[llvm] [RISCV] Enable rematerialization for scalar loads (PR #166774)

[Philip Reames via llvm-commits]

================
@@ -200,3 +200,169 @@ for.end:                                          ; preds = %for.inc, %entry
 }
 
 declare i32 @foo(i32, i32, i32, i32, i32, i32)
+
+define void @remat_load(i32 %0, i32 %1, i32 %2, i32 %3, i32 %4, i32 %5, i32 %6, i32 %7, double %8, double %9, double %10, double %11, double %12, double %13, double %14, double %15, i8 %stackarg0, i16 %stackarg1, i32 %stackarg2, i64 %stackarg3, half %stackarg4, bfloat %stackarg5, float %stackarg6, double %stackarg7, ptr %p) nounwind {
+; CHECK-LABEL: remat_load:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    addi sp, sp, -208
+; CHECK-NEXT:    sd ra, 200(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s0, 192(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s1, 184(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s2, 176(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s3, 168(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s4, 160(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s5, 152(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s6, 144(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s7, 136(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s8, 128(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s9, 120(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s10, 112(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    sd s11, 104(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs0, 96(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs1, 88(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs2, 80(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs3, 72(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs4, 64(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs5, 56(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs6, 48(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs7, 40(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs8, 32(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs9, 24(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs10, 16(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fsd fs11, 8(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    fld fa5, 264(sp)
+; CHECK-NEXT:    flw fa4, 256(sp)
+; CHECK-NEXT:    flh fa3, 248(sp)
+; CHECK-NEXT:    flh fa2, 240(sp)
+; CHECK-NEXT:    ld a0, 272(sp)
+; CHECK-NEXT:    lbu a4, 208(sp)
+; CHECK-NEXT:    lh a3, 216(sp)
+; CHECK-NEXT:    lw a2, 224(sp)
+; CHECK-NEXT:    ld a1, 232(sp)
+; CHECK-NEXT:    sb a4, 0(a0)
+; CHECK-NEXT:    sh a3, 0(a0)
+; CHECK-NEXT:    sw a2, 0(a0)
+; CHECK-NEXT:    sd a1, 0(a0)
+; CHECK-NEXT:    fsh fa2, 0(a0)
+; CHECK-NEXT:    fsh fa3, 0(a0)
+; CHECK-NEXT:    fsw fa4, 0(a0)
+; CHECK-NEXT:    fsd fa5, 0(a0)
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    ld a0, 272(sp)
+; CHECK-NEXT:    lbu a1, 208(sp)
+; CHECK-NEXT:    sb a1, 0(a0)
+; CHECK-NEXT:    lh a1, 216(sp)
+; CHECK-NEXT:    sh a1, 0(a0)
+; CHECK-NEXT:    lw a1, 224(sp)
+; CHECK-NEXT:    sw a1, 0(a0)
+; CHECK-NEXT:    ld a1, 232(sp)
+; CHECK-NEXT:    sd a1, 0(a0)
+; CHECK-NEXT:    flh fa5, 240(sp)
+; CHECK-NEXT:    fsh fa5, 0(a0)
+; CHECK-NEXT:    flh fa5, 248(sp)
+; CHECK-NEXT:    fsh fa5, 0(a0)
+; CHECK-NEXT:    flw fa5, 256(sp)
+; CHECK-NEXT:    fsw fa5, 0(a0)
+; CHECK-NEXT:    fld fa5, 264(sp)
+; CHECK-NEXT:    fsd fa5, 0(a0)
+; CHECK-NEXT:    ld ra, 200(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s0, 192(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s1, 184(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s2, 176(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s3, 168(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s4, 160(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s5, 152(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s6, 144(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s7, 136(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s8, 128(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s9, 120(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s10, 112(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    ld s11, 104(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs0, 96(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs1, 88(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs2, 80(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs3, 72(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs4, 64(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs5, 56(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs6, 48(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs7, 40(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs8, 32(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs9, 24(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs10, 16(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    fld fs11, 8(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    addi sp, sp, 208
+; CHECK-NEXT:    ret
+entry:
+  ; Add a use of the stack arguments here so that we will have to load them from
+  ; the stack before the inline asm
----------------
preames wrote:

Add "because otherwise we'd be exercising the machine scheduler, not rematerialization"

https://github.com/llvm/llvm-project/pull/166774