Lines Matching refs:precise
532 /* correct NaN behaviour with precise */
542 //~f(16|32)! v1: (precise)%res8_tmp = @min @ub, %a
1208 /* precise arithmetic */
1210 //~gfx9! v1: (precise)%res0 = v_fma_f32 %a, %a, %res0_cvt
1211 //~gfx10! v1: (precise)%res0 = v_fma_mix_f32 %a, %a, lo(%a16)
1213 writeout(0, fma(a, a, f2f32(a16), bld.precise()));
1216 //! v2b: (precise)%res1 = v_mul_f16 %a16, %res1_cvt
1218 writeout(1, fmul(a16, f2f16(a), bld.precise()));
1221 //! v2b: (precise)%res2 = v_add_f16 %a16, %res2_cvt
1223 writeout(2, fadd(a16, f2f16(a), bld.precise()));
1226 //! v2b: (precise)%res3 = v_fma_f16 %a16, %a16, %res3_cvt
1228 writeout(3, fma(a16, a16, f2f16(a), bld.precise()));
1230 /* precise conversions */
1231 //! v2b: (precise)%res4_cvt = v_cvt_f16_f32 %a
1234 writeout(4, fmul(a16, f2f16(a, bld.precise())));
1236 //! v2b: (precise)%res5_cvt = v_cvt_f16_f32 %a
1239 writeout(5, fadd(a16, f2f16(a, bld.precise())));
1241 //! v2b: (precise)%res6_cvt = v_cvt_f16_f32 %a
1244 writeout(6, fma(a16, a16, f2f16(a, bld.precise())));
1398 //! v1: (precise)%res0 = v_cvt_f32_f16 %res0_tmp
1400 writeout(0, f2f32(fmul(a16, a16), bld.precise()));
1402 //! v2b: (precise)%res1_tmp = v_mul_f16 %a16, %a16
1405 writeout(1, f2f32(fmul(a16, a16, bld.precise())));
1528 /* the optimization is precise for 32-bit on GFX9 */
1530 //~gfx10! v1: (precise)%res0_tmp = v_fma_mix_f32 lo(%a16), %b, -0
1533 writeout(0, fadd(fmul(f2f32(a16), b, bld.precise()), c));
1535 //~gfx9! v1: (precise)%res1 = v_fma_mix_f32 lo(%a16), %b, %c
1537 //~gfx10! v1: (precise)%res1 = v_add_f32 %res1_tmp, %c
1539 writeout(1, fadd(fmul(f2f32(a16), b), c, bld.precise()));