From e8b5f2040d8a3eb3f021d296b82856ae8ce028d3 Mon Sep 17 00:00:00 2001
From: nimlgen <138685161+nimlgen@users.noreply.github.com>
Date: Sun, 2 Jun 2024 21:47:24 +0300
Subject: [PATCH] nv faster signal on dma queue (#4789)

---
 tinygrad/runtime/ops_nv.py | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/tinygrad/runtime/ops_nv.py b/tinygrad/runtime/ops_nv.py
index f936412db9..31bfd9dab4 100644
--- a/tinygrad/runtime/ops_nv.py
+++ b/tinygrad/runtime/ops_nv.py
@@ -186,6 +186,12 @@ class HWCopyQueue(HWQueue):
     self.next_cmd_index += 1
     return self
 
+  def signal(self, signal, value=0):
+    self.q += [nvmethod(4, nv_gpu.NVC6B5_SET_SEMAPHORE_A, 4), *nvdata64(ctypes.addressof(from_mv(signal))), value, 4]
+    self.q += [nvmethod(4, nv_gpu.NVC6B5_LAUNCH_DMA, 1), 0x14]
+    self.next_cmd_index += 1
+    return self
+
   def submit(self, dev:NVDevice):
     if len(self.q) == 0: return
     dev.dma_put_value = self._submit(dev, dev.dma_gpu_ring, dev.dma_put_value, dev.dma_gpfifo_entries,