On 15/06/2022 11:27, Jan Beulich wrote:
> In order to re-use (also in subsequent patches) existing code and tables
> as much as possible, simply introduce a new boolean field in emulator
> state indicating whether an insn is one with a half-precision source.
> Everything else then follows "naturally".
This is slightly misleading. What you mean is it has an FP16 source, I
think?
There are non-FP16 instructions which have half-precision inputs, like
the F16C group.
> --- a/tools/tests/x86_emulator/x86-emulate.h
> +++ b/tools/tests/x86_emulator/x86-emulate.h
> @@ -182,6 +182,7 @@ void wrpkru(unsigned int val);
> #define cpu_has_avx512_4fmaps (cp.feat.avx512_4fmaps && xcr0_mask(0xe6))
> #define cpu_has_avx512_vp2intersect (cp.feat.avx512_vp2intersect &&
> xcr0_mask(0xe6))
> #define cpu_has_serialize cp.feat.serialize
> +#define cpu_has_avx512_fp16 (cp.feat.avx512_fp16 && xcr0_mask(0xe6))
> #define cpu_has_avx_vnni (cp.feat.avx_vnni && xcr0_mask(6))
> #define cpu_has_avx512_bf16 (cp.feat.avx512_bf16 && xcr0_mask(0xe6))
>
> --- a/xen/arch/x86/x86_emulate/decode.c
> +++ b/xen/arch/x86/x86_emulate/decode.c
> @@ -518,6 +518,7 @@ static const struct ext0f3a_table {
> [0x7a ... 0x7b] = { .simd_size = simd_scalar_opc, .four_op = 1 },
> [0x7c ... 0x7d] = { .simd_size = simd_packed_fp, .four_op = 1 },
> [0x7e ... 0x7f] = { .simd_size = simd_scalar_opc, .four_op = 1 },
> + [0xc2] = { .simd_size = simd_any_fp, .d8s = d8s_vl },
This is a new instruction, isn't it?
~Andrew
