[llvm] 2a7b8ab - [RISCV] Use add.uw for (or (and X, 0xFFFFFFFF), Y) if Y has zeroes in the lower 32 bits. (#65402)

[via llvm-commits]

Author: Craig Topper
Date: 2023-09-05T21:05:53-07:00
New Revision: 2a7b8ab07c26cd426a132278618191f53821299d

URL: https://github.com/llvm/llvm-project/commit/2a7b8ab07c26cd426a132278618191f53821299d
DIFF: https://github.com/llvm/llvm-project/commit/2a7b8ab07c26cd426a132278618191f53821299d.diff

LOG: [RISCV] Use add.uw for (or (and X, 0xFFFFFFFF), Y) if Y has zeroes in the lower 32 bits. (#65402)

Added: 
    

Modified: 
    llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
    llvm/test/CodeGen/RISCV/rv64zba.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
index 0cee7b9a1cebc7..a21c3d132636be 100644
--- a/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
+++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZb.td
@@ -739,6 +739,9 @@ def : Pat<(i64 (add (and GPR:$rs1, 0xFFFFFFFF), non_imm12:$rs2)),
           (ADD_UW GPR:$rs1, GPR:$rs2)>;
 def : Pat<(i64 (and GPR:$rs, 0xFFFFFFFF)), (ADD_UW GPR:$rs, (XLenVT X0))>;
 
+def : Pat<(i64 (or_is_add (and GPR:$rs1, 0xFFFFFFFF), non_imm12:$rs2)),
+          (ADD_UW GPR:$rs1, GPR:$rs2)>;
+
 def : Pat<(i64 (add (shl (and GPR:$rs1, 0xFFFFFFFF), (i64 1)), non_imm12:$rs2)),
           (SH1ADD_UW GPR:$rs1, GPR:$rs2)>;
 def : Pat<(i64 (add (shl (and GPR:$rs1, 0xFFFFFFFF), (i64 2)), non_imm12:$rs2)),

diff  --git a/llvm/test/CodeGen/RISCV/rv64zba.ll b/llvm/test/CodeGen/RISCV/rv64zba.ll
index 2f82d56eb9c050..9b472523875e79 100644
--- a/llvm/test/CodeGen/RISCV/rv64zba.ll
+++ b/llvm/test/CodeGen/RISCV/rv64zba.ll
@@ -1782,3 +1782,44 @@ define i64 @shl_31_sh3add(ptr %0, i32 signext %1) {
   %6 = load i64, ptr %5, align 8
   ret i64 %6
 }
+
+define i64 @pack_i64(i64 %a, i64 %b) nounwind {
+; RV64I-LABEL: pack_i64:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    slli a1, a1, 32
+; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: pack_i64:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    slli a1, a1, 32
+; RV64ZBA-NEXT:    add.uw a0, a0, a1
+; RV64ZBA-NEXT:    ret
+  %shl = and i64 %a, 4294967295
+  %shl1 = shl i64 %b, 32
+  %or = or i64 %shl1, %shl
+  ret i64 %or
+}
+
+define i64 @pack_i64_2(i32 signext %a, i32 signext %b) nounwind {
+; RV64I-LABEL: pack_i64_2:
+; RV64I:       # %bb.0:
+; RV64I-NEXT:    slli a0, a0, 32
+; RV64I-NEXT:    srli a0, a0, 32
+; RV64I-NEXT:    slli a1, a1, 32
+; RV64I-NEXT:    or a0, a1, a0
+; RV64I-NEXT:    ret
+;
+; RV64ZBA-LABEL: pack_i64_2:
+; RV64ZBA:       # %bb.0:
+; RV64ZBA-NEXT:    slli a1, a1, 32
+; RV64ZBA-NEXT:    add.uw a0, a0, a1
+; RV64ZBA-NEXT:    ret
+  %zexta = zext i32 %a to i64
+  %zextb = zext i32 %b to i64
+  %shl1 = shl i64 %zextb, 32
+  %or = or i64 %shl1, %zexta
+  ret i64 %or
+}