[llvm] cddeb78 - [RISCV] Add test cases showing failure to remove mask on rotate amounts.
Craig Topper via llvm-commits
llvm-commits at lists.llvm.org
Tue May 24 09:41:46 PDT 2022
Author: Craig Topper
Date: 2022-05-24T09:41:00-07:00
New Revision: cddeb78e8d5b7b924c920fb702d3d883b2661b12
URL: https://github.com/llvm/llvm-project/commit/cddeb78e8d5b7b924c920fb702d3d883b2661b12
DIFF: https://github.com/llvm/llvm-project/commit/cddeb78e8d5b7b924c920fb702d3d883b2661b12.diff
LOG: [RISCV] Add test cases showing failure to remove mask on rotate amounts.
This is similar to tests I added in
e2f410feeab27a8bb2c015fc02bb8527702e401f that had to be reverted.
I've modified them to avoid the bug that is being fixed by D126036.
Added:
Modified:
llvm/test/CodeGen/RISCV/rotl-rotr.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/RISCV/rotl-rotr.ll b/llvm/test/CodeGen/RISCV/rotl-rotr.ll
index fd5437e3cb147..0cd4866515ad3 100644
--- a/llvm/test/CodeGen/RISCV/rotl-rotr.ll
+++ b/llvm/test/CodeGen/RISCV/rotl-rotr.ll
@@ -550,3 +550,313 @@ define i64 @rotr_64_mask(i64 %x, i64 %y) nounwind {
%d = or i64 %b, %c
ret i64 %d
}
+
+; Test that we're able to remove a mask on the rotate amount that has more than
+; one use.
+define signext i32 @rotl_32_mask_shared(i32 signext %a, i32 signext %b, i32 signext %amt) nounwind {
+; RV32I-LABEL: rotl_32_mask_shared:
+; RV32I: # %bb.0:
+; RV32I-NEXT: sll a3, a0, a2
+; RV32I-NEXT: neg a4, a2
+; RV32I-NEXT: srl a0, a0, a4
+; RV32I-NEXT: or a0, a3, a0
+; RV32I-NEXT: sll a1, a1, a2
+; RV32I-NEXT: add a0, a0, a1
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: rotl_32_mask_shared:
+; RV64I: # %bb.0:
+; RV64I-NEXT: sllw a3, a0, a2
+; RV64I-NEXT: negw a4, a2
+; RV64I-NEXT: srlw a0, a0, a4
+; RV64I-NEXT: or a0, a3, a0
+; RV64I-NEXT: sllw a1, a1, a2
+; RV64I-NEXT: addw a0, a0, a1
+; RV64I-NEXT: ret
+;
+; RV32ZBB-LABEL: rotl_32_mask_shared:
+; RV32ZBB: # %bb.0:
+; RV32ZBB-NEXT: andi a3, a2, 31
+; RV32ZBB-NEXT: rol a0, a0, a3
+; RV32ZBB-NEXT: sll a1, a1, a2
+; RV32ZBB-NEXT: add a0, a0, a1
+; RV32ZBB-NEXT: ret
+;
+; RV64ZBB-LABEL: rotl_32_mask_shared:
+; RV64ZBB: # %bb.0:
+; RV64ZBB-NEXT: rolw a0, a0, a2
+; RV64ZBB-NEXT: sllw a1, a1, a2
+; RV64ZBB-NEXT: addw a0, a0, a1
+; RV64ZBB-NEXT: ret
+ %maskedamt = and i32 %amt, 31
+ %1 = tail call i32 @llvm.fshl.i32(i32 %a, i32 %a, i32 %maskedamt)
+ %2 = shl i32 %b, %maskedamt
+ %3 = add i32 %1, %2
+ ret i32 %3
+}
+declare i32 @llvm.fshl.i32(i32, i32, i32)
+
+define signext i64 @rotl_64_mask_shared(i64 signext %a, i64 signext %b, i64 signext %amt) nounwind {
