[LoongArch] Switch to the Machine Scheduler (#83759)

The SelectionDAG scheduling preference now becomes source order
scheduling (machine scheduler generates better code -- even without
there being a machine model defined for LoongArch yet).

Most of the test changes are trivial instruction reorderings and
differing register allocations, without any obvious performance impact.

This is similar to commit: 3d0fbaf

Author: wangleiat

llvm/lib/Target/LoongArch/LoongArchSubtarget.h

@@ -113,6 +113,7 @@ class LoongArchSubtarget : public LoongArchGenSubtargetInfo {
   Align getPrefFunctionAlignment() const { return PrefFunctionAlignment; }
   Align getPrefLoopAlignment() const { return PrefLoopAlignment; }
   unsigned getMaxBytesForAlignment() const { return MaxBytesForAlignment; }
+  bool enableMachineScheduler() const override { return true; }
 };
 } // end namespace llvm
 

llvm/test/CodeGen/LoongArch/alloca.ll

@@ -126,15 +126,15 @@ define void @alloca_callframe(i32 %n) nounwind {
 ; LA32-NEXT:    st.w $a1, $sp, 8
 ; LA32-NEXT:    ori $a1, $zero, 10
 ; LA32-NEXT:    st.w $a1, $sp, 4
-; LA32-NEXT:    ori $a1, $zero, 9
-; LA32-NEXT:    st.w $a1, $sp, 0
+; LA32-NEXT:    ori $t0, $zero, 9
 ; LA32-NEXT:    ori $a1, $zero, 2
 ; LA32-NEXT:    ori $a2, $zero, 3
 ; LA32-NEXT:    ori $a3, $zero, 4
 ; LA32-NEXT:    ori $a4, $zero, 5
 ; LA32-NEXT:    ori $a5, $zero, 6
 ; LA32-NEXT:    ori $a6, $zero, 7
 ; LA32-NEXT:    ori $a7, $zero, 8
+; LA32-NEXT:    st.w $t0, $sp, 0
 ; LA32-NEXT:    bl %plt(func)
 ; LA32-NEXT:    addi.w $sp, $sp, 16
 ; LA32-NEXT:    addi.w $sp, $fp, -16
@@ -162,15 +162,15 @@ define void @alloca_callframe(i32 %n) nounwind {
 ; LA64-NEXT:    st.d $a1, $sp, 16
 ; LA64-NEXT:    ori $a1, $zero, 10
 ; LA64-NEXT:    st.d $a1, $sp, 8
-; LA64-NEXT:    ori $a1, $zero, 9
-; LA64-NEXT:    st.d $a1, $sp, 0
+; LA64-NEXT:    ori $t0, $zero, 9
 ; LA64-NEXT:    ori $a1, $zero, 2
 ; LA64-NEXT:    ori $a2, $zero, 3
 ; LA64-NEXT:    ori $a3, $zero, 4
 ; LA64-NEXT:    ori $a4, $zero, 5
 ; LA64-NEXT:    ori $a5, $zero, 6
 ; LA64-NEXT:    ori $a6, $zero, 7
 ; LA64-NEXT:    ori $a7, $zero, 8
+; LA64-NEXT:    st.d $t0, $sp, 0
 ; LA64-NEXT:    bl %plt(func)
 ; LA64-NEXT:    addi.d $sp, $sp, 32
 ; LA64-NEXT:    addi.d $sp, $fp, -16

llvm/test/CodeGen/LoongArch/alsl.ll

@@ -53,12 +53,12 @@ entry:
 define i64 @alsl_i64(i64 signext %a, i64 signext %b) nounwind {
 ; LA32-LABEL: alsl_i64:
 ; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    slli.w $a1, $a1, 4
 ; LA32-NEXT:    srli.w $a4, $a0, 28
+; LA32-NEXT:    slli.w $a1, $a1, 4
 ; LA32-NEXT:    or $a1, $a1, $a4
-; LA32-NEXT:    add.w $a1, $a3, $a1
 ; LA32-NEXT:    alsl.w $a0, $a0, $a2, 4
 ; LA32-NEXT:    sltu $a2, $a0, $a2
+; LA32-NEXT:    add.w $a1, $a3, $a1
 ; LA32-NEXT:    add.w $a1, $a1, $a2
 ; LA32-NEXT:    ret
 ;
@@ -189,14 +189,14 @@ entry:
 define i64 @mul_add_i64(i64 signext %a, i64 signext %b) nounwind {
 ; LA32-LABEL: mul_add_i64:
 ; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    slli.w $a4, $a1, 4
-; LA32-NEXT:    sub.w $a1, $a4, $a1
 ; LA32-NEXT:    ori $a4, $zero, 15
 ; LA32-NEXT:    mulh.wu $a4, $a0, $a4
+; LA32-NEXT:    slli.w $a5, $a1, 4
+; LA32-NEXT:    sub.w $a1, $a5, $a1
 ; LA32-NEXT:    add.w $a1, $a4, $a1
+; LA32-NEXT:    slli.w $a4, $a0, 4
+; LA32-NEXT:    sub.w $a0, $a4, $a0
 ; LA32-NEXT:    add.w $a1, $a3, $a1
-; LA32-NEXT:    slli.w $a3, $a0, 4
-; LA32-NEXT:    sub.w $a0, $a3, $a0
 ; LA32-NEXT:    add.w $a0, $a2, $a0
 ; LA32-NEXT:    sltu $a2, $a0, $a2
 ; LA32-NEXT:    add.w $a1, $a1, $a2
@@ -342,9 +342,9 @@ define i64 @mul_add_neg_i64(i64 signext %a, i64 signext %b) nounwind {
 ; LA32-NEXT:    mulh.wu $a4, $a0, $a4
 ; LA32-NEXT:    sub.w $a4, $a4, $a0
 ; LA32-NEXT:    add.w $a1, $a4, $a1
+; LA32-NEXT:    slli.w $a4, $a0, 4
+; LA32-NEXT:    sub.w $a0, $a0, $a4
 ; LA32-NEXT:    add.w $a1, $a3, $a1
-; LA32-NEXT:    slli.w $a3, $a0, 4
-; LA32-NEXT:    sub.w $a0, $a0, $a3
 ; LA32-NEXT:    add.w $a0, $a2, $a0
 ; LA32-NEXT:    sltu $a2, $a0, $a2
 ; LA32-NEXT:    add.w $a1, $a1, $a2