Amdgpu backend fixes a missing comma in register class names.
A missing comma in the `REG_CLASS_NAMES` definition for the AMDgpu backend is now corrected, fixing minor display issues in dump files.
A small but impactful bug in the AMDgpu backend’s gcn.h file is fixed, restoring a missing comma in the REG_CLASS_NAMES macro. This correction resolves an issue primarily affecting the display of register class information in compiler dump files, ensuring proper formatting and readability for diagnostic output.
In Details
The REG_CLASS_NAMES macro within gcc/config/gcn/gcn.h for the AMDgpu GCN backend declares an array of string literals that represent the names of various register classes. A missing comma in this C array initializer meant that the subsequent string literal was effectively concatenated, leading to incorrect or malformed register class names in compiler diagnostic dumps. Though harmless to code generation, this fix ensures that internal compiler dumps and debugging outputs, which rely on these names, accurately reflect the AMDgpu's register allocation scheme, avoiding confusion for develope…
For Context
Compilers often have specific code that targets different computer processors, like the AMDgpu graphics cards. This specialized code needs to understand the unique features of each processor, including how it organizes its memory and internal storage units called 'registers'. To help with this, the compiler defines names for these different kinds of registers. Internally, the compiler uses these names for debugging and analysis, often printing them out in special 'dump files' that developers can read. A small mistake, like a missing comma in a list of these register names, can cause these internal diagnostic files to display incorrectly, making it harder for compiler developers to understand what's happening. This fix simply corrects that minor error, ensuring that the internal debugging information is always displayed properly.