From 17a12a27ec1f0e44b4c379cac8fc9536074b4ed4 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@sifive.com>
Date: Fri, 15 Sep 2023 18:53:53 -0700
Subject: [PATCH] [RISCV] Add test case to show bad codegen for unaligned i64
 store of a large constant.

On the first split we create two i32 trunc stores and a srl to shift
the high part down. The srl gets constant folded, but to produce
a new i32 constant. But the truncstore for the low store still uses
the original constant.

This original constant then gets converted to a constant pool
before we revisit the stores to further split them. The constant
pool prevents further constant folding of the additional srls.

After legalization is done, we run DAGCombiner and get some constant
folding of srl via computeKnownBits which can peek through the constant
pool load. This can create new constants that also need a constant pool.
---
 .../CodeGen/RISCV/unaligned-load-store.ll     | 72 +++++++++++++++++++
 1 file changed, 72 insertions(+)

diff --git a/llvm/test/CodeGen/RISCV/unaligned-load-store.ll b/llvm/test/CodeGen/RISCV/unaligned-load-store.ll
index 429f0543b41b..5626c353f1cd 100644
--- a/llvm/test/CodeGen/RISCV/unaligned-load-store.ll
+++ b/llvm/test/CodeGen/RISCV/unaligned-load-store.ll
@@ -415,3 +415,75 @@ define void @merge_stores_i32_i64(ptr %p) {
   store i32 0, ptr %p2
   ret void
 }
+
+; FIXME: We shouldn't generate multiple constant pools entries with shifted
+; values.
+;.LCPI0_0:
+;        .quad   280223976814164                 # 0xfedcba987654
+;.LCPI0_1:
+;        .quad   71737338064426034               # 0xfedcba98765432
+;.LCPI0_2:
+;        .quad   -81985529216486896              # 0xfedcba9876543210
+define void @store_large_constant(ptr %x) {
+; RV32I-LABEL: store_large_constant:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    li a1, 254
+; RV32I-NEXT:    sb a1, 7(a0)
+; RV32I-NEXT:    li a1, 220
+; RV32I-NEXT:    sb a1, 6(a0)
+; RV32I-NEXT:    li a1, 186
+; RV32I-NEXT:    sb a1, 5(a0)
+; RV32I-NEXT:    li a1, 152
+; RV32I-NEXT:    sb a1, 4(a0)
+; RV32I-NEXT:    li a1, 118
+; RV32I-NEXT:    sb a1, 3(a0)
+; RV32I-NEXT:    li a1, 84
+; RV32I-NEXT:    sb a1, 2(a0)
+; RV32I-NEXT:    li a1, 50
+; RV32I-NEXT:    sb a1, 1(a0)
+; RV32I-NEXT:    li a1, 16
+; RV32I-NEXT:    sb a1, 0(a0)
+; RV32I-NEXT:    ret
+;
+; RV64I-LABEL: store_large_constant:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    li a1, 254
+; RV64I-NEXT:    sb a1, 7(a0)
+; RV64I-NEXT:    li a1, 220
+; RV64I-NEXT:    sb a1, 6(a0)
+; RV64I-NEXT:    li a1, 186
+; RV64I-NEXT:    sb a1, 5(a0)
+; RV64I-NEXT:    li a1, 152
+; RV64I-NEXT:    sb a1, 4(a0)
+; RV64I-NEXT:    li a1, 118
+; RV64I-NEXT:    lui a2, %hi(.LCPI16_0)
+; RV64I-NEXT:    ld a2, %lo(.LCPI16_0)(a2)
+; RV64I-NEXT:    lui a3, %hi(.LCPI16_1)
+; RV64I-NEXT:    ld a3, %lo(.LCPI16_1)(a3)
+; RV64I-NEXT:    lui a4, %hi(.LCPI16_2)
+; RV64I-NEXT:    ld a4, %lo(.LCPI16_2)(a4)
+; RV64I-NEXT:    sb a1, 3(a0)
+; RV64I-NEXT:    sb a2, 2(a0)
+; RV64I-NEXT:    sb a3, 1(a0)
+; RV64I-NEXT:    sb a4, 0(a0)
+; RV64I-NEXT:    ret
+;
+; RV32I-FAST-LABEL: store_large_constant:
+; RV32I-FAST:       # %bb.0:
+; RV32I-FAST-NEXT:    lui a1, 1043916
+; RV32I-FAST-NEXT:    addi a1, a1, -1384
+; RV32I-FAST-NEXT:    sw a1, 4(a0)
+; RV32I-FAST-NEXT:    lui a1, 484675
+; RV32I-FAST-NEXT:    addi a1, a1, 528
+; RV32I-FAST-NEXT:    sw a1, 0(a0)
+; RV32I-FAST-NEXT:    ret
+;
+; RV64I-FAST-LABEL: store_large_constant:
+; RV64I-FAST:       # %bb.0:
+; RV64I-FAST-NEXT:    lui a1, %hi(.LCPI16_0)
+; RV64I-FAST-NEXT:    ld a1, %lo(.LCPI16_0)(a1)
+; RV64I-FAST-NEXT:    sd a1, 0(a0)
+; RV64I-FAST-NEXT:    ret
+  store i64 18364758544493064720, ptr %x, align 1
+  ret void
+}