+; RV32I-LABEL: rotl_64_mask_shared:
+; RV32I: # %bb.0:
+; RV32I-NEXT: slli a5, a4, 26
+; RV32I-NEXT: srli a5, a5, 31
+; RV32I-NEXT: mv a7, a0
+; RV32I-NEXT: bnez a5, .LBB9_2
+; RV32I-NEXT: # %bb.1:
+; RV32I-NEXT: mv a7, a1
+; RV32I-NEXT: .LBB9_2:
+; RV32I-NEXT: andi a6, a4, 63
+; RV32I-NEXT: sll t0, a7, a4
+; RV32I-NEXT: bnez a5, .LBB9_4
+; RV32I-NEXT: # %bb.3:
+; RV32I-NEXT: mv a1, a0
+; RV32I-NEXT: .LBB9_4:
+; RV32I-NEXT: srli a0, a1, 1
+; RV32I-NEXT: not t1, a4
+; RV32I-NEXT: srl a0, a0, t1
+; RV32I-NEXT: or a5, t0, a0
+; RV32I-NEXT: sll a1, a1, a4
+; RV32I-NEXT: srli a0, a7, 1
+; RV32I-NEXT: srl a7, a0, t1
+; RV32I-NEXT: addi a0, a6, -32
+; RV32I-NEXT: or a1, a1, a7
+; RV32I-NEXT: bltz a0, .LBB9_6
+; RV32I-NEXT: # %bb.5:
+; RV32I-NEXT: sll a3, a2, a0
+; RV32I-NEXT: mv a0, a1
+; RV32I-NEXT: j .LBB9_7
+; RV32I-NEXT: .LBB9_6:
+; RV32I-NEXT: sll a0, a3, a4
+; RV32I-NEXT: srli a3, a2, 1
+; RV32I-NEXT: xori a6, a6, 31
+; RV32I-NEXT: srl a3, a3, a6
+; RV32I-NEXT: or a3, a0, a3
+; RV32I-NEXT: sll a0, a2, a4
+; RV32I-NEXT: add a0, a1, a0
+; RV32I-NEXT: .LBB9_7:
+; RV32I-NEXT: sltu a1, a0, a1
+; RV32I-NEXT: add a2, a5, a3
+; RV32I-NEXT: add a1, a2, a1
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: rotl_64_mask_shared:
+; RV64I: # %bb.0:
+; RV64I-NEXT: sll a3, a0, a2
+; RV64I-NEXT: neg a4, a2
+; RV64I-NEXT: srl a0, a0, a4
+; RV64I-NEXT: or a0, a3, a0
+; RV64I-NEXT: sll a1, a1, a2
+; RV64I-NEXT: add a0, a0, a1
+; RV64I-NEXT: ret
+;
+; RV32ZBB-LABEL: rotl_64_mask_shared:
+; RV32ZBB: # %bb.0:
+; RV32ZBB-NEXT: slli a5, a4, 26
+; RV32ZBB-NEXT: srli a5, a5, 31
+; RV32ZBB-NEXT: mv a7, a0
+; RV32ZBB-NEXT: bnez a5, .LBB9_2
+; RV32ZBB-NEXT: # %bb.1:
+; RV32ZBB-NEXT: mv a7, a1
+; RV32ZBB-NEXT: .LBB9_2:
+; RV32ZBB-NEXT: andi a6, a4, 63
+; RV32ZBB-NEXT: sll t0, a7, a4
+; RV32ZBB-NEXT: bnez a5, .LBB9_4
+; RV32ZBB-NEXT: # %bb.3:
+; RV32ZBB-NEXT: mv a1, a0
+; RV32ZBB-NEXT: .LBB9_4:
+; RV32ZBB-NEXT: srli a0, a1, 1
+; RV32ZBB-NEXT: not t1, a4
+; RV32ZBB-NEXT: srl a0, a0, t1
+; RV32ZBB-NEXT: or a5, t0, a0
+; RV32ZBB-NEXT: sll a1, a1, a4
+; RV32ZBB-NEXT: srli a0, a7, 1
+; RV32ZBB-NEXT: srl a7, a0, t1
+; RV32ZBB-NEXT: addi a0, a6, -32
+; RV32ZBB-NEXT: or a1, a1, a7
+; RV32ZBB-NEXT: bltz a0, .LBB9_6
+; RV32ZBB-NEXT: # %bb.5:
+; RV32ZBB-NEXT: sll a3, a2, a0
+; RV32ZBB-NEXT: mv a0, a1
+; RV32ZBB-NEXT: j .LBB9_7
+; RV32ZBB-NEXT: .LBB9_6:
+; RV32ZBB-NEXT: sll a0, a3, a4
+; RV32ZBB-NEXT: srli a3, a2, 1
+; RV32ZBB-NEXT: xori a6, a6, 31
+; RV32ZBB-NEXT: srl a3, a3, a6
+; RV32ZBB-NEXT: or a3, a0, a3
+; RV32ZBB-NEXT: sll a0, a2, a4
