[PATCH] R600/SI: Add a VALU pattern for i64 xor
Tom Stellard
thomas.stellard at amd.com
Thu Jun 12 12:04:49 PDT 2014
---
lib/Target/R600/SIInstructions.td | 11 +++++++----
test/CodeGen/R600/xor.ll | 27 +++++++++++++++++++++++++++
2 files changed, 34 insertions(+), 4 deletions(-)
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index acdbc9b..7caf414 100644
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -1735,15 +1735,18 @@ def : Pat <
// VOP2 Patterns
//===----------------------------------------------------------------------===//
-def : Pat <
- (or i64:$src0, i64:$src1),
+class BinOp64Pat <SDNode node, Instruction inst> : Pat <
+ (node i64:$src0, i64:$src1),
(INSERT_SUBREG (INSERT_SUBREG (i64 (IMPLICIT_DEF)),
- (V_OR_B32_e32 (EXTRACT_SUBREG i64:$src0, sub0),
+ (inst (EXTRACT_SUBREG i64:$src0, sub0),
(EXTRACT_SUBREG i64:$src1, sub0)), sub0),
- (V_OR_B32_e32 (EXTRACT_SUBREG i64:$src0, sub1),
+ (inst (EXTRACT_SUBREG i64:$src0, sub1),
(EXTRACT_SUBREG i64:$src1, sub1)), sub1)
>;
+def : BinOp64Pat <or, V_OR_B32_e32>;
+def : BinOp64Pat <xor, V_XOR_B32_e32>;
+
class SextInReg <ValueType vt, int ShiftAmt> : Pat <
(sext_inreg i32:$src0, vt),
(V_ASHRREV_I32_e32 ShiftAmt, (V_LSHLREV_B32_e32 ShiftAmt, $src0))
diff --git a/test/CodeGen/R600/xor.ll b/test/CodeGen/R600/xor.ll
index 0043041..c60aba7 100644
--- a/test/CodeGen/R600/xor.ll
+++ b/test/CodeGen/R600/xor.ll
@@ -91,6 +91,7 @@ define void @vector_not_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in0, i32
ret void
}
+<<<<<<< HEAD
; SI-CHECK-LABEL: @vector_xor_i64
; SI-CHECK: V_XOR_B32_e32
; SI-CHECK: V_XOR_B32_e32
@@ -130,3 +131,29 @@ define void @vector_not_i64(i64 addrspace(1)* %out, i64 addrspace(1)* %in0, i64
store i64 %result, i64 addrspace(1)* %out
ret void
}
+
+; Test that we have a pattern to match xor inside a branch.
+; Note that in the future the backend may be smart enough to
+; use an SALU instruction for this.
+
+; SI-CHECK-LABEL: @xor_cf
+; SI-CHECK: V_XOR
+; SI-CHECK: V_XOR
+define void @xor_cf(i64 addrspace(1)* %out, i64 addrspace(1)* %in, i64 %a, i64 %b) {
+entry:
+ %0 = icmp eq i64 %a, 0
+ br i1 %0, label %if, label %else
+
+if:
+ %1 = xor i64 %a, %b
+ br label %endif
+
+else:
+ %2 = load i64 addrspace(1)* %in
+ br label %endif
+
+endif:
+ %3 = phi i64 [%1, %if], [%2, %else]
+ store i64 %3, i64 addrspace(1)* %out
+ ret void
+}
--
1.8.1.5
More information about the llvm-commits
mailing list