Re: [Qemu-devel] [PATCH 07/18] tcg: add vector addition operations

[Richard Henderson]

On 01/17/2017 01:07 AM, Kirill Batuzov wrote:
> +/***************************************/
> +/* 64-bit and 128-bit vector arithmetic.          */
> +
> +static inline void *tcg_v128_swap_slot(int n)
> +{
> +    return &tcg_ctx.v128_swap[n * 16];
> +}
> +
> +/* Find a memory location for 128-bit TCG variable. */
> +static inline void tcg_v128_to_ptr(TCGv_v128 tmp, TCGv_ptr base, int slot,
> +                                   TCGv_ptr *real_base, intptr_t *real_offset,
> +                                   int is_read)

None of this needs to be inline in tcg-op.h.  All of it should be out-of-line 
in tcg-op.c.


> @@ -750,6 +778,7 @@ struct TCGContext {
>      void *code_gen_buffer;
>      size_t code_gen_buffer_size;
>      void *code_gen_ptr;
> +    uint8_t v128_swap[16 * 3];

This is not thread-safe.
Shouldn't use space in TCGContext; should use space on stack.

Since there is no function call that is live, you can re-use the space for 
on-stack arguments.  There is TCG_STATIC_CALL_ARGS_SIZE (128) bytes allocated 
for that.  Which should be more than enough.


r~