<p dir="ltr"><br>
On Jun 9, 2016 4:10 AM, "Kenneth Graunke" <<a href="mailto:kenneth@whitecape.org">kenneth@whitecape.org</a>> wrote:<br>
><br>
> Skylake changes the representation of shared local memory size:<br>
><br>
> Size | 0 kB | 1 kB | 2 kB | 4 kB | 8 kB | 16 kB | 32 kB | 64 kB |<br>
> -------------------------------------------------------------------<br>
> Gen7-8 | 0 | none | none | 1 | 2 | 3 | 4 | 5 |<br>
> -------------------------------------------------------------------<br>
> Gen9+ | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 |<br>
><br>
> The old formula would substantially underallocate the amount of space.<br>
> This fixes GPU hangs on Skylake when running with full thread counts.<br>
><br>
> Cc: "12.0" <<a href="mailto:mesa-stable@lists.freedesktop.org">mesa-stable@lists.freedesktop.org</a>><br>
> Signed-off-by: Kenneth Graunke <<a href="mailto:kenneth@whitecape.org">kenneth@whitecape.org</a>><br>
> ---<br>
> src/mesa/drivers/dri/i965/gen7_cs_state.c | 15 ++++++++++-----<br>
> 1 file changed, 10 insertions(+), 5 deletions(-)<br>
><br>
> diff --git a/src/mesa/drivers/dri/i965/gen7_cs_state.c b/src/mesa/drivers/dri/i965/gen7_cs_state.c<br>
> index 750aa2c..aff1f4e 100644<br>
> --- a/src/mesa/drivers/dri/i965/gen7_cs_state.c<br>
> +++ b/src/mesa/drivers/dri/i965/gen7_cs_state.c<br>
> @@ -150,11 +150,16 @@ brw_upload_cs_state(struct brw_context *brw)<br>
> assert(prog_data->total_shared <= 64 * 1024);<br>
> uint32_t slm_size = 0;<br>
> if (prog_data->total_shared > 0) {<br>
> - /* slm_size is in 4k increments, but must be a power of 2. */<br>
> - slm_size = 4 * 1024;<br>
> - while (slm_size < prog_data->total_shared)<br>
> - slm_size <<= 1;<br>
> - slm_size /= 4 * 1024;<br>
> + /* Shared Local Memory Size is specified as powers of two. */<br>
> + slm_size = util_next_power_of_two(prog_data->total_shared);<br>
> +<br>
> + if (brw->gen >= 9) {<br>
> + /* Use a minimum of 1kB; turn an exponent of 10 (1024 kB) into 1. */<br>
> + slm_size = ffs(MAX2(slm_size, 1024)) - 10;<br>
> + } else {<br>
> + /* Use a minimum of 4kB; convert to the pre-Gen9 representation. */<br>
> + slm_size = MAX2(slm_size, 4096) / 4096;</p>
<p dir="ltr">According to your chart, 16k should end up with 3, but this logic will produce 4. The old comment said it was in increments of 4k, so I'm guessing just the chart needs to be adjusted.</p>
<p dir="ltr">> + }<br>
> }<br>
><br>
> desc[dw++] =<br>
> --<br>
> 2.8.3<br>
><br>
> _______________________________________________<br>
> mesa-dev mailing list<br>
> <a href="mailto:mesa-dev@lists.freedesktop.org">mesa-dev@lists.freedesktop.org</a><br>
> <a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a><br>
</p>