/* { 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" /* ** ld4q_f16_base: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_base, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0), z0 = svld4q (p0, x0)) /* ** ld4q_f16_index: ** add (x[0-9]), x0, x1, lsl #?1 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld4q_f16_index, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + x1), z0 = svld4q (p0, x0 + x1)) /* ** ld4q_f16_index2: ** add (x[0-9]), x0, x1, lsl #?2 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld4q_f16_index2, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + x1 * 2), z0 = svld4q (p0, x0 + x1 * 2)) /* ** ld4q_f16_index4: ** add (x[0-9]), x0, x1, lsl #?3 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld4q_f16_index4, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + x1 * 4), z0 = svld4q (p0, x0 + x1 * 4)) /* ** ld4q_f16_index8: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, x1, lsl #?4\] ** ret */ TEST_LOAD (ld4q_f16_index8, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + x1 * 8), z0 = svld4q (p0, x0 + x1 * 8)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f16_1: ** incb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_1, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + svcnth ()), z0 = svld4q (p0, x0 + svcnth ())) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f16_2: ** incb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_2, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + svcnth () * 2), z0 = svld4q (p0, x0 + svcnth () * 2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f16_3: ** incb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_3, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + svcnth () * 3), z0 = svld4q (p0, x0 + svcnth () * 3)) /* ** ld4q_f16_4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #4, mul vl\] ** ret */ TEST_LOAD (ld4q_f16_4, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + svcnth () * 4), z0 = svld4q (p0, x0 + svcnth () * 4)) /* ** ld4q_f16_28: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #28, mul vl\] ** ret */ TEST_LOAD (ld4q_f16_28, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + svcnth () * 28), z0 = svld4q (p0, x0 + svcnth () * 28)) /* ** ld4q_f16_32: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_f16_32, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 + svcnth () * 32), z0 = svld4q (p0, x0 + svcnth () * 32)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f16_m1: ** decb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_m1, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 - svcnth ()), z0 = svld4q (p0, x0 - svcnth ())) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f16_m2: ** decb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_m2, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 - svcnth () * 2), z0 = svld4q (p0, x0 - svcnth () * 2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f16_m3: ** decb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f16_m3, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 - svcnth () * 3), z0 = svld4q (p0, x0 - svcnth () * 3)) /* ** ld4q_f16_m4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-4, mul vl\] ** ret */ TEST_LOAD (ld4q_f16_m4, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 - svcnth () * 4), z0 = svld4q (p0, x0 - svcnth () * 4)) /* ** ld4q_f16_m32: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-32, mul vl\] ** ret */ TEST_LOAD (ld4q_f16_m32, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 - svcnth () * 32), z0 = svld4q (p0, x0 - svcnth () * 32)) /* ** ld4q_f16_m36: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_f16_m36, svfloat16x4_t, float16_t, z0 = svld4q_f16 (p0, x0 - svcnth () * 36), z0 = svld4q (p0, x0 - svcnth () * 36)) /* ** ld4q_vnum_f16_0: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_0, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 0), z0 = svld4q_vnum (p0, x0, 0)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f16_1: ** incb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_1, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 1), z0 = svld4q_vnum (p0, x0, 1)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f16_2: ** incb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_2, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 2), z0 = svld4q_vnum (p0, x0, 2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f16_3: ** incb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_3, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 3), z0 = svld4q_vnum (p0, x0, 3)) /* ** ld4q_vnum_f16_4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #4, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f16_4, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 4), z0 = svld4q_vnum (p0, x0, 4)) /* ** ld4q_vnum_f16_28: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #28, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f16_28, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 28), z0 = svld4q_vnum (p0, x0, 28)) /* ** ld4q_vnum_f16_32: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_vnum_f16_32, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, 32), z0 = svld4q_vnum (p0, x0, 32)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f16_m1: ** decb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_m1, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, -1), z0 = svld4q_vnum (p0, x0, -1)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f16_m2: ** decb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_m2, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, -2), z0 = svld4q_vnum (p0, x0, -2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f16_m3: ** decb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f16_m3, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, -3), z0 = svld4q_vnum (p0, x0, -3)) /* ** ld4q_vnum_f16_m4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-4, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f16_m4, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, -4), z0 = svld4q_vnum (p0, x0, -4)) /* ** ld4q_vnum_f16_m32: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-32, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f16_m32, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, -32), z0 = svld4q_vnum (p0, x0, -32)) /* ** ld4q_vnum_f16_m36: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_vnum_f16_m36, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, -36), z0 = svld4q_vnum (p0, x0, -36)) /* ** ld4q_vnum_f16_x1: ** cntb (x[0-9]+) ** ( ** madd (x[0-9]+), (?:x1, \1|\1, x1), x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[\2\] ** | ** mul (x[0-9]+), (?:x1, \1|\1, x1) ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, \3\] ** ) ** ret */ TEST_LOAD (ld4q_vnum_f16_x1, svfloat16x4_t, float16_t, z0 = svld4q_vnum_f16 (p0, x0, x1), z0 = svld4q_vnum (p0, x0, x1))