/* { dg-do assemble { target aarch64_asm_sve2p1_ok } } */ /* { dg-do compile { target { ! aarch64_asm_sve2p1_ok } } } */ /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" { target { ! ilp32 } } } } */ #include "test_sve_acle.h" #pragma GCC target "+sve2p1" #ifdef STREAMING_COMPATIBLE #pragma GCC target "+sme2" #endif /* ** ld1_f16_base: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_base, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0), z0 = svld1_x4 (pn8, x0)) /* ** ld1_f16_index: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, x1, lsl #?1\] ** ret */ TEST_LOAD_COUNT (ld1_f16_index, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + x1), z0 = svld1_x4 (pn8, x0 + x1)) /* Moving the constant into a register would also be OK. */ /* ** ld1_f16_1: ** incb x0 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_1, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + svcnth ()), z0 = svld1_x4 (pn8, x0 + svcnth ())) /* Moving the constant into a register would also be OK. */ /* ** ld1_f16_2: ** incb x0, all, mul #2 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_2, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + svcnth () * 2), z0 = svld1_x4 (pn8, x0 + svcnth () * 2)) /* Moving the constant into a register would also be OK. */ /* ** ld1_f16_3: ** incb x0, all, mul #3 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_3, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + svcnth () * 3), z0 = svld1_x4 (pn8, x0 + svcnth () * 3)) /* ** ld1_f16_4: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #4, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_f16_4, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + svcnth () * 4), z0 = svld1_x4 (pn8, x0 + svcnth () * 4)) /* ** ld1_f16_28: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #28, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_f16_28, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + svcnth () * 28), z0 = svld1_x4 (pn8, x0 + svcnth () * 28)) /* ** ld1_f16_32: ** [^{]* ** ld1h {z0\.h - z3\.h}, pn8/z, \[x[0-9]+\] ** ret */ TEST_LOAD_COUNT (ld1_f16_32, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 + svcnth () * 32), z0 = svld1_x4 (pn8, x0 + svcnth () * 32)) /* Moving the constant into a register would also be OK. */ /* ** ld1_f16_m1: ** decb x0 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_m1, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 - svcnth ()), z0 = svld1_x4 (pn8, x0 - svcnth ())) /* Moving the constant into a register would also be OK. */ /* ** ld1_f16_m2: ** decb x0, all, mul #2 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_m2, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 - svcnth () * 2), z0 = svld1_x4 (pn8, x0 - svcnth () * 2)) /* Moving the constant into a register would also be OK. */ /* ** ld1_f16_m3: ** decb x0, all, mul #3 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_m3, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 - svcnth () * 3), z0 = svld1_x4 (pn8, x0 - svcnth () * 3)) /* ** ld1_f16_m4: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #-4, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_f16_m4, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 - svcnth () * 4), z0 = svld1_x4 (pn8, x0 - svcnth () * 4)) /* ** ld1_f16_m32: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #-32, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_f16_m32, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 - svcnth () * 32), z0 = svld1_x4 (pn8, x0 - svcnth () * 32)) /* ** ld1_f16_m36: ** [^{]* ** ld1h {z0\.h - z3\.h}, pn8/z, \[x[0-9]+\] ** ret */ TEST_LOAD_COUNT (ld1_f16_m36, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn8, x0 - svcnth () * 36), z0 = svld1_x4 (pn8, x0 - svcnth () * 36)) /* ** ld1_f16_z17: ** ld1h {z[^\n]+}, pn8/z, \[x0\] ** mov [^\n]+ ** mov [^\n]+ ** mov [^\n]+ ** mov [^\n]+ ** ret */ TEST_LOAD_COUNT (ld1_f16_z17, svfloat16x4_t, float16_t, z17 = svld1_f16_x4 (pn8, x0), z17 = svld1_x4 (pn8, x0)) /* ** ld1_f16_z22: ** ld1h {z[^\n]+}, pn8/z, \[x0\] ** mov [^\n]+ ** mov [^\n]+ ** mov [^\n]+ ** mov [^\n]+ ** ret */ TEST_LOAD_COUNT (ld1_f16_z22, svfloat16x4_t, float16_t, z22 = svld1_f16_x4 (pn8, x0), z22 = svld1_x4 (pn8, x0)) /* ** ld1_f16_z28: ** ld1h {z28\.h(?: - |, )z31\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_z28, svfloat16x4_t, float16_t, z28 = svld1_f16_x4 (pn8, x0), z28 = svld1_x4 (pn8, x0)) /* ** ld1_f16_pn0: ** mov p([89]|1[0-5])\.b, p0\.b ** ld1h {z0\.h(?: - |, )z3\.h}, pn\1/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_pn0, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn0, x0), z0 = svld1_x4 (pn0, x0)) /* ** ld1_f16_pn7: ** mov p([89]|1[0-5])\.b, p7\.b ** ld1h {z0\.h(?: - |, )z3\.h}, pn\1/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_pn7, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn7, x0), z0 = svld1_x4 (pn7, x0)) /* ** ld1_f16_pn15: ** ld1h {z0\.h(?: - |, )z3\.h}, pn15/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_f16_pn15, svfloat16x4_t, float16_t, z0 = svld1_f16_x4 (pn15, x0), z0 = svld1_x4 (pn15, x0)) /* ** ld1_vnum_f16_0: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_0, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 0), z0 = svld1_vnum_x4 (pn8, x0, 0)) /* Moving the constant into a register would also be OK. */ /* ** ld1_vnum_f16_1: ** incb x0 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_1, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 1), z0 = svld1_vnum_x4 (pn8, x0, 1)) /* Moving the constant into a register would also be OK. */ /* ** ld1_vnum_f16_2: ** incb x0, all, mul #2 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_2, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 2), z0 = svld1_vnum_x4 (pn8, x0, 2)) /* Moving the constant into a register would also be OK. */ /* ** ld1_vnum_f16_3: ** incb x0, all, mul #3 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_3, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 3), z0 = svld1_vnum_x4 (pn8, x0, 3)) /* ** ld1_vnum_f16_4: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #4, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_4, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 4), z0 = svld1_vnum_x4 (pn8, x0, 4)) /* ** ld1_vnum_f16_28: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #28, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_28, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 28), z0 = svld1_vnum_x4 (pn8, x0, 28)) /* ** ld1_vnum_f16_32: ** [^{]* ** ld1h {z0\.h - z3\.h}, pn8/z, \[x[0-9]+\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_32, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, 32), z0 = svld1_vnum_x4 (pn8, x0, 32)) /* Moving the constant into a register would also be OK. */ /* ** ld1_vnum_f16_m1: ** decb x0 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_m1, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, -1), z0 = svld1_vnum_x4 (pn8, x0, -1)) /* Moving the constant into a register would also be OK. */ /* ** ld1_vnum_f16_m2: ** decb x0, all, mul #2 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_m2, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, -2), z0 = svld1_vnum_x4 (pn8, x0, -2)) /* Moving the constant into a register would also be OK. */ /* ** ld1_vnum_f16_m3: ** decb x0, all, mul #3 ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_m3, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, -3), z0 = svld1_vnum_x4 (pn8, x0, -3)) /* ** ld1_vnum_f16_m4: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #-4, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_m4, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, -4), z0 = svld1_vnum_x4 (pn8, x0, -4)) /* ** ld1_vnum_f16_m32: ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, #-32, mul vl\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_m32, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, -32), z0 = svld1_vnum_x4 (pn8, x0, -32)) /* ** ld1_vnum_f16_m36: ** [^{]* ** ld1h {z0\.h - z3\.h}, pn8/z, \[x[0-9]+\] ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_m36, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, -36), z0 = svld1_vnum_x4 (pn8, x0, -36)) /* ** ld1_vnum_f16_x1: ** cntb (x[0-9]+) ** ( ** madd (x[0-9]+), (?:x1, \1|\1, x1), x0 ** ld1h {z0\.h - z3\.h}, pn8/z, \[\2\] ** | ** mul (x[0-9]+), (?:x1, \1|\1, x1) ** ld1h {z0\.h - z3\.h}, pn8/z, \[x0, \3\] ** ) ** ret */ TEST_LOAD_COUNT (ld1_vnum_f16_x1, svfloat16x4_t, float16_t, z0 = svld1_vnum_f16_x4 (pn8, x0, x1), z0 = svld1_vnum_x4 (pn8, x0, x1))