[llvm] [AMDGPU] Fix opsel for scaled MFMA operations (PR #140183)

[Matt Arsenault via llvm-commits]

================
@@ -514,7 +514,7 @@ v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:31], v[32:47], v48, v49
 // ERR: :[[@LINE+1]]:{{[0-9]+}}: error: instruction not supported on this GPU
 v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:23], v[24:29], v[32:47], v48, v49 blgp:2
 
-// GFX950: v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:21], v[24:29], v[32:47], v48, v49 op_sel_hi:[0,0,0] cbsz:2 blgp:3 ; encoding: [0x00,0x00,0xac,0xd3,0x30,0x63,0x02,0x18,0x00,0x0a,0xae,0xd3,0x10,0x31,0x82,0x64]
+// GFX950: v_mfma_scale_f32_32x32x64_f8f6f4 v[0:15], v[16:21], v[24:29], v[32:47], v48, v49 op_sel:[0,1,0] op_sel_hi:[0,1,0] cbsz:2 blgp:3 ; encoding: [0x00,0x10,0xac,0xd3,0x30,0x63,0x02,0x10,0x00,0x0a,0xae,0xd3,0x10,0x31,0x82,0x64]
----------------
arsenm wrote:

This doesn't seem like enough lines changed. Do we have every permutation of values tested? 

https://github.com/llvm/llvm-project/pull/140183