3q_vnum_f16_24: ** addvl (x[0-9]+), x0, #24 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_vnum_f16_24, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, 24), z0 = svld3q_vnum (p0, x0, 24)) /* Moving the constant into a register would also be OK. */ /* ** ld3q_vnum_f16_m1: ** decb x0 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_vnum_f16_m1, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, -1), z0 = svld3q_vnum (p0, x0, -1)) /* Moving the constant into a register would also be OK. */ /* ** ld3q_vnum_f16_m2: ** decb x0, all, mul #2 ** ld3q {z0\.q - z2\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld3q_vnum_f16_m2, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, -2), z0 = svld3q_vnum (p0, x0, -2)) /* ** ld3q_vnum_f16_m3: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #-3, mul vl\] ** ret */ TEST_LOAD (ld3q_vnum_f16_m3, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, -3), z0 = svld3q_vnum (p0, x0, -3)) /* ** ld3q_vnum_f16_m24: ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, #-24, mul vl\] ** ret */ TEST_LOAD (ld3q_vnum_f16_m24, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, -24), z0 = svld3q_vnum (p0, x0, -24)) /* ** ld3q_vnum_f16_m27: ** addvl (x[0-9]+), x0, #-27 ** ld3q {z0\.q - z2\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld3q_vnum_f16_m27, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, -27), z0 = svld3q_vnum (p0, x0, -27)) /* ** ld3q_vnum_f16_x1: ** cntb (x[0-9]+) ** ( ** madd (x[0-9]+), (?:x1, \1|\1, x1), x0 ** ld3q {z0\.q - z2\.q}, p0/z, \[\2\] ** | ** mul (x[0-9]+), (?:x1, \1|\1, x1) ** ld3q {z0\.q - z2\.q}, p0/z, \[x0, \3\] ** ) ** ret */ TEST_LOAD (ld3q_vnum_f16_x1, svfloat16x3_t, float16_t, z0 = svld3q_vnum_f16 (p0, x0, x1), z0 = svld3q_vnum (p0, x0, x1))