[llvm] r297873 - AMDGPU: Fix unnecessary ands when packing f16 vectors

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Wed Mar 15 12:04:27 PDT 2017


Author: arsenm
Date: Wed Mar 15 14:04:26 2017
New Revision: 297873

URL: http://llvm.org/viewvc/llvm-project?rev=297873&view=rev
Log:
AMDGPU: Fix unnecessary ands when packing f16 vectors

computeKnownBits didn't handle fp_to_fp16 to report
the high bits as 0. ARM maps the generic node to an instruction
that does not modify the high bits of the register, so introduce
a target node where the high bits are known 0.

Modified:
    llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
    llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h
    llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td
    llvm/trunk/lib/Target/AMDGPU/EvergreenInstructions.td
    llvm/trunk/lib/Target/AMDGPU/SIInstructions.td
    llvm/trunk/lib/Target/AMDGPU/VOP1Instructions.td
    llvm/trunk/test/CodeGen/AMDGPU/fabs.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/fadd.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/fmul.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/fptrunc.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/fsub.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.ceil.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.cos.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.floor.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.fma.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.fmuladd.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.log2.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.minnum.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.rint.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.sin.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.sqrt.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/llvm.trunc.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/select.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/sitofp.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/uitofp.f16.ll
    llvm/trunk/test/CodeGen/AMDGPU/v_mac_f16.ll

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.cpp Wed Mar 15 14:04:26 2017
@@ -271,6 +271,7 @@ AMDGPUTargetLowering::AMDGPUTargetLoweri
 
   setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
   setOperationAction(ISD::FP_TO_FP16, MVT::f64, Custom);
+  setOperationAction(ISD::FP_TO_FP16, MVT::f32, Custom);
 
   const MVT ScalarIntVTs[] = { MVT::i32, MVT::i64 };
   for (MVT VT : ScalarIntVTs) {
@@ -2100,15 +2101,19 @@ SDValue AMDGPUTargetLowering::LowerFP64_
 }
 
 SDValue AMDGPUTargetLowering::LowerFP_TO_FP16(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc DL(Op);
+  SDValue N0 = Op.getOperand(0);
+
+  // Convert to target node to get known bits
+  if (N0.getValueType() == MVT::f32)
+    return DAG.getNode(AMDGPUISD::FP_TO_FP16, DL, Op.getValueType(), N0);
 
   if (getTargetMachine().Options.UnsafeFPMath) {
     // There is a generic expand for FP_TO_FP16 with unsafe fast math.
     return SDValue();
   }
 
-  SDLoc DL(Op);
-  SDValue N0 = Op.getOperand(0);
-  assert (N0.getSimpleValueType() == MVT::f64);
+  assert(N0.getSimpleValueType() == MVT::f64);
 
   // f64 -> f16 conversion using round-to-nearest-even rounding mode.
   const unsigned ExpMask = 0x7ff;
@@ -3478,6 +3483,7 @@ const char* AMDGPUTargetLowering::getTar
   NODE_NAME_CASE(CVT_F32_UBYTE2)
   NODE_NAME_CASE(CVT_F32_UBYTE3)
   NODE_NAME_CASE(CVT_PKRTZ_F16_F32)
+  NODE_NAME_CASE(FP_TO_FP16)
   NODE_NAME_CASE(BUILD_VERTICAL_VECTOR)
   NODE_NAME_CASE(CONST_DATA_PTR)
   NODE_NAME_CASE(PC_ADD_REL_OFFSET)
@@ -3577,6 +3583,13 @@ void AMDGPUTargetLowering::computeKnownB
 
     break;
   }
+  case AMDGPUISD::FP_TO_FP16: {
+    unsigned BitWidth = KnownZero.getBitWidth();
+
+    // High bits are zero.
+    KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - 16);
+    break;
+  }
   }
 }
 

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUISelLowering.h Wed Mar 15 14:04:26 2017
@@ -326,6 +326,10 @@ enum NodeType : unsigned {
   // with round to zero.
   CVT_PKRTZ_F16_F32,
 
+  // Same as the standard node, except the high bits of the resulting integer
+  // are known 0.
+  FP_TO_FP16,
+
   /// This node is for VLIW targets and it is used to represent a vector
   /// that is stored in consecutive registers with the same channel.
   /// For example:

Modified: llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/AMDGPUInstrInfo.td Wed Mar 15 14:04:26 2017
@@ -83,6 +83,8 @@ def AMDGPUrsq_clamp : SDNode<"AMDGPUISD:
 def AMDGPUldexp : SDNode<"AMDGPUISD::LDEXP", AMDGPULdExpOp>;
 
 def AMDGPUpkrtz_f16_f32 : SDNode<"AMDGPUISD::CVT_PKRTZ_F16_F32", AMDGPUFPPackOp>;
+def AMDGPUfp_to_f16 : SDNode<"AMDGPUISD::FP_TO_FP16" , SDTFPToIntOp>;
+
 
 def AMDGPUfp_class : SDNode<"AMDGPUISD::FP_CLASS", AMDGPUFPClassOp>;
 

