/* { dg-do assemble { target aarch64_asm_sve2p1_ok } } */ /* { dg-do compile { target { ! aarch64_asm_sve2p1_ok } } } */ /* { dg-final { check-function-bodies "**" "" "-DCHECK_ASM" } } */ #include "test_sve_acle.h" #pragma GCC target "+sve2p1" #ifdef STREAMING_COMPATIBLE #pragma GCC target "+sme2" #endif /* ** bfmlslt_lane_0_f32_tied1: ** bfmlslt z0\.s, z4\.h, z5\.h\[0\] ** ret */ TEST_DUAL_Z (bfmlslt_lane_0_f32_tied1, svfloat32_t, svbfloat16_t, z0 = svbfmlslt_lane_f32 (z0, z4, z5, 0), z0 = svbfmlslt_lane (z0, z4, z5, 0)) /* ** bfmlslt_lane_0_f32_tied2: ** mov (z[0-9]+)\.d, z0\.d ** movprfx z0, z4 ** bfmlslt z0\.s, \1\.h, z1\.h\[0\] ** ret */ TEST_DUAL_Z_REV (bfmlslt_lane_0_f32_tied2, svfloat32_t, svbfloat16_t, z0_res = svbfmlslt_lane_f32 (z4, z0, z1, 0), z0_res = svbfmlslt_lane (z4, z0, z1, 0)) /* ** bfmlslt_lane_0_f32_tied3: ** mov (z[0-9]+)\.d, z0\.d ** movprfx z0, z4 ** bfmlslt z0\.s, z1\.h, \1\.h\[0\] ** ret */ TEST_DUAL_Z_REV (bfmlslt_lane_0_f32_tied3, svfloat32_t, svbfloat16_t, z0_res = svbfmlslt_lane_f32 (z4, z1, z0, 0), z0_res = svbfmlslt_lane (z4, z1, z0, 0)) /* ** bfmlslt_lane_0_f32_untied: ** movprfx z0, z1 ** bfmlslt z0\.s, z4\.h, z5\.h\[0\] ** ret */ TEST_DUAL_Z (bfmlslt_lane_0_f32_untied, svfloat32_t, svbfloat16_t, z0 = svbfmlslt_lane_f32 (z1, z4, z5, 0), z0 = svbfmlslt_lane (z1, z4, z5, 0)) /* ** bfmlslt_lane_1_f32: ** bfmlslt z0\.s, z4\.h, z5\.h\[1\] ** ret */ TEST_DUAL_Z (bfmlslt_lane_1_f32, svfloat32_t, svbfloat16_t, z0 = svbfmlslt_lane_f32 (z0, z4, z5, 1), z0 = svbfmlslt_lane (z0, z4, z5, 1)) /* ** bfmlslt_lane_7_f32: ** bfmlslt z0\.s, z4\.h, z5\.h\[7\] ** ret */ TEST_DUAL_Z (bfmlslt_lane_7_f32, svfloat32_t, svbfloat16_t, z0 = svbfmlslt_lane_f32 (z0, z4, z5, 7), z0 = svbfmlslt_lane (z0, z4, z5, 7)) /* ** bfmlslt_lane_z8_f32: ** str d8, \[sp, -16\]! ** mov (z[0-7])\.d, z8\.d ** bfmlslt z0\.s, z1\.h, \1\.h\[1\] ** ldr d8, \[sp\], 16 ** ret */ TEST_DUAL_LANE_REG (bfmlslt_lane_z8_f32, svfloat32_t, svbfloat16_t, z8, z0 = svbfmlslt_lane_f32 (z0, z1, z8, 1), z0 = svbfmlslt_lane (z0, z1, z8, 1)) /* ** bfmlslt_lane_z16_f32: ** mov (z[0-7])\.d, z16\.d ** bfmlslt z0\.s, z1\.h, \1\.h\[1\] ** ret */ TEST_DUAL_LANE_REG (bfmlslt_lane_z16_f32, svfloat32_t, svbfloat16_t, z16, z0 = svbfmlslt_lane_f32 (z0, z1, z16, 1), z0 = svbfmlslt_lane (z0, z1, z16, 1))