R600: Add target nodes for BFM and BFI

llvm-svn: 205235

5 files changed, 14 insertions, 3 deletions

diff --git a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
index 797063d0e10..2fbf6257241 100644
--- a/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -1176,6 +1176,8 @@ const char* AMDGPUTargetLowering::getTargetNodeName(unsigned Opcode) const {
   NODE_NAME_CASE(UMIN)
   NODE_NAME_CASE(BFE_U32)
   NODE_NAME_CASE(BFE_I32)
+  NODE_NAME_CASE(BFI)
+  NODE_NAME_CASE(BFM)
   NODE_NAME_CASE(URECIP)
   NODE_NAME_CASE(DOT4)
   NODE_NAME_CASE(EXPORT)
diff --git a/llvm/lib/Target/R600/AMDGPUISelLowering.h b/llvm/lib/Target/R600/AMDGPUISelLowering.h
index 48298f20979..a01961601de 100644
--- a/llvm/lib/Target/R600/AMDGPUISelLowering.h
+++ b/llvm/lib/Target/R600/AMDGPUISelLowering.h
@@ -186,6 +186,8 @@ enum {
   DOT4,
   BFE_U32, // Extract range of bits with zero extension to 32-bits.
   BFE_I32, // Extract range of bits with sign extension to 32-bits.
+  BFI, // (src0 & src1) | (~src0 & src2)
+  BFM, // Insert a range of bits into a 32-bit word.
   TEXTURE_FETCH,
   EXPORT,
   CONST_ADDRESS,
diff --git a/llvm/lib/Target/R600/AMDGPUInstrInfo.td b/llvm/lib/Target/R600/AMDGPUInstrInfo.td
index 2138bd23a36..69d80592cf8 100644
--- a/llvm/lib/Target/R600/AMDGPUInstrInfo.td
+++ b/llvm/lib/Target/R600/AMDGPUInstrInfo.td
@@ -89,4 +89,6 @@ def AMDGPUround : SDNode<"ISD::FROUND",
 
 def AMDGPUbfe_u32 : SDNode<"AMDGPUISD::BFE_U32", AMDGPUDTIntTernaryOp>;
 def AMDGPUbfe_i32 : SDNode<"AMDGPUISD::BFE_I32", AMDGPUDTIntTernaryOp>;
+def AMDGPUbfi : SDNode<"AMDGPUISD::BFI", AMDGPUDTIntTernaryOp>;
+def AMDGPUbfm : SDNode<"AMDGPUISD::BFM", SDTIntBinOp>;
 
diff --git a/llvm/lib/Target/R600/EvergreenInstructions.td b/llvm/lib/Target/R600/EvergreenInstructions.td
index 384b98eebe8..93151b4b51e 100644
--- a/llvm/lib/Target/R600/EvergreenInstructions.td
+++ b/llvm/lib/Target/R600/EvergreenInstructions.td
@@ -281,8 +281,11 @@ def BFE_INT_eg : R600_3OP <0x4, "BFE_INT",
 // XXX: This pattern is broken, disabling for now.  See comment in
 // AMDGPUInstructions.td for more info.
 //  def : BFEPattern <BFE_UINT_eg>;
+def BFI_INT_eg : R600_3OP <0x06, "BFI_INT",
+  [(set i32:$dst, (AMDGPUbfi i32:$src0, i32:$src1, i32:$src2))],
+  VecALU
+>;
 
-def BFI_INT_eg : R600_3OP <0x06, "BFI_INT", [], VecALU>;
 defm : BFIPatterns <BFI_INT_eg>;
 
 def MULADD_UINT24_eg : R600_3OP <0x10, "MULADD_UINT24",
diff --git a/llvm/lib/Target/R600/SIInstructions.td b/llvm/lib/Target/R600/SIInstructions.td
index 55d31eaa5bf..1a1676f8f56 100644
--- a/llvm/lib/Target/R600/SIInstructions.td
+++ b/llvm/lib/Target/R600/SIInstructions.td
@@ -989,7 +989,8 @@ defm V_XOR_B32 : VOP2_32 <0x0000001d, "V_XOR_B32", []>;
 
 } // End isCommutable = 1
 
-defm V_BFM_B32 : VOP2_32 <0x0000001e, "V_BFM_B32", []>;
+defm V_BFM_B32 : VOP2_32 <0x0000001e, "V_BFM_B32",
+  [(set i32:$dst, (AMDGPUbfm i32:$src0, i32:$src1))]>;
 defm V_MAC_F32 : VOP2_32 <0x0000001f, "V_MAC_F32", []>;
 defm V_MADMK_F32 : VOP2_32 <0x00000020, "V_MADMK_F32", []>;
 defm V_MADAK_F32 : VOP2_32 <0x00000021, "V_MADAK_F32", []>;
@@ -1064,7 +1065,8 @@ def V_BFE_I32 : VOP3_32 <0x00000149, "V_BFE_I32",
   [(set i32:$dst, (AMDGPUbfe_i32 i32:$src0, i32:$src1, i32:$src2))]>;
 }
 
-def V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32", []>;
+def V_BFI_B32 : VOP3_32 <0x0000014a, "V_BFI_B32",
+  [(set i32:$dst, (AMDGPUbfi i32:$src0, i32:$src1, i32:$src2))]>;
 defm : BFIPatterns <V_BFI_B32>;
 def V_FMA_F32 : VOP3_32 <0x0000014b, "V_FMA_F32",
   [(set f32:$dst, (fma f32:$src0, f32:$src1, f32:$src2))]