[llvm] [ExpandMemCmp][RISCV] Expand memcmp/bcmp for aligned pointers on strict-align targets (PR #209738)
Luke Lau via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 16 05:14:42 PDT 2026
================
@@ -2722,194 +2382,38 @@ entry:
%ret = icmp eq i32 %bcmp, 0
ret i1 %ret
}
-
-define i1 @bcmp_lt_zero(ptr %s1, ptr %s2) nounwind optsize {
-; CHECK-ALIGNED-RV32-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV32: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-NEXT: srli a0, a0, 31
-; CHECK-ALIGNED-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV64: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-NEXT: srli a0, a0, 63
-; CHECK-ALIGNED-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-NEXT: ret
-;
-; CHECK-ALIGNED-RV32-ZBB-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV32-ZBB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-ZBB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-ZBB-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-ZBB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-ZBB-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-ZBB-NEXT: srli a0, a0, 31
-; CHECK-ALIGNED-RV32-ZBB-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-ZBB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-ZBB-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-ZBB-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV64-ZBB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-ZBB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-ZBB-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-ZBB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-ZBB-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-ZBB-NEXT: srli a0, a0, 63
-; CHECK-ALIGNED-RV64-ZBB-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-ZBB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-ZBB-NEXT: ret
-;
-; CHECK-ALIGNED-RV32-ZBKB-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV32-ZBKB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: srli a0, a0, 31
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-ZBKB-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV64-ZBKB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: srli a0, a0, 63
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: ret
-;
-; CHECK-ALIGNED-RV32-V-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV32-V: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-V-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-V-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-V-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-V-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-V-NEXT: srli a0, a0, 31
-; CHECK-ALIGNED-RV32-V-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-V-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-V-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-V-LABEL: bcmp_lt_zero:
-; CHECK-ALIGNED-RV64-V: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-V-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-V-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-V-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-V-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-V-NEXT: srli a0, a0, 63
-; CHECK-ALIGNED-RV64-V-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-V-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-V-NEXT: ret
-;
-; CHECK-UNALIGNED-LABEL: bcmp_lt_zero:
-; CHECK-UNALIGNED: # %bb.0: # %entry
-; CHECK-UNALIGNED-NEXT: li a0, 0
-; CHECK-UNALIGNED-NEXT: ret
+
+define i1 @bcmp_lt_zero(ptr %s1, ptr %s2) nounwind optsize {
+; CHECK-LABEL: bcmp_lt_zero:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: li a0, 0
+; CHECK-NEXT: ret
entry:
%bcmp = call signext i32 @bcmp(ptr %s1, ptr %s2, iXLen 4)
%ret = icmp slt i32 %bcmp, 0
ret i1 %ret
}
define i1 @bcmp_gt_zero(ptr %s1, ptr %s2) nounwind optsize {
-; CHECK-ALIGNED-RV32-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV32: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV64: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-NEXT: ret
-;
-; CHECK-ALIGNED-RV32-ZBB-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV32-ZBB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-ZBB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-ZBB-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-ZBB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-ZBB-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-ZBB-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV32-ZBB-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-ZBB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-ZBB-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-ZBB-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV64-ZBB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-ZBB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-ZBB-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-ZBB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-ZBB-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-ZBB-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV64-ZBB-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-ZBB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-ZBB-NEXT: ret
-;
-; CHECK-ALIGNED-RV32-ZBKB-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV32-ZBKB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-ZBKB-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-ZBKB-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV64-ZBKB: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-ZBKB-NEXT: ret
-;
-; CHECK-ALIGNED-RV32-V-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV32-V: # %bb.0: # %entry
-; CHECK-ALIGNED-RV32-V-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV32-V-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
-; CHECK-ALIGNED-RV32-V-NEXT: li a2, 4
-; CHECK-ALIGNED-RV32-V-NEXT: call bcmp
-; CHECK-ALIGNED-RV32-V-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV32-V-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
-; CHECK-ALIGNED-RV32-V-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV32-V-NEXT: ret
-;
-; CHECK-ALIGNED-RV64-V-LABEL: bcmp_gt_zero:
-; CHECK-ALIGNED-RV64-V: # %bb.0: # %entry
-; CHECK-ALIGNED-RV64-V-NEXT: addi sp, sp, -16
-; CHECK-ALIGNED-RV64-V-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
-; CHECK-ALIGNED-RV64-V-NEXT: li a2, 4
-; CHECK-ALIGNED-RV64-V-NEXT: call bcmp
-; CHECK-ALIGNED-RV64-V-NEXT: sgtz a0, a0
-; CHECK-ALIGNED-RV64-V-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
-; CHECK-ALIGNED-RV64-V-NEXT: addi sp, sp, 16
-; CHECK-ALIGNED-RV64-V-NEXT: ret
+; CHECK-ALIGNED-LABEL: bcmp_gt_zero:
+; CHECK-ALIGNED: # %bb.0: # %entry
+; CHECK-ALIGNED-NEXT: lbu a2, 0(a0)
+; CHECK-ALIGNED-NEXT: lbu a3, 1(a0)
+; CHECK-ALIGNED-NEXT: lbu a4, 2(a0)
+; CHECK-ALIGNED-NEXT: lbu a0, 3(a0)
+; CHECK-ALIGNED-NEXT: lbu a5, 0(a1)
+; CHECK-ALIGNED-NEXT: lbu a6, 1(a1)
+; CHECK-ALIGNED-NEXT: lbu a7, 2(a1)
+; CHECK-ALIGNED-NEXT: lbu a1, 3(a1)
----------------
lukel97 wrote:
Is this producing more loads than MaxLoadsPerMemcmpOptSize? I think it's set to 4.
https://github.com/llvm/llvm-project/pull/209738
More information about the llvm-commits
mailing list