+; RV32ZBB-NEXT: add a0, a1, a0
+; RV32ZBB-NEXT: .LBB9_7:
+; RV32ZBB-NEXT: sltu a1, a0, a1
+; RV32ZBB-NEXT: add a2, a5, a3
+; RV32ZBB-NEXT: add a1, a2, a1
+; RV32ZBB-NEXT: ret
+;
+; RV64ZBB-LABEL: rotl_64_mask_shared:
+; RV64ZBB: # %bb.0:
+; RV64ZBB-NEXT: andi a3, a2, 63
+; RV64ZBB-NEXT: rol a0, a0, a3
+; RV64ZBB-NEXT: sll a1, a1, a2
+; RV64ZBB-NEXT: add a0, a0, a1
+; RV64ZBB-NEXT: ret
+ %maskedamt = and i64 %amt, 63
+ %1 = tail call i64 @llvm.fshl.i64(i64 %a, i64 %a, i64 %maskedamt)
+ %2 = shl i64 %b, %maskedamt
+ %3 = add i64 %1, %2
+ ret i64 %3
+}
+declare i64 @llvm.fshl.i64(i64, i64, i64)
+
+define signext i32 @rotr_32_mask_shared(i32 signext %a, i32 signext %b, i32 signext %amt) nounwind {
+; RV32I-LABEL: rotr_32_mask_shared:
+; RV32I: # %bb.0:
+; RV32I-NEXT: srl a3, a0, a2
+; RV32I-NEXT: neg a4, a2
+; RV32I-NEXT: sll a0, a0, a4
+; RV32I-NEXT: or a0, a3, a0
+; RV32I-NEXT: sll a1, a1, a2
+; RV32I-NEXT: add a0, a0, a1
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: rotr_32_mask_shared:
+; RV64I: # %bb.0:
+; RV64I-NEXT: srlw a3, a0, a2
+; RV64I-NEXT: negw a4, a2
+; RV64I-NEXT: sllw a0, a0, a4
+; RV64I-NEXT: or a0, a3, a0
+; RV64I-NEXT: sllw a1, a1, a2
+; RV64I-NEXT: addw a0, a0, a1
+; RV64I-NEXT: ret
+;
+; RV32ZBB-LABEL: rotr_32_mask_shared:
+; RV32ZBB: # %bb.0:
+; RV32ZBB-NEXT: andi a3, a2, 31
+; RV32ZBB-NEXT: ror a0, a0, a3
+; RV32ZBB-NEXT: sll a1, a1, a2
+; RV32ZBB-NEXT: add a0, a0, a1
+; RV32ZBB-NEXT: ret
+;
+; RV64ZBB-LABEL: rotr_32_mask_shared:
+; RV64ZBB: # %bb.0:
+; RV64ZBB-NEXT: rorw a0, a0, a2
+; RV64ZBB-NEXT: sllw a1, a1, a2
+; RV64ZBB-NEXT: addw a0, a0, a1
+; RV64ZBB-NEXT: ret
+ %maskedamt = and i32 %amt, 31
+ %1 = tail call i32 @llvm.fshr.i32(i32 %a, i32 %a, i32 %maskedamt)
+ %2 = shl i32 %b, %maskedamt
+ %3 = add i32 %1, %2
+ ret i32 %3
+}
+declare i32 @llvm.fshr.i32(i32, i32, i32)
+
+define signext i64 @rotr_64_mask_shared(i64 signext %a, i64 signext %b, i64 signext %amt) nounwind {
+; RV32I-LABEL: rotr_64_mask_shared:
+; RV32I: # %bb.0:
+; RV32I-NEXT: andi a7, a4, 32
+; RV32I-NEXT: mv a6, a1
+; RV32I-NEXT: beqz a7, .LBB11_2
+; RV32I-NEXT: # %bb.1:
+; RV32I-NEXT: mv a6, a0
+; RV32I-NEXT: .LBB11_2:
+; RV32I-NEXT: andi a5, a4, 63
+; RV32I-NEXT: srl t0, a6, a4
+; RV32I-NEXT: beqz a7, .LBB11_4
+; RV32I-NEXT: # %bb.3:
+; RV32I-NEXT: mv a0, a1
+; RV32I-NEXT: .LBB11_4:
+; RV32I-NEXT: slli a1, a0, 1
+; RV32I-NEXT: not a7, a4
+; RV32I-NEXT: sll a1, a1, a7
+; RV32I-NEXT: or a1, a1, t0
+; RV32I-NEXT: srl t0, a0, a4
+; RV32I-NEXT: slli a0, a6, 1
+; RV32I-NEXT: sll a6, a0, a7
