RE: [PATCH v11 12/15] target/hexagon: prepare frontend for parser

[Taylor Simpson]

> -----Original Message-----
> From: Anton Johansson <anjo@rev.ng>
> Sent: Thursday, August 4, 2022 5:56 AM
> To: qemu-devel@nongnu.org
> Cc: ale@rev.ng; anjo@rev.ng; babush@rev.ng; nizzo@rev.ng; Taylor Simpson
> <tsimpson@quicinc.com>; Brian Cain <bcain@quicinc.com>; Michael Lambert
> <mlambert@quicinc.com>; richard.henderson@linaro.org;
> alex.bennee@linaro.org
> Subject: [PATCH v11 12/15] target/hexagon: prepare frontend for parser
> 
> This patch adds the necessary changes to the Hexagon frontend, required by
> idef-parser to deal with a :mem_noshuf packet with a store in slot 1 and
> predicated load in slot 0.
> 
> Signed-off-by: Anton Johansson <anjo@rev.ng>
> ---
>  target/hexagon/cpu.h       |  8 ++++++++
>  target/hexagon/translate.c | 22 ++++++++++++++++++++++
> target/hexagon/translate.h |  2 ++
>  3 files changed, 32 insertions(+)
> 
> -git a/target/hexagon/translate.c b/target/hexagon/translate.c index
> e3e250fd4f..72bf8d591b 100644
> 494,6 +498,22 @@ void process_store(DisasContext *ctx, Packet *pkt, int
> slot_num)
> 
>  static void process_store_log(DisasContext *ctx, Packet *pkt)  {
> +    /*
> +     * Here we deal with the special case of a :mem_noshuf packet with a
> +     * predicated load in slot 0 with a store in slot 1. If the predicated
> +     * branch wasn't taken during packet execution, then store in slot 1
> +     * will not have been executed, corresponding to hex_did_store_s1 being 
> 0.
> +     * If this is the case, process the store here.
> +     */
> +    if (ctx->insn_is_noshuf_pload) {
> +        TCGLabel *l = gen_new_label();
> +        /* Reset s1_store_processed so process_store actually emits a store 
> */
> +        ctx->s1_store_processed = false;
> +        tcg_gen_brcondi_tl(TCG_COND_EQ, hex_did_s1_store, 1, l);
> +        process_store(ctx, pkt, 1);
> +        gen_set_label(l);
> +    }
> +

Let's follow the convention in CHECK_NOSHUF_PRED where we branch around the 
probe of the load when the predicate is false and always call process_store.  
Then, the mem_noshuf store will always have been processed - so we don't need 
the runtime state to tell us.

Thanks,
Taylor