Modified: llvm/trunk/lib/Target/AMDGPU/EvergreenInstructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/EvergreenInstructions.td?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/EvergreenInstructions.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/EvergreenInstructions.td Wed Mar 15 14:04:26 2017
@@ -438,7 +438,7 @@ defm CUBE_eg : CUBE_Common<0xC0>;
 def ADDC_UINT : R600_2OP_Helper <0x52, "ADDC_UINT", AMDGPUcarry>;
 def SUBB_UINT : R600_2OP_Helper <0x53, "SUBB_UINT", AMDGPUborrow>;
 
-def FLT32_TO_FLT16 : R600_1OP_Helper <0xA2, "FLT32_TO_FLT16", fp_to_f16, VecALU>;
+def FLT32_TO_FLT16 : R600_1OP_Helper <0xA2, "FLT32_TO_FLT16", AMDGPUfp_to_f16, VecALU>;
 def FLT16_TO_FLT32 : R600_1OP_Helper <0xA3, "FLT16_TO_FLT32", f16_to_fp, VecALU>;
 def BCNT_INT : R600_1OP_Helper <0xAA, "BCNT_INT", ctpop, VecALU>;
 def FFBH_UINT : R600_1OP_Helper <0xAB, "FFBH_UINT", AMDGPUffbh_u32, VecALU>;

Modified: llvm/trunk/lib/Target/AMDGPU/SIInstructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/SIInstructions.td?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/SIInstructions.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/SIInstructions.td Wed Mar 15 14:04:26 2017
@@ -467,7 +467,7 @@ def : Pat <
 
 // fp_to_fp16 patterns
 def : Pat <
-  (i32 (fp_to_f16 (f32 (VOP3Mods0 f32:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)))),
+  (i32 (AMDGPUfp_to_f16 (f32 (VOP3Mods0 f32:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)))),
   (V_CVT_F16_F32_e64 $src0_modifiers, f32:$src0, $clamp, $omod)
 >;
 

Modified: llvm/trunk/lib/Target/AMDGPU/VOP1Instructions.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AMDGPU/VOP1Instructions.td?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AMDGPU/VOP1Instructions.td (original)
+++ llvm/trunk/lib/Target/AMDGPU/VOP1Instructions.td Wed Mar 15 14:04:26 2017
@@ -328,7 +328,7 @@ def : Pat<
 >;
 
 def : Pat<
-    (i16 (fp_to_f16 f32:$src)),
+    (i16 (AMDGPUfp_to_f16 f32:$src)),
     (V_CVT_F16_F32_e32 $src)
 >;
 

Modified: llvm/trunk/test/CodeGen/AMDGPU/fabs.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fabs.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fabs.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fabs.f16.ll Wed Mar 15 14:04:26 2017
@@ -123,8 +123,8 @@ define void @fabs_free_v2f16(<2 x half>
 ; CI: v_cvt_f32_f16_e32
 ; CI: v_cvt_f32_f16_e32
 ; CI: v_mul_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
-; CI: v_mul_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
 ; CI: v_cvt_f16_f32
+; CI: v_mul_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
 ; CI: v_cvt_f16_f32
 
 ; VI: v_lshrrev_b32_e32 v{{[0-9]+}}, 16,

Modified: llvm/trunk/test/CodeGen/AMDGPU/fadd.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fadd.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fadd.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fadd.f16.ll Wed Mar 15 14:04:26 2017
@@ -59,24 +59,28 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}fadd_v2f16
+; GCN-LABEL: {{^}}fadd_v2f16:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
-; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
+; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_add_f32_e32 v[[R_F32_0:[0-9]+]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
 ; SI:  v_add_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI:  v_add_f16_e32 v[[R_F16_0:[0-9]+]], v[[B_V2_F16]], v[[A_V2_F16]]
 ; VI:  v_add_f16_e32 v[[R_F16_1:[0-9]+]], v[[B_F16_1]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fadd_v2f16(
@@ -91,7 +95,7 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}fadd_v2f16_imm_a
+; GCN-LABEL: {{^}}fadd_v2f16_imm_a:
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
 ; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
@@ -103,9 +107,8 @@ entry:
 