+; RV32I-NEXT: addi a0, a5, -32
+; RV32I-NEXT: or a6, a6, t0
+; RV32I-NEXT: bltz a0, .LBB11_6
+; RV32I-NEXT: # %bb.5:
+; RV32I-NEXT: sll a3, a2, a0
+; RV32I-NEXT: mv a0, a6
+; RV32I-NEXT: j .LBB11_7
+; RV32I-NEXT: .LBB11_6:
+; RV32I-NEXT: sll a0, a3, a4
+; RV32I-NEXT: srli a3, a2, 1
+; RV32I-NEXT: xori a5, a5, 31
+; RV32I-NEXT: srl a3, a3, a5
+; RV32I-NEXT: or a3, a0, a3
+; RV32I-NEXT: sll a0, a2, a4
+; RV32I-NEXT: add a0, a6, a0
+; RV32I-NEXT: .LBB11_7:
+; RV32I-NEXT: sltu a2, a0, a6
+; RV32I-NEXT: add a1, a1, a3
+; RV32I-NEXT: add a1, a1, a2
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: rotr_64_mask_shared:
+; RV64I: # %bb.0:
+; RV64I-NEXT: srl a3, a0, a2
+; RV64I-NEXT: neg a4, a2
+; RV64I-NEXT: sll a0, a0, a4
+; RV64I-NEXT: or a0, a3, a0
+; RV64I-NEXT: sll a1, a1, a2
+; RV64I-NEXT: add a0, a0, a1
+; RV64I-NEXT: ret
+;
+; RV32ZBB-LABEL: rotr_64_mask_shared:
+; RV32ZBB: # %bb.0:
+; RV32ZBB-NEXT: andi a7, a4, 32
+; RV32ZBB-NEXT: mv a6, a1
+; RV32ZBB-NEXT: beqz a7, .LBB11_2
+; RV32ZBB-NEXT: # %bb.1:
+; RV32ZBB-NEXT: mv a6, a0
+; RV32ZBB-NEXT: .LBB11_2:
+; RV32ZBB-NEXT: andi a5, a4, 63
+; RV32ZBB-NEXT: srl t0, a6, a4
+; RV32ZBB-NEXT: beqz a7, .LBB11_4
+; RV32ZBB-NEXT: # %bb.3:
+; RV32ZBB-NEXT: mv a0, a1
+; RV32ZBB-NEXT: .LBB11_4:
+; RV32ZBB-NEXT: slli a1, a0, 1
+; RV32ZBB-NEXT: not a7, a4
+; RV32ZBB-NEXT: sll a1, a1, a7
+; RV32ZBB-NEXT: or a1, a1, t0
+; RV32ZBB-NEXT: srl t0, a0, a4
+; RV32ZBB-NEXT: slli a0, a6, 1
+; RV32ZBB-NEXT: sll a6, a0, a7
+; RV32ZBB-NEXT: addi a0, a5, -32
+; RV32ZBB-NEXT: or a6, a6, t0
+; RV32ZBB-NEXT: bltz a0, .LBB11_6
+; RV32ZBB-NEXT: # %bb.5:
+; RV32ZBB-NEXT: sll a3, a2, a0
+; RV32ZBB-NEXT: mv a0, a6
+; RV32ZBB-NEXT: j .LBB11_7
+; RV32ZBB-NEXT: .LBB11_6:
+; RV32ZBB-NEXT: sll a0, a3, a4
+; RV32ZBB-NEXT: srli a3, a2, 1
+; RV32ZBB-NEXT: xori a5, a5, 31
+; RV32ZBB-NEXT: srl a3, a3, a5
+; RV32ZBB-NEXT: or a3, a0, a3
+; RV32ZBB-NEXT: sll a0, a2, a4
+; RV32ZBB-NEXT: add a0, a6, a0
+; RV32ZBB-NEXT: .LBB11_7:
+; RV32ZBB-NEXT: sltu a2, a0, a6
+; RV32ZBB-NEXT: add a1, a1, a3
+; RV32ZBB-NEXT: add a1, a1, a2
+; RV32ZBB-NEXT: ret
+;
+; RV64ZBB-LABEL: rotr_64_mask_shared:
+; RV64ZBB: # %bb.0:
+; RV64ZBB-NEXT: andi a3, a2, 63
+; RV64ZBB-NEXT: ror a0, a0, a3
+; RV64ZBB-NEXT: sll a1, a1, a2
+; RV64ZBB-NEXT: add a0, a0, a1
+; RV64ZBB-NEXT: ret
+ %maskedamt = and i64 %amt, 63
+ %1 = tail call i64 @llvm.fshr.i64(i64 %a, i64 %a, i64 %maskedamt)
+ %2 = shl i64 %b, %maskedamt
+ %3 = add i64 %1, %2
+ ret i64 %3
+}
+declare i64 @llvm.fshr.i64(i64, i64, i64)
More information about the llvm-commits
mailing list