[AMDGPU] SIPeepholeSDWA: Add V_CNDMASK_B32 support

[frederik-h]

The isConvertibleToSDWA function form llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp contains this FIXME:

// FIXME: has SDWA but require handling of implicit VCC use
if (Opc == AMDGPU::V_CNDMASK_B32_e32)
    return false;

[llvmbot]

@llvm/issue-subscribers-backend-amdgpu

Author: Frederik Harwath (frederik-h)

The `isConvertibleToSDWA` function form `llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp` contains this FIXME: ```c++ // FIXME: has SDWA but require handling of implicit VCC use if (Opc == AMDGPU::V_CNDMASK_B32_e32) return false; ```

[frederik-h]

Note that the AMDGPU::V_CNDMASK_B32_e32 instruction seems to be unproblematic, but the function converts e64 to e32 opcodes at the top and hence the if statement rejects both opcodes. The conversion of AMDGPU::V_CNDMASK_B32_e64 to an SDWA instruction requires that the explicit src2/carryin be provided through vcc.