[llvm] [NVPTX] Prevent fptrunc of v2f32 from being folded into store (PR #149571)

[LLVM Continuous Integration via llvm-commits]

llvm-ci wrote:

LLVM Buildbot has detected a new failure on builder `ml-opt-rel-x86-64` running on `ml-opt-rel-x86-64-b1` while building `llvm` at step 6 "test-build-unified-tree-check-all".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/185/builds/22430

<details>
<summary>Here is the relevant piece of the build log for the reference</summary>

```
Step 6 (test-build-unified-tree-check-all) failure: test (failure)
******************** TEST 'LLVM :: CodeGen/NVPTX/f16x2-instructions.ll' FAILED ********************
Exit Code: 1

Command Output (stderr):
--
/b/ml-opt-rel-x86-64-b1/build/bin/llc < /b/ml-opt-rel-x86-64-b1/llvm-project/llvm/test/CodeGen/NVPTX/f16x2-instructions.ll -mtriple=nvptx64-nvidia-cuda -mcpu=sm_53           -O0 -disable-post-ra -frame-pointer=all -verify-machineinstrs  | /b/ml-opt-rel-x86-64-b1/build/bin/FileCheck -allow-deprecated-dag-overlap -check-prefixes CHECK,CHECK-F16 /b/ml-opt-rel-x86-64-b1/llvm-project/llvm/test/CodeGen/NVPTX/f16x2-instructions.ll # RUN: at line 3
+ /b/ml-opt-rel-x86-64-b1/build/bin/llc -mtriple=nvptx64-nvidia-cuda -mcpu=sm_53 -O0 -disable-post-ra -frame-pointer=all -verify-machineinstrs
+ /b/ml-opt-rel-x86-64-b1/build/bin/FileCheck -allow-deprecated-dag-overlap -check-prefixes CHECK,CHECK-F16 /b/ml-opt-rel-x86-64-b1/llvm-project/llvm/test/CodeGen/NVPTX/f16x2-instructions.ll
/b/ml-opt-rel-x86-64-b1/llvm-project/llvm/test/CodeGen/NVPTX/f16x2-instructions.ll:1507:15: error: CHECK-NEXT: expected string not found in input
; CHECK-NEXT: ld.param.v2.b32 {%r1, %r2}, [test_fptrunc_2xfloat_param_0];
              ^
<stdin>:960:10: note: scanning from here
// %bb.0:
         ^
<stdin>:961:2: note: possible intended match here
 ld.param.b64 %rd1, [test_fptrunc_2xfloat_param_0];
 ^

Input file: <stdin>
Check file: /b/ml-opt-rel-x86-64-b1/llvm-project/llvm/test/CodeGen/NVPTX/f16x2-instructions.ll

-dump-input=help explains the following input dump.

Input was:
<<<<<<
             .
             .
             .
           955: { 
           956:  .reg .b16 %rs<3>; 
           957:  .reg .b32 %r<4>; 
           958:  .reg .b64 %rd<2>; 
           959:  
           960: // %bb.0: 
next:1507'0              X error: no match found
           961:  ld.param.b64 %rd1, [test_fptrunc_2xfloat_param_0]; 
next:1507'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
next:1507'1      ?                                                   possible intended match
           962:  mov.b64 {%r1, %r2}, %rd1; 
next:1507'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~
           963:  cvt.rn.f16.f32 %rs1, %r2; 
next:1507'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~
           964:  cvt.rn.f16.f32 %rs2, %r1; 
next:1507'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~
           965:  mov.b32 %r3, {%rs2, %rs1}; 
next:1507'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
           966:  st.param.b32 [func_retval0], %r3; 
next:1507'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
             .
             .
...

```

</details>

https://github.com/llvm/llvm-project/pull/149571