Multi instruction pattern help

Project Infrastructure LLVM Dev List Archives
1

I am trying to get a multi instruction pattern to work and seem to be
running into trouble.

The problem itself is fairly simple. I need to go from 64bit floats to
32bit integers. As the backend doesn't support this natively but has a
way of converting it, I'd prefer to get this working via tablegen.

What I thought would work from the previous discussion is the following:

def : Pat<(fp_to_uint (f64 GPR:$src0)),

          (fp_to_uint (f32 (dp_to_fp (f64 GPR:$src0))))>;

Which when it runs across a 64bit float, it does a double to single
conversion, and then calls the 32bit float to int routine.

However, tablegen fails with the following error:

1>anonymous.2: (fp_to_uint:isInt GPR:f64:$src0)

1>f:\hq\main\sw\appeng\tools\hpc\opencl\llvm\win32\AMDIL\..\bin\Win32\De
bug\TableGen.exe: In anonymous.2: Could not infer all types in pattern!

Is it possible to get an example explaining what is going on and how the
patterns are used? Maybe show how you can go from three different
patterns to a single pattern using multi-instruction pattern.

Thanks

Micah Villmow

Systems Engineer

Advanced Technology & Performance

Advanced Micro Devices Inc.

4555 Great America Pkwy,

Santa Clara, CA. 95054

P: 408-572-6219

F: 408-572-6596

2

This is trying to tell you that it inferred that fp_to_uint returns an integer type, but it doesn’t know which one. Try disambiguating either the input or output with an explicit type, ilke (i32 (fp_to_uint (…

3

Chris,

Thanks for the help, this will help me with writing more patterns, but
I am still hitting another roadblock. I attempted what you suggested and
it fixed that issue, but then it started giving a warning that there is
an unknown node in the resulting pattern.

// unsigned int: f64->i32 ==> f64->f32 + f32->i32

def : Pat<(i32 (fp_to_uint (f64 GPR:$src0))),

          (i32 (fp_to_uint (f32 (dp_to_fp (f64 GPR:$src0)))))>;

1>Building AMDil.td instruction selector implementation with tblgen

1>(fp_to_uint:i32 (dp_to_fp:f32 GPR:f64:$src0))

1>f:\hq\main\sw\appeng\tools\hpc\opencl\llvm\win32\AMDIL\..\bin\Win32\De
bug\TableGen.exe: Unknown node in result pattern!

both fp_to_uint and dp_to_fp are used in other patterns, so I am sure
that the patterns separate from each other work and the above pattern
passes the instruction information stage, so not sure why the
instruction selector cannot parse it. Any help would be useful.

Thanks,

4

What are you trying to do here? Your pattern is transforming target independent nodes into other target independent nodes (fp_to_uint, etc)? That’s not going to work. Patterns are meant to transform to target instruction nodes.

Evan

5

Yes, I ran into that problem a couple of weeks ago. There are cases
where doing this kind of transformation is useful but tblgen doesn't support
it yet. Much of instcombine could be expressed with such patterns, for
example.

                                                  -Dave

6

Evan,

Thanks, I understand better now. This is what I eventually came up
with.

def : Pat<(i32 (fp_to_sint (f64 GPR:$src0))),

          (i32 (FTOI (f32 (DTOF (f64 GPR:$src0)))))>;

Micah