[llvm] AMDGPU: Add tonearest and towardzero roundings for intrinsic llvm.fptrunc.round (PR #104486)

[Matt Arsenault via llvm-commits]

================
@@ -23,6 +32,16 @@ define amdgpu_gs half @v_fptrunc_round_f32_to_f16_downward(float %a) {
   ret half %res
 }
 
+define amdgpu_gs half @v_fptrunc_round_f32_to_f16_towardzero(float %a) {
+; CHECK-LABEL: v_fptrunc_round_f32_to_f16_towardzero:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 3
+; CHECK-NEXT:    v_cvt_f16_f32_e32 v0, v0
+; CHECK-NEXT:    ; return to shader part epilog
+  %res = call half @llvm.fptrunc.round.f16.f32(float %a, metadata !"round.towardzero")
+  ret half %res
+}
+
----------------
arsenm wrote:

can you also include a test where multiple modes are used in the same function 

https://github.com/llvm/llvm-project/pull/104486