Update documentation for buffer fat pointers

[krzysz00]

Now that we've got (minus some issues around datatypes and invariant loads) working lowerings for address space 7, update the table in the AMDGPU usage guide to proerly indicate the nature of these address spaces.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Krzysztof Drewniak (krzysz00)

Changes

Now that we've got (minus some issues around datatypes and invariant loads) working lowerings for address space 7, update the table in the AMDGPU usage guide to proerly indicate the nature of these address spaces.

---

Full diff: https://github.com/llvm/llvm-project/pull/92034.diff

1 Files Affected:

• (modified) llvm/docs/AMDGPUUsage.rst (+2-2)

diff --git a/llvm/docs/AMDGPUUsage.rst b/llvm/docs/AMDGPUUsage.rst
index 029db00134c09..edf58b562a449 100644
--- a/llvm/docs/AMDGPUUsage.rst
+++ b/llvm/docs/AMDGPUUsage.rst
@@ -824,8 +824,8 @@ supported for the ``amdgcn`` target.
      Constant                              4               constant    *same as global* 64      0x0000000000000000
      Private                               5               private     scratch          32      0xFFFFFFFF
      Constant 32-bit                       6               *TODO*                               0x00000000
-     Buffer Fat Pointer (experimental)     7               *TODO*
-     Buffer Resource (experimental)        8               *TODO*
+     Buffer Fat Pointer                    7               N/A         N/A              160     0
+     Buffer Resource                       8               N/A         V#               128     0x00000000000000000000000000000000
      Buffer Strided Pointer (experimental) 9               *TODO*
      Streamout Registers                   128             N/A         GS_REGS
      ===================================== =============== =========== ================ ======= ============================

[krzysz00]

One thing I noticed while writing this: should the AMDGPU metadata streamer be updated so that ptr addrspace(8) inputs to kernels get the type "image" and not "global buffer"?

[arsenm]

One thing I noticed while writing this: should the AMDGPU metadata streamer be updated so that ptr addrspace(8) inputs to kernels get the type "image" and not "global buffer"?

I'm not sure what "image" is supposed to mean there, but an arbitrary resource descriptor isn't necessarily an image

[krzysz00]

@arsenm I meant this code right here https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/AMDGPU/AMDGPUHSAMetadataStreamer.cpp#L120-L136 - it seems like if you provide type hints to your kernel, various buffer descriptors become "image" in the metadata?