[llvm] r343030 - AMDGPU: Add Selection patterns to support add of one bit.

[Changpeng Fang via llvm-commits]

Author: chfang
Date: Tue Sep 25 14:21:18 2018
New Revision: 343030

URL: http://llvm.org/viewvc/llvm-project?rev=343030&view=rev
Log:
AMDGPU: Add Selection patterns to support add of one bit.

Summary:
  We generate s_xor to lower add of i1s in general cases, and s_not to
lower add with a one-bit imm of -1 (true).

Reviewers:
  rampitec

Differential Revision:
  https://reviews.llvm.org/D52518

Added:
    llvm/trunk/test/CodeGen/AMDGPU/add_i1.ll
Modified:
    llvm/trunk/lib/Target/AMDGPU/SIInstructions.td

Modified: llvm/trunk/lib/Target/AMDGPU/SIInstructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIInstructions.td?rev=343030&r1=343029&r2=343030&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInstructions.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIInstructions.td Tue Sep 25 14:21:18 2018
@@ -1323,6 +1323,18 @@ def : GCNPat <
 >;
 
 def : GCNPat <
+  (i1 (add i1:$src0, i1:$src1)),
+  (S_XOR_B64 $src0, $src1)
+>;
+
+let AddedComplexity = 1 in {
+def : GCNPat <
+  (i1 (add i1:$src0, (i1 -1))),
+  (S_NOT_B64 $src0)
+>;
+}
+
+def : GCNPat <
   (f16 (sint_to_fp i1:$src)),
   (V_CVT_F16_F32_e32 (V_CNDMASK_B32_e64 (i32 0), (i32 CONST.FP32_NEG_ONE), $src))
 >;

Added: llvm/trunk/test/CodeGen/AMDGPU/add_i1.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/add_i1.ll?rev=343030&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/add_i1.ll (added)
+++ llvm/trunk/test/CodeGen/AMDGPU/add_i1.ll Tue Sep 25 14:21:18 2018
@@ -0,0 +1,21 @@
+; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+
+; GCN-LABEL: {{^}}add_var_var_i1:
+; GCN: s_xor_b64
+define amdgpu_kernel void @add_var_var_i1(i1 addrspace(1)* %out, i1 addrspace(1)* %in0, i1 addrspace(1)* %in1) {
+  %a = load volatile i1, i1 addrspace(1)* %in0
+  %b = load volatile i1, i1 addrspace(1)* %in1
+  %add = add i1 %a, %b
+  store i1 %add, i1 addrspace(1)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}add_var_imm_i1:
+; GCN: s_not_b64
+define amdgpu_kernel void @add_var_imm_i1(i1 addrspace(1)* %out, i1 addrspace(1)* %in) {
+  %a = load volatile i1, i1 addrspace(1)* %in
+  %add = add i1 %a, 1
+  store i1 %add, i1 addrspace(1)* %out
+  ret void
+}