[PATCH] D46853: AMDGPU: Add disasm tests for deep learning instructions + fix v_fmac_f32 disasm

[Konstantin Zhuravlyov via Phabricator via llvm-commits]

kzhuravl added inline comments.


================
Comment at: lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp:244-247
   if (Res && (MI.getOpcode() == AMDGPU::V_MAC_F32_e64_vi ||
               MI.getOpcode() == AMDGPU::V_MAC_F32_e64_si ||
-              MI.getOpcode() == AMDGPU::V_MAC_F16_e64_vi)) {
+              MI.getOpcode() == AMDGPU::V_MAC_F16_e64_vi ||
+              MI.getOpcode() == AMDGPU::V_FMAC_F32_e64_vi)) {
----------------
arsenm wrote:
> Could we avoid this problem in the future by testing for a tied src2 operand?
Ok, let me prepare the chagne.


Repository:
  rL LLVM

https://reviews.llvm.org/D46853