/* { dg-do assemble { target aarch64_asm_sve-b16b16_ok } } */ /* { dg-do compile { target { ! aarch64_asm_sve-b16b16_ok } } } */ /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */ #include "test_sve_acle.h" #pragma GCC target "+sve-b16b16" #ifdef STREAMING_COMPATIBLE #pragma GCC target "+sme2" #endif /* ** mul_bf16_m_tied1: ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_m_tied1, svbfloat16_t, z0 = svmul_bf16_m (p0, z0, z1), z0 = svmul_m (p0, z0, z1)) /* ** mul_bf16_m_tied2: ** mov (z[0-9]+)\.d, z0\.d ** movprfx z0, z1 ** bfmul z0\.h, p0/m, z0\.h, \1\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_m_tied2, svbfloat16_t, z0 = svmul_bf16_m (p0, z1, z0), z0 = svmul_m (p0, z1, z0)) /* ** mul_bf16_m_untied: ** movprfx z0, z1 ** bfmul z0\.h, p0/m, z0\.h, z2\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_m_untied, svbfloat16_t, z0 = svmul_bf16_m (p0, z1, z2), z0 = svmul_m (p0, z1, z2)) /* ** mul_h4_bf16_m_tied1: ** mov (z[0-9]+\.h), h4 ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_ZD (mul_h4_bf16_m_tied1, svbfloat16_t, __bf16, z0 = svmul_n_bf16_m (p0, z0, d4), z0 = svmul_m (p0, z0, d4)) /* ** mul_h4_bf16_m_untied: ** mov (z[0-9]+\.h), h4 ** movprfx z0, z1 ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_ZD (mul_h4_bf16_m_untied, svbfloat16_t, __bf16, z0 = svmul_n_bf16_m (p0, z1, d4), z0 = svmul_m (p0, z1, d4)) /* ** mul_1_bf16_m: ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_1_bf16_m, svbfloat16_t, z0 = svmul_n_bf16_m (p0, z0, 1), z0 = svmul_m (p0, z0, 1)) /* ** mul_0p5_bf16_m: ** fmov (z[0-9]+\.h), #1\.75(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_0p5_bf16_m, svbfloat16_t, z0 = svmul_n_bf16_m (p0, z0, (bfloat16_t) (0.5)), z0 = svmul_m (p0, z0, (bfloat16_t) (0.5))) /* ** mul_m1_bf16_m: ** fmov (z[0-9]+\.h), #-1\.875(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_m1_bf16_m, svbfloat16_t, z0 = svmul_n_bf16_m (p0, z0, -1), z0 = svmul_m (p0, z0, -1)) /* ** mul_m0p5_bf16_m: ** fmov (z[0-9]+\.h), #-1\.75(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_m0p5_bf16_m, svbfloat16_t, z0 = svmul_n_bf16_m (p0, z0, (bfloat16_t) (-0.5)), z0 = svmul_m (p0, z0, (bfloat16_t) (-0.5))) /* ** mul_m2_bf16_m_tied1: ** fmov (z[0-9]+\.h), #-2\.0(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_m2_bf16_m_tied1, svbfloat16_t, z0 = svmul_n_bf16_m (p0, z0, -2), z0 = svmul_m (p0, z0, -2)) /* ** mul_m2_bf16_m_untied: ** fmov (z[0-9]+\.h), #-2\.0(?:e\+0)? ** movprfx z0, z1 ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_m2_bf16_m_untied, svbfloat16_t, z0 = svmul_n_bf16_m (p0, z1, -2), z0 = svmul_m (p0, z1, -2)) /* ** mul_bf16_z_tied1: ** movprfx z0\.h, p0/z, z0\.h ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_z_tied1, svbfloat16_t, z0 = svmul_bf16_z (p0, z0, z1), z0 = svmul_z (p0, z0, z1)) /* ** mul_bf16_z_tied2: ** movprfx z0\.h, p0/z, z0\.h ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_z_tied2, svbfloat16_t, z0 = svmul_bf16_z (p0, z1, z0), z0 = svmul_z (p0, z1, z0)) /* ** mul_bf16_z_untied: ** ( ** movprfx z0\.h, p0/z, z1\.h ** bfmul z0\.h, p0/m, z0\.h, z2\.h ** | ** movprfx z0\.h, p0/z, z2\.h ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ) ** ret */ TEST_UNIFORM_Z (mul_bf16_z_untied, svbfloat16_t, z0 = svmul_bf16_z (p0, z1, z2), z0 = svmul_z (p0, z1, z2)) /* ** mul_h4_bf16_z_tied1: ** mov (z[0-9]+\.h), h4 ** movprfx z0\.h, p0/z, z0\.h ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_ZD (mul_h4_bf16_z_tied1, svbfloat16_t, __bf16, z0 = svmul_n_bf16_z (p0, z0, d4), z0 = svmul_z (p0, z0, d4)) /* ** mul_h4_bf16_z_untied: ** mov (z[0-9]+\.h), h4 ** ( ** movprfx z0\.h, p0/z, z1\.h ** bfmul z0\.h, p0/m, z0\.h, \1 ** | ** movprfx z0\.h, p0/z, \1 ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ) ** ret */ TEST_UNIFORM_ZD (mul_h4_bf16_z_untied, svbfloat16_t, __bf16, z0 = svmul_n_bf16_z (p0, z1, d4), z0 = svmul_z (p0, z1, d4)) /* ** mul_1_bf16_z: ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)? ** movprfx z0\.h, p0/z, z0\.h ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_1_bf16_z, svbfloat16_t, z0 = svmul_n_bf16_z (p0, z0, 1), z0 = svmul_z (p0, z0, 1)) /* ** mul_bf16_x_tied1: ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_x_tied1, svbfloat16_t, z0 = svmul_bf16_x (p0, z0, z1), z0 = svmul_x (p0, z0, z1)) /* ** mul_bf16_x_tied2: ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_Z (mul_bf16_x_tied2, svbfloat16_t, z0 = svmul_bf16_x (p0, z1, z0), z0 = svmul_x (p0, z1, z0)) /* ** mul_bf16_x_untied: ** ( ** movprfx z0, z1 ** bfmul z0\.h, p0/m, z0\.h, z2\.h ** | ** movprfx z0, z2 ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ) ** ret */ TEST_UNIFORM_Z (mul_bf16_x_untied, svbfloat16_t, z0 = svmul_bf16_x (p0, z1, z2), z0 = svmul_x (p0, z1, z2)) /* ** mul_h4_bf16_x_tied1: ** mov (z[0-9]+\.h), h4 ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_ZD (mul_h4_bf16_x_tied1, svbfloat16_t, __bf16, z0 = svmul_n_bf16_x (p0, z0, d4), z0 = svmul_x (p0, z0, d4)) /* ** mul_h4_bf16_x_untied: ** mov z0\.h, h4 ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_ZD (mul_h4_bf16_x_untied, svbfloat16_t, __bf16, z0 = svmul_n_bf16_x (p0, z1, d4), z0 = svmul_x (p0, z1, d4)) /* ** mul_1_bf16_x_tied1: ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, \1 ** ret */ TEST_UNIFORM_Z (mul_1_bf16_x_tied1, svbfloat16_t, z0 = svmul_n_bf16_x (p0, z0, 1), z0 = svmul_x (p0, z0, 1)) /* ** mul_1_bf16_x_untied: ** fmov z0\.h, #1\.875(?:e\+0)? ** bfmul z0\.h, p0/m, z0\.h, z1\.h ** ret */ TEST_UNIFORM_Z (mul_1_bf16_x_untied, svbfloat16_t, z0 = svmul_n_bf16_x (p0, z1, 1), z0 = svmul_x (p0, z1, 1)) /* ** ptrue_mul_bf16_x_tied1: ** bfmul z0\.h, (z0\.h, z1\.h|z1\.h, z0\.h) ** ret */ TEST_UNIFORM_Z (ptrue_mul_bf16_x_tied1, svbfloat16_t, z0 = svmul_bf16_x (svptrue_b16 (), z0, z1), z0 = svmul_x (svptrue_b16 (), z0, z1)) /* ** ptrue_mul_bf16_x_tied2: ** bfmul z0\.h, (z0\.h, z1\.h|z1\.h, z0\.h) ** ret */ TEST_UNIFORM_Z (ptrue_mul_bf16_x_tied2, svbfloat16_t, z0 = svmul_bf16_x (svptrue_b16 (), z1, z0), z0 = svmul_x (svptrue_b16 (), z1, z0)) /* ** ptrue_mul_bf16_x_untied: ** bfmul z0\.h, (z1\.h, z2\.h|z2\.h, z1\.h) ** ret */ TEST_UNIFORM_Z (ptrue_mul_bf16_x_untied, svbfloat16_t, z0 = svmul_bf16_x (svptrue_b16 (), z1, z2), z0 = svmul_x (svptrue_b16 (), z1, z2)) /* ** ptrue_mul_1_bf16_x_tied1: ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)? ** bfmul z0\.h, (z0\.h, \1|\1, z0\.h) ** ret */ TEST_UNIFORM_Z (ptrue_mul_1_bf16_x_tied1, svbfloat16_t, z0 = svmul_n_bf16_x (svptrue_b16 (), z0, 1), z0 = svmul_x (svptrue_b16 (), z0, 1)) /* ** ptrue_mul_1_bf16_x_untied: ** fmov (z[0-9]+\.h), #1\.875(?:e\+0)? ** bfmul z0\.h, (z1\.h, \1|\1, z1\.h) ** ret */ TEST_UNIFORM_Z (ptrue_mul_1_bf16_x_untied, svbfloat16_t, z0 = svmul_n_bf16_x (svptrue_b16 (), z1, 1), z0 = svmul_x (svptrue_b16 (), z1, 1))