Merging r231259:

------------------------------------------------------------------------
r231259 | marek.olsak | 2015-03-04 12:33:45 -0500 (Wed, 04 Mar 2015) | 4 lines

R600/SI: Add an intrinsic for S_FLBIT_I32 / V_FFBH_I32

Required by OpenGL (ARB_gpu_shader5).

------------------------------------------------------------------------

git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_36@236263 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/AMDGPUIntrinsics.td b/lib/Target/R600/AMDGPUIntrinsics.td
index eee9c29038d05a70e06199eeb416d5777ec1ba64..ab489cd2a4abe4f21460d345063f08360b08f0bd 100644 (file)
--- a/lib/Target/R600/AMDGPUIntrinsics.td
+++ b/lib/Target/R600/AMDGPUIntrinsics.td
@@ -68,6 +68,7 @@ let TargetPrefix = "AMDGPU", isTarget = 1 in {
   def int_AMDGPU_bfe_u32 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_bfm : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_brev : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>;
+  def int_AMDGPU_flbit_i32 : Intrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>;
   def int_AMDGPU_barrier_local  : Intrinsic<[], [], []>;
   def int_AMDGPU_barrier_global  : Intrinsic<[], [], []>;
 }

diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
index 5d05f06c94b41f7fa0d8a84216a19f7fb6d249f6..5ab33b491ce33c022186431460e6fc2040f9ec17 100644 (file)
--- a/lib/Target/R600/SIInstrInfo.cpp
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -1295,6 +1295,7 @@ unsigned SIInstrInfo::getVALUOp(const MachineInstr &MI) {
   case AMDGPU::S_BCNT1_I32_B32: return AMDGPU::V_BCNT_U32_B32_e64;
   case AMDGPU::S_FF1_I32_B32: return AMDGPU::V_FFBL_B32_e32;
   case AMDGPU::S_FLBIT_I32_B32: return AMDGPU::V_FFBH_U32_e32;
+  case AMDGPU::S_FLBIT_I32: return AMDGPU::V_FFBH_I32_e64;
   }
 }
 

diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td
index cc51e61c203a01387905587f0eba6e3e9c591cfe..ee4bb4c13411c6c1a6fa86574d879c7bcbda2ad5 100644 (file)
--- a/lib/Target/R600/SIInstructions.td
+++ b/lib/Target/R600/SIInstructions.td
@@ -152,9 +152,11 @@ defm S_FLBIT_I32_B32 : SOP1_32 <sop1<0x15, 0x12>, "s_flbit_i32_b32",
   [(set i32:$dst, (ctlz_zero_undef i32:$src0))]
 >;
 
-//defm S_FLBIT_I32_B64 : SOP1_32 <sop1<0x16, 0x13>, "s_flbit_i32_b64", []>;
-defm S_FLBIT_I32 : SOP1_32 <sop1<0x17, 0x14>, "s_flbit_i32", []>;
-//defm S_FLBIT_I32_I64 : SOP1_32 <sop1<0x18, 0x15>, "s_flbit_i32_i64", []>;
+defm S_FLBIT_I32_B64 : SOP1_32_64 <sop1<0x16, 0x13>, "s_flbit_i32_b64", []>;
+defm S_FLBIT_I32 : SOP1_32 <sop1<0x17, 0x14>, "s_flbit_i32",
+  [(set i32:$dst, (int_AMDGPU_flbit_i32 i32:$src0))]
+>;
+defm S_FLBIT_I32_I64 : SOP1_32_64 <sop1<0x18, 0x15>, "s_flbit_i32_i64", []>;
 defm S_SEXT_I32_I8 : SOP1_32 <sop1<0x19, 0x16>, "s_sext_i32_i8",
   [(set i32:$dst, (sext_inreg i32:$src0, i8))]
 >;

diff --git a/test/CodeGen/R600/llvm.AMDGPU.flbit.i32.ll b/test/CodeGen/R600/llvm.AMDGPU.flbit.i32.ll
new file mode 100644 (file)
index 0000000..19fbee8
--- /dev/null
+++ b/test/CodeGen/R600/llvm.AMDGPU.flbit.i32.ll
@@ -0,0 +1,28 @@
+; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
+
+declare i32 @llvm.AMDGPU.flbit.i32(i32) nounwind readnone
+
+; FUNC-LABEL: {{^}}s_flbit:
+; SI: s_load_dword [[VAL:s[0-9]+]],
+; SI: s_flbit_i32 [[SRESULT:s[0-9]+]], [[VAL]]
+; SI: v_mov_b32_e32 [[VRESULT:v[0-9]+]], [[SRESULT]]
+; SI: buffer_store_dword [[VRESULT]],
+; SI: s_endpgm
+define void @s_flbit(i32 addrspace(1)* noalias %out, i32 %val) nounwind {
+  %r = call i32 @llvm.AMDGPU.flbit.i32(i32 %val) nounwind readnone
+  store i32 %r, i32 addrspace(1)* %out, align 4
+  ret void
+}
+
+; FUNC-LABEL: {{^}}v_flbit:
+; SI: buffer_load_dword [[VAL:v[0-9]+]],
+; SI: v_ffbh_i32_e32 [[RESULT:v[0-9]+]], [[VAL]]
+; SI: buffer_store_dword [[RESULT]],
+; SI: s_endpgm
+define void @v_flbit(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
+  %val = load i32 addrspace(1)* %valptr, align 4
+  %r = call i32 @llvm.AMDGPU.flbit.i32(i32 %val) nounwind readnone
+  store i32 %r, i32 addrspace(1)* %out, align 4
+  ret void
+}