nir: Add AGX atomic intrinsics

This is a piece of cake with unified atomics :-) This will let us do our
addressing math tricks nice and easily.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/23529>

diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index 19f2443..84e1db9 100644 (file)
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -754,6 +754,9 @@ intrinsic("load_vulkan_descriptor", src_comp=[-1], dest_comp=0,
 #
 # IR3 global operations take 32b vec2 as memory address. IR3 doesn't support
 # float atomics.
+#
+# AGX global variants take a 64-bit base address plus a 32-bit offset in words.
+# The offset is sign-extended or zero-extended based on the SIGN_EXTEND index.
 
 intrinsic("deref_atomic",  src_comp=[-1, 1], dest_comp=1, indices=[ACCESS, ATOMIC_OP])
 intrinsic("ssbo_atomic",  src_comp=[-1, 1, 1], dest_comp=1, indices=[ACCESS, ATOMIC_OP])
@@ -763,6 +766,7 @@ intrinsic("global_atomic",  src_comp=[1, 1], dest_comp=1, indices=[ATOMIC_OP])
 intrinsic("global_atomic_2x32",  src_comp=[2, 1], dest_comp=1, indices=[ATOMIC_OP])
 intrinsic("global_atomic_amd",  src_comp=[1, 1, 1], dest_comp=1, indices=[BASE, ATOMIC_OP])
 intrinsic("global_atomic_ir3",  src_comp=[2, 1], dest_comp=1, indices=[BASE, ATOMIC_OP])
+intrinsic("global_atomic_agx",  src_comp=[1, 1, 1], dest_comp=1, indices=[ATOMIC_OP, SIGN_EXTEND])
 
 intrinsic("deref_atomic_swap",  src_comp=[-1, 1, 1], dest_comp=1, indices=[ACCESS, ATOMIC_OP])
 intrinsic("ssbo_atomic_swap",  src_comp=[-1, 1, 1, 1], dest_comp=1, indices=[ACCESS, ATOMIC_OP])
@@ -772,6 +776,7 @@ intrinsic("global_atomic_swap",  src_comp=[1, 1, 1], dest_comp=1, indices=[ATOMI
 intrinsic("global_atomic_swap_2x32",  src_comp=[2, 1, 1], dest_comp=1, indices=[ATOMIC_OP])
 intrinsic("global_atomic_swap_amd",  src_comp=[1, 1, 1, 1], dest_comp=1, indices=[BASE, ATOMIC_OP])
 intrinsic("global_atomic_swap_ir3",  src_comp=[2, 1, 1], dest_comp=1, indices=[BASE, ATOMIC_OP])
+intrinsic("global_atomic_swap_agx",  src_comp=[1, 1, 1, 1], dest_comp=1, indices=[ATOMIC_OP, SIGN_EXTEND])
 
 def system_value(name, dest_comp, indices=[], bit_sizes=[32]):
     intrinsic("load_" + name, [], dest_comp, indices,