[llvm] [NVPTX] Basic support for fp128 as a storage type (PR #136006)

[Artem Belevich via llvm-commits]

================
@@ -4,12 +4,15 @@
 target datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
 target triple = "nvptx64-nvidia-cuda"
 
-; Check that we can handle global variables of large integer type.
+; Check that we can handle global variables of large integer and fp128 type.
 
 ; (lsb) 0x0102'0304'0506...0F10 (msb)
 @gv = addrspace(1) externally_initialized global i128 21345817372864405881847059188222722561, align 16
 ; CHECK: .visible .global .align 16 .b8 gv[16] = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16};
 
+ at gv_fp128 = addrspace(1) externally_initialized global fp128 0xL33333333333333334004033333333333, align 16
----------------
Artem-B wrote:

It would be good to make the constant composed of bytes with sequential values, so we can see that all the bytes laded in the correct order.

https://github.com/llvm/llvm-project/pull/136006