summaryrefslogtreecommitdiffstats
path: root/ComputeShaders/mulMatByRowTiledEx.hlsl
diff options
context:
space:
mode:
authorKonstantin <const@const.me>2023-01-24 17:40:50 +0100
committerKonstantin <const@const.me>2023-01-24 17:40:50 +0100
commite78815de53005336c3746bcd85c2e789a43f1b94 (patch)
tree5fe25b6d4024bbcca7402ebcc28910417c54758d /ComputeShaders/mulMatByRowTiledEx.hlsl
parent9d6fb00973218ebc397e15819be3964423b81f91 (diff)
Performance tuning on AMD iGPU
Diffstat (limited to 'ComputeShaders/mulMatByRowTiledEx.hlsl')
-rw-r--r--ComputeShaders/mulMatByRowTiledEx.hlsl2
1 files changed, 1 insertions, 1 deletions
diff --git a/ComputeShaders/mulMatByRowTiledEx.hlsl b/ComputeShaders/mulMatByRowTiledEx.hlsl
index bc0c0c6..d286ec2 100644
--- a/ComputeShaders/mulMatByRowTiledEx.hlsl
+++ b/ComputeShaders/mulMatByRowTiledEx.hlsl
@@ -3,7 +3,7 @@
static const uint TILE_SIZE = 32;
#endif
#ifndef THREADS_Y
-static const uint THREADS_Y = 16;
+static const uint THREADS_Y = 8;
#endif
// First tensor, reshaped into dense column major horizontal panels of size [ width, TILE_SIZE ]