[llvm] [AMDGPU] Implement vop3p complex pattern optmization for gisel (PR #130234)

[Diana Picus via llvm-commits]

================
@@ -263,8 +262,7 @@ define i32 @v_sdot2_fnegf32_c(<2 x i16> %a, <2 x i16> %b, float %c) {
 ; GFX10-LABEL: v_sdot2_fnegf32_c:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_xor_b32_e32 v2, 0x80000000, v2
-; GFX10-NEXT:    v_dot2_i32_i16 v0, v0, v1, v2
+; GFX10-NEXT:    v_dot2_i32_i16 v0, v0, v1, v2 neg_lo:[0,0,1] neg_hi:[0,0,1]
----------------
rovka wrote:

I think we [don't support](https://rocm.docs.amd.com/projects/llvm-project/en/docs-6.2.4/LLVM/llvm/html/AMDGPU/AMDGPUAsmGFX1030.html#vop3p) the neg modifiers for integer operands.

https://github.com/llvm/llvm-project/pull/130234