From d9f8636c3d44a70f114d910fa31c15a25846e344 Mon Sep 17 00:00:00 2001
From: Zhang Zheng <32410583+ZzSean@users.noreply.github.com>
Date: Tue, 7 Jun 2022 10:30:03 +0800
Subject: [PATCH] Supoort more dimensions in forward fast layer_norm kernel
 (#43226)

---
 .../operators/fused/fused_layernorm_residual_dropout_bias.h   | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h b/paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h
index f72f73438c0..fc044e0bafa 100644
--- a/paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h
+++ b/paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h
@@ -481,10 +481,12 @@ void LaunchLayernormResidualDropoutBias(
   LAUNCH_FUSED_FAST_LN_KERNEL_BASE(1536); \
   LAUNCH_FUSED_FAST_LN_KERNEL_BASE(1792); \
   LAUNCH_FUSED_FAST_LN_KERNEL_BASE(2048); \
+  LAUNCH_FUSED_FAST_LN_KERNEL_BASE(3072); \
   LAUNCH_FUSED_FAST_LN_KERNEL_BASE(4096)
 
   bool can_call_fast_ln_kernel = false;
-  if (((cols >= 768 && cols <= 2048 && cols % 256 == 0) || cols == 4096) &&
+  if (((cols >= 768 && cols <= 2048 && cols % 256 == 0) || cols == 3072 ||
+       cols == 4096) &&
       scale != nullptr && layernorm_bias != nullptr) {
     can_call_fast_ln_kernel = true;
   }
-- 
GitLab