Canonicalize header guards into a common format.

[oota-llvm.git] / lib / Target / R600 / EvergreenInstructions.td

diff --git a/lib/Target/R600/EvergreenInstructions.td b/lib/Target/R600/EvergreenInstructions.td
index 0d0b9c751a4f1f172a0ac0ac34680a4fa1f923d1..a83567a6acaa72fef2bc98ca596733b6d6f0b799 100644 (file)
--- a/lib/Target/R600/EvergreenInstructions.td
+++ b/lib/Target/R600/EvergreenInstructions.td
@@ -69,6 +69,7 @@ def EXP_IEEE_eg : EXP_IEEE_Common<0x81>;
 def LOG_IEEE_eg : LOG_IEEE_Common<0x83>;
 def RECIP_CLAMPED_eg : RECIP_CLAMPED_Common<0x84>;
 def RECIPSQRT_IEEE_eg : RECIPSQRT_IEEE_Common<0x89>;
+defm : RsqPat<RECIPSQRT_IEEE_eg, f32>;
 def SIN_eg : SIN_Common<0x8D>;
 def COS_eg : COS_Common<0x8E>;
 
@@ -256,6 +257,12 @@ def VTX_READ_GLOBAL_128_eg : VTX_READ_128_eg <1,
 
 let Predicates = [isEGorCayman] in {
 
+// Should be predicated on FeatureFP64
+// def FMA_64 : R600_3OP <
+//   0xA, "FMA_64",
+//   [(set f64:$dst, (fma f64:$src0, f64:$src1, f64:$src2))]
+// >;
+
 // BFE_UINT - bit_extract, an optimization for mask and shift
 // Src0 = Input
 // Src1 = Offset
@@ -312,6 +319,7 @@ def BIT_ALIGN_INT_eg : R600_3OP <0xC, "BIT_ALIGN_INT", [], VecALU>;
 def : ROTRPattern <BIT_ALIGN_INT_eg>;
 def MULADD_eg : MULADD_Common<0x14>;
 def MULADD_IEEE_eg : MULADD_IEEE_Common<0x18>;
+def FMA_eg : FMA_Common<0x7>;
 def ASHR_eg : ASHR_Common<0x15>;
 def LSHR_eg : LSHR_Common<0x16>;
 def LSHL_eg : LSHL_Common<0x17>;
@@ -326,7 +334,10 @@ def MUL_UINT24_eg : R600_2OP <0xB5, "MUL_UINT24",
 def DOT4_eg : DOT4_Common<0xBE>;
 defm CUBE_eg : CUBE_Common<0xC0>;
 
-def BCNT_INT : R600_1OP_Helper <0xAA, "BCNT_INT", ctpop>;
+def BCNT_INT : R600_1OP_Helper <0xAA, "BCNT_INT", ctpop, VecALU>;
+
+def FFBH_UINT : R600_1OP_Helper <0xAB, "FFBH_UINT", ctlz_zero_undef, VecALU>;
+def FFBL_INT : R600_1OP_Helper <0xAC, "FFBL_INT", cttz_zero_undef, VecALU>;
 
 let hasSideEffects = 1 in {
   def MOVA_INT_eg : R600_1OP <0xCC, "MOVA_INT", [], VecALU>;
@@ -348,7 +359,7 @@ def FLT_TO_UINT_eg : FLT_TO_UINT_Common<0x9A> {
 def UINT_TO_FLT_eg : UINT_TO_FLT_Common<0x9C>;
 
 def GROUP_BARRIER : InstR600 <
-    (outs), (ins), "  GROUP_BARRIER", [(int_AMDGPU_barrier_local)], AnyALU>,
+    (outs), (ins), "  GROUP_BARRIER", [(int_AMDGPU_barrier_local), (int_AMDGPU_barrier_global)], AnyALU>,
     R600ALU_Word0,
     R600ALU_Word1_OP2 <0x54> {
 
@@ -377,6 +388,11 @@ def GROUP_BARRIER : InstR600 <
   let ALUInst = 1;
 }
 
+def : Pat <
+       (int_AMDGPU_barrier_global),
+       (GROUP_BARRIER)
+>;
+
 //===----------------------------------------------------------------------===//
 // LDS Instructions
 //===----------------------------------------------------------------------===//