/* { 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_f32_base: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_base, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0), z0 = svld4q (p0, x0)) /* ** ld4q_f32_index: ** add (x[0-9]), x0, x1, lsl #?2 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld4q_f32_index, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + x1), z0 = svld4q (p0, x0 + x1)) /* ** ld4q_f32_index2: ** add (x[0-9]), x0, x1, lsl #?3 ** ld4q {z0\.q - z3\.q}, p0/z, \[\1\] ** ret */ TEST_LOAD (ld4q_f32_index2, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + x1 * 2), z0 = svld4q (p0, x0 + x1 * 2)) /* ** ld4q_f32_index4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, x1, lsl #?4\] ** ret */ TEST_LOAD (ld4q_f32_index4, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + x1 * 4), z0 = svld4q (p0, x0 + x1 * 4)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f32_1: ** incb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_1, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + svcntw ()), z0 = svld4q (p0, x0 + svcntw ())) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f32_2: ** incb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_2, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + svcntw () * 2), z0 = svld4q (p0, x0 + svcntw () * 2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f32_3: ** incb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_3, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + svcntw () * 3), z0 = svld4q (p0, x0 + svcntw () * 3)) /* ** ld4q_f32_4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #4, mul vl\] ** ret */ TEST_LOAD (ld4q_f32_4, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + svcntw () * 4), z0 = svld4q (p0, x0 + svcntw () * 4)) /* ** ld4q_f32_28: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #28, mul vl\] ** ret */ TEST_LOAD (ld4q_f32_28, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + svcntw () * 28), z0 = svld4q (p0, x0 + svcntw () * 28)) /* ** ld4q_f32_32: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_f32_32, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 + svcntw () * 32), z0 = svld4q (p0, x0 + svcntw () * 32)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f32_m1: ** decb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_m1, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 - svcntw ()), z0 = svld4q (p0, x0 - svcntw ())) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f32_m2: ** decb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_m2, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 - svcntw () * 2), z0 = svld4q (p0, x0 - svcntw () * 2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_f32_m3: ** decb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_f32_m3, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 - svcntw () * 3), z0 = svld4q (p0, x0 - svcntw () * 3)) /* ** ld4q_f32_m4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-4, mul vl\] ** ret */ TEST_LOAD (ld4q_f32_m4, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 - svcntw () * 4), z0 = svld4q (p0, x0 - svcntw () * 4)) /* ** ld4q_f32_m32: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-32, mul vl\] ** ret */ TEST_LOAD (ld4q_f32_m32, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 - svcntw () * 32), z0 = svld4q (p0, x0 - svcntw () * 32)) /* ** ld4q_f32_m36: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_f32_m36, svfloat32x4_t, float32_t, z0 = svld4q_f32 (p0, x0 - svcntw () * 36), z0 = svld4q (p0, x0 - svcntw () * 36)) /* ** ld4q_vnum_f32_0: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_0, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 0), z0 = svld4q_vnum (p0, x0, 0)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f32_1: ** incb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_1, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 1), z0 = svld4q_vnum (p0, x0, 1)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f32_2: ** incb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_2, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 2), z0 = svld4q_vnum (p0, x0, 2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f32_3: ** incb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_3, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 3), z0 = svld4q_vnum (p0, x0, 3)) /* ** ld4q_vnum_f32_4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #4, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f32_4, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 4), z0 = svld4q_vnum (p0, x0, 4)) /* ** ld4q_vnum_f32_28: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #28, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f32_28, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 28), z0 = svld4q_vnum (p0, x0, 28)) /* ** ld4q_vnum_f32_32: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_vnum_f32_32, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, 32), z0 = svld4q_vnum (p0, x0, 32)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f32_m1: ** decb x0 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_m1, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, -1), z0 = svld4q_vnum (p0, x0, -1)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f32_m2: ** decb x0, all, mul #2 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_m2, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, -2), z0 = svld4q_vnum (p0, x0, -2)) /* Moving the constant into a register would also be OK. */ /* ** ld4q_vnum_f32_m3: ** decb x0, all, mul #3 ** ld4q {z0\.q - z3\.q}, p0/z, \[x0\] ** ret */ TEST_LOAD (ld4q_vnum_f32_m3, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, -3), z0 = svld4q_vnum (p0, x0, -3)) /* ** ld4q_vnum_f32_m4: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-4, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f32_m4, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, -4), z0 = svld4q_vnum (p0, x0, -4)) /* ** ld4q_vnum_f32_m32: ** ld4q {z0\.q - z3\.q}, p0/z, \[x0, #-32, mul vl\] ** ret */ TEST_LOAD (ld4q_vnum_f32_m32, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, -32), z0 = svld4q_vnum (p0, x0, -32)) /* ** ld4q_vnum_f32_m36: ** [^{]* ** ld4q {z0\.q - z3\.q}, p0/z, \[x[0-9]+\] ** ret */ TEST_LOAD (ld4q_vnum_f32_m36, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, -36), z0 = svld4q_vnum (p0, x0, -36)) /* ** ld4q_vnum_f32_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_f32_x1, svfloat32x4_t, float32_t, z0 = svld4q_vnum_f32 (p0, x0, x1), z0 = svld4q_vnum (p0, x0, x1))