[llvm] [AMDGPU][SDAG] Legalise v2i32 or/xor/and instructions to make use of 64-bit wide instructions (PR #140694)

[Matt Arsenault via llvm-commits]

================
@@ -2378,16 +2379,54 @@ def : AMDGPUPat <
 let True16Predicate = NotHasTrue16BitInsts in {
 def : ROTRPattern <V_ALIGNBIT_B32_e64>;
 
+def : AMDGPUPat <
+  (rotr v2i32:$src0, v2i32:$src1),
+  (REG_SEQUENCE VReg_64,
+    (V_ALIGNBIT_B32_e64
+      (i32 (EXTRACT_SUBREG VReg_64:$src0, sub0)), 
+      (i32 (EXTRACT_SUBREG VReg_64:$src0, sub0)), 
+      (i32 (EXTRACT_SUBREG VReg_64:$src1, sub0))), sub0,
+    (V_ALIGNBIT_B32_e64
+      (i32 (EXTRACT_SUBREG VReg_64:$src0, sub1)),
+      (i32 (EXTRACT_SUBREG VReg_64:$src0, sub1)), 
+      (i32 (EXTRACT_SUBREG VReg_64:$src1, sub1))), sub1)
+>;
+
+// Prevents regression in fneg-modifier-casting.ll along with modifications to XorCombine() when v2i32 or is legal.
+def : AMDGPUPat <
+  (fneg (select i1:$src0, (f32 (bitconvert i32:$src1)), (f32 (bitconvert i32:$src2)))),
+    (V_CNDMASK_B32_e64 (i32 1), $src2, (i32 1), $src1, $src0)>;
----------------
arsenm wrote:

That should be solved in the fneg formation combines, we should have pushed this fneg into the source position of the select. The heuristics for where to put source modifiers is more context dependent, and shouldn't be done in selection 

https://github.com/llvm/llvm-project/pull/140694