From c4cceb425c2788243f0330f2a73e4368b1ed5346 Mon Sep 17 00:00:00 2001
From: Jason Volk <jason@zemos.net>
Date: Mon, 3 Oct 2022 22:14:23 +0000
Subject: [PATCH] ircd::gpt::gpu: Use explicit broadcast for local access.

---
 ircd/gpt_gpu.cl | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/ircd/gpt_gpu.cl b/ircd/gpt_gpu.cl
index 4c3d0d468..95013d0c2 100644
--- a/ircd/gpt_gpu.cl
+++ b/ircd/gpt_gpu.cl
@@ -742,11 +742,13 @@ ircd_gpt_lm_logsm(__global struct ircd_gpt_ctrl *const ctrl,
 	if(li == 0)
 		samax.mu = mu[li];
 
+	ircd_simt_broadcast_flldr(mu, ln, li);
+
 	sum[li] = 0.0f;
 	for(uint ti = start; ti < stop; ++ti)
 	{
 		const float
-		sub = logit[ti] - samax.mu,
+		sub = logit[ti] - mu[li],
 		res = native_exp(sub);
 
 		sum[li] += res;
@@ -764,7 +766,7 @@ ircd_gpt_lm_logsm(__global struct ircd_gpt_ctrl *const ctrl,
 	for(uint ti = start; ti < stop; ++ti)
 	{
 		const float
-		sub = logit[ti] - samax.mu,
+		sub = logit[ti] - mu[li],
 		res = lambda[li] * native_exp(sub);
 
 		logit[ti] = res;