1876127044972728402, 1876147238508138659

2026-08-25 21:59:10 -04:00
1 parent baf931947b
commit c2f0e52b35
@@ -0,0 +1,37 @@
+---
+title: "After a few hours of trying workaround and seeing a whole lot of perf bugs in the process."
+type: archive
+source: twitter
+source_url: "https://x.com/NOTimothyLottes/status/1876127044972728402"
+author: "NOTimothyLottes"
+handle: NOTimothyLottes
+post_id: "1876127044972728402"
+date: 2025-01-06
+archived: 2026-08-25
+draft: false
+tags:
+  - archive
+  - twitter
+  - NOTimothyLottes
+description: "After a few hours of trying workaround and seeing a whole lot of perf bugs in the process."
+in_reply_to: ""
+parent_post_id: "1876120138283729159"
+---
+
+## Source
+
+- URL: https://x.com/NOTimothyLottes/status/1876127044972728402
+- Author: NOTimothyLottes (@NOTimothyLottes)
+- Posted: 2025-01-06 04:41:58
+
+## Branch
+
+**1/** **@NOTimothyLottes** ^1876127044972728402
+
+After a few hours of trying workaround and seeing a whole lot of perf bugs in the process. The irony is this is just a trivial HEX dump print shader. And if I can find this many problems in just one shader, image how many you'd find in a whole application!
+
+![](https://pbs.twimg.com/media/GglXeRrXsAAU2aW?format=png&name=orig)
+
+## Related
+
+- Spine: [[archive/threads/NOTimothyLottes/2025-01-06-one-of-the-downsides-of-v-cndmask-b32-on-amd-is]]
@@ -0,0 +1,43 @@
+---
+title: "@NOTimothyLottes GLSL bitfieldInsert does something else?"
+type: archive
+source: twitter
+source_url: "https://x.com/axelgneiting/status/1876147238508138659"
+author: "Axel Gneiting"
+handle: axelgneiting
+post_id: "1876147238508138659"
+date: 2025-01-06
+archived: 2026-08-25
+draft: false
+tags:
+  - archive
+  - twitter
+  - NOTimothyLottes
+description: "@NOTimothyLottes GLSL bitfieldInsert does something else?"
+in_reply_to: ""
+parent_post_id: "1876112593510949253"
+---
+
+## Source
+
+- URL: https://x.com/axelgneiting/status/1876147238508138659
+- Author: Axel Gneiting (@axelgneiting)
+- Posted: 2025-01-06 06:02:13
+
+## Branch
+
+**1/** **@axelgneiting** ^1876147238508138659
+
+**@NOTimothyLottes**
+
+GLSL bitfieldInsert does something else?
+
+**2/** **@NOTimothyLottes** ^1876152977704247326
+
+**@axelgneiting**
+
+Yeah the GLSL bitfieldInsert() and V_BFI_B32 are different. The HW instruction supports an arbitrary bitmask instead of one generated from just {offset, size}
+
+## Related
+
+- Spine: [[archive/threads/NOTimothyLottes/2025-01-06-one-of-the-downsides-of-v-cndmask-b32-on-amd-is]]
@@ -50,12 +50,16 @@ One of my favorite AMD instructions V_BFI_B32 is unfortunately one that no intri
 ![](https://pbs.twimg.com/media/GglKA6nW4AEtOIC?format=png&name=orig)
 ![](https://pbs.twimg.com/media/GglKXy4XcAAA_Og?format=png&name=orig)

+Branches: [[archive/threads/NOTimothyLottes/2025-01-06-one-of-the-downsides-of-v-cndmask-b32-on-amd-is/2025-01-06-axelgneiting-glsl-bitfieldinsert-does-something-else]]
+
 **5/** **@NOTimothyLottes** ^1876120138283729159

 Just looking at the next 5 lines of disassembly shows another 2 perf bugs: the compiler transforms 3 operations into 5 operations because it cannot handle mixed packed and unpacked stuff. The 2 V_ANDs can be merged, and the LSHL and CVT should just be one op

 ![](https://pbs.twimg.com/media/GglQsALXYAAEfWI?format=png&name=orig)

+Branches: [[archive/threads/NOTimothyLottes/2025-01-06-one-of-the-downsides-of-v-cndmask-b32-on-amd-is/2025-01-06-NOTimothyLottes-after-a-few-hours-of-trying-workaround-and-seeing]]
+
 **6/** **@NOTimothyLottes** ^1876126316816380333

 No way to work around those problems. Looks like bitfieldExtract 'v4' back-propagates something where the compiler ignores the .y in the packed logic, and then makes a mess of things scalarizing it. BFE only uses the 5-bit LSB of 'v4', so it's safe to leave junk in the other bits