[llvm] r284743 - [AMDGPU] add fcopysign(f64, f32) pattern

[Valery Pykhtin via llvm-commits]

Author: vpykhtin
Date: Thu Oct 20 11:17:54 2016
New Revision: 284743

URL: http://llvm.org/viewvc/llvm-project?rev=284743&view=rev
Log:
[AMDGPU] add fcopysign(f64, f32) pattern

Differential revision: https://reviews.llvm.org/D25827

Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructions.td
    llvm/trunk/test/CodeGen/AMDGPU/fcopysign.f64.ll

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructions.td?rev=284743&r1=284742&r2=284743&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructions.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUInstructions.td Thu Oct 20 11:17:54 2016
@@ -540,6 +540,15 @@ multiclass BFIPatterns <Instruction BFI_
                (i32 (EXTRACT_SUBREG $src0, sub1)),
                (i32 (EXTRACT_SUBREG $src1, sub1))), sub1)
   >;
+
+  def : Pat <
+    (f64 (fcopysign f64:$src0, f32:$src1)),
+    (REG_SEQUENCE RC64,
+      (i32 (EXTRACT_SUBREG $src0, sub0)), sub0,
+      (BFI_INT (LoadImm32 0x7fffffff),
+               (i32 (EXTRACT_SUBREG $src0, sub1)),
+               $src1), sub1)
+  >;
 }
 
 // SHA-256 Ma patterns

Modified: llvm/trunk/test/CodeGen/AMDGPU/fcopysign.f64.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fcopysign.f64.ll?rev=284743&r1=284742&r2=284743&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fcopysign.f64.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fcopysign.f64.ll Thu Oct 20 11:17:54 2016
@@ -23,6 +23,22 @@ define void @test_copysign_f64(double ad
   ret void
 }
 
+; FUNC-LABEL: {{^}}test_copysign_f64_f32:
+; GCN-DAG: s_load_dwordx2 s{{\[}}[[SMAG_LO:[0-9]+]]:[[SMAG_HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}
+; GCN-DAG: s_load_dword s[[SSIGN:[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}
+; GCN-DAG: s_mov_b32 [[SCONST:s[0-9]+]], 0x7fffffff
+; GCN-DAG: v_mov_b32_e32 v[[VMAG_HI:[0-9]+]], s[[SMAG_HI]]
+; GCN-DAG: v_mov_b32_e32 v[[VSIGN:[0-9]+]], s[[SSIGN]]
+; GCN-DAG: v_bfi_b32 v[[VRESULT_HI:[0-9]+]], [[SCONST]], v[[VMAG_HI]], v[[VSIGN]]
+; GCN-DAG: v_mov_b32_e32 v[[VMAG_LO:[0-9]+]], s[[SMAG_LO]]
+; GCN: buffer_store_dwordx2 v{{\[}}[[VMAG_LO]]:[[VRESULT_HI]]{{\]}}
+define void @test_copysign_f64_f32(double addrspace(1)* %out, double %mag, float %sign) nounwind {
+  %c = fpext float %sign to double
+  %result = call double @llvm.copysign.f64(double %mag, double %c)
+  store double %result, double addrspace(1)* %out, align 8
+  ret void
+}
+
 ; FUNC-LABEL: {{^}}test_copysign_v2f64:
 ; GCN: s_endpgm
 define void @test_copysign_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %mag, <2 x double> %sign) nounwind {