til: add til_fragmenter_slice_per_cpu_x16() variant

This restores the original til_fragmenter_slice_per_cpu() while
adding an explicit x16 variant for what
til_fragmenter_slice_per_cpu() had become.

The impetus for this is realizing the x16 multiplier is terrible
for sparkler's crappy threading, and it really need strictly
n_cpus slices for the threading to be beneficial.

So I'm just getting rid of the hidden x16 in favor of making it
explicit in an _x16 variant.

Subsequent commit will pivot all the non-sparkler callers to
til_fragmenter_slice_per_cpu_x16().

1 files changed, 8 insertions, 1 deletions

diff --git a/src/til.c b/src/til.c
index 5d9a91f..0aa8efd 100644
--- a/src/til.c
+++ b/src/til.c
@@ -795,9 +795,16 @@ int til_module_setup_finalize(const til_module_t *module, const til_settings_t *
 }
 
 
-/* generic fragmenter using a horizontal slice per cpu according to context->n_cpus (multiplied by a constant factor) */
+/* generic fragmenter using a horizontal slice per cpu according to context->n_cpus */
 int til_fragmenter_slice_per_cpu(til_module_context_t *context, const til_fb_fragment_t *fragment, unsigned number, til_fb_fragment_t *res_fragment)
 {
+	return til_fb_fragment_slice_single(fragment, context->n_cpus, number, res_fragment);
+}
+
+
+/* generic fragmenter using a horizontal slice per cpu according to context->n_cpus * 16 */
+int til_fragmenter_slice_per_cpu_x16(til_module_context_t *context, const til_fb_fragment_t *fragment, unsigned number, til_fb_fragment_t *res_fragment)
+{
 	/* The *16 is to combat leaving CPUs idle waiting for others to finish their work.
 	 *
 	 * Even though there's some overhead in scheduling smaller work units,