 ; VI:  v_add_f16_e32 v[[R_F16_0:[0-9]+]], 1.0, v[[B_V2_F16]]
 ; VI:  v_add_f16_e32 v[[R_F16_1:[0-9]+]], 2.0, v[[B_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fadd_v2f16_imm_a(
@@ -118,7 +121,7 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}fadd_v2f16_imm_b
+; GCN-LABEL: {{^}}fadd_v2f16_imm_b:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
 ; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
@@ -129,9 +132,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_add_f16_e32 v[[R_F16_0:[0-9]+]], 2.0, v[[A_V2_F16]]
 ; VI:  v_add_f16_e32 v[[R_F16_1:[0-9]+]], 1.0, v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fadd_v2f16_imm_b(

Modified: llvm/trunk/test/CodeGen/AMDGPU/fmul.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fmul.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fmul.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fmul.f16.ll Wed Mar 15 14:04:26 2017
@@ -60,24 +60,27 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_v2f16
+; GCN-LABEL: {{^}}fmul_v2f16:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_mul_f32_e32 v[[R_F32_0:[0-9]+]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
 ; SI:  v_mul_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI:  v_mul_f16_e32 v[[R_F16_0:[0-9]+]], v[[B_V2_F16]], v[[A_V2_F16]]
 ; VI:  v_mul_f16_e32 v[[R_F16_1:[0-9]+]], v[[B_F16_1]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fmul_v2f16(
@@ -92,7 +95,7 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_v2f16_imm_a
+; GCN-LABEL: {{^}}fmul_v2f16_imm_a:
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
 ; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
@@ -103,9 +106,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_mul_f16_e32 v[[R_F16_0:[0-9]+]], 0x4200, v[[B_V2_F16]]
 ; VI:  v_mul_f16_e32 v[[R_F16_1:[0-9]+]], 4.0, v[[B_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fmul_v2f16_imm_a(
@@ -118,7 +120,7 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}fmul_v2f16_imm_b
+; GCN-LABEL: {{^}}fmul_v2f16_imm_b:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
 ; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
@@ -129,9 +131,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_mul_f16_e32 v[[R_F16_0:[0-9]+]], 4.0, v[[A_V2_F16]]
 ; VI:  v_mul_f16_e32 v[[R_F16_1:[0-9]+]], 0x4200, v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fmul_v2f16_imm_b(

Modified: llvm/trunk/test/CodeGen/AMDGPU/fptrunc.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fptrunc.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fptrunc.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fptrunc.f16.ll Wed Mar 15 14:04:26 2017
@@ -38,9 +38,8 @@ entry:
 ; GCN:     buffer_load_dwordx2 v{{\[}}[[A_F32_0:[0-9]+]]:[[A_F32_1:[0-9]+]]{{\]}}
 ; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[A_F32_0]]
 ; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
-; SIVI-DAG: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; SIVI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; SIVI:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; SIVI:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GFX9-FLUSH: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GFX9-FLUSH: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]
@@ -61,14 +60,13 @@ entry:
 
 ; GCN-LABEL: {{^}}fptrunc_v2f64_to_v2f16:
 ; GCN: buffer_load_dwordx4 v{{\[}}[[A_F64_0:[0-9]+]]:[[A_F64_3:[0-9]+]]{{\]}}
-; GCN: v_cvt_f32_f64_e32 v[[A_F32_0:[0-9]+]], v{{\[}}[[A_F64_0]]:{{[0-9]+}}{{\]}}
-; GCN: v_cvt_f32_f64_e32 v[[A_F32_1:[0-9]+]], v{{\[}}{{[0-9]+}}:[[A_F64_3]]{{\]}}
-; GCN: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[A_F32_0]]
-; GCN: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
+; GCN-DAG: v_cvt_f32_f64_e32 v[[A_F32_0:[0-9]+]], v{{\[}}[[A_F64_0]]:{{[0-9]+}}{{\]}}
+; GCN-DAG: v_cvt_f32_f64_e32 v[[A_F32_1:[0-9]+]], v{{\[}}{{[0-9]+}}:[[A_F64_3]]{{\]}}
+; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[A_F32_0]]
+; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[A_F32_1]]
 
-; SIVI: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; SIVI: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GFX9-FLUSH: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GFX9-FLUSH: v_lshl_or_b32 v[[R_V2_F16:[0-9]+]], v[[R_F16_1]], 16, v[[R_F16_LO]]

Modified: llvm/trunk/test/CodeGen/AMDGPU/fsub.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/fsub.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/fsub.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/fsub.f16.ll Wed Mar 15 14:04:26 2017
@@ -63,24 +63,27 @@ entry:
 ; GCN-LABEL: {{^}}fsub_v2f16:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
-; SIVI-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; SIVI-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; SI: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI: v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
-; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_subrev_f32_e32 v[[R_F32_0:[0-9]+]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
 ; SI:  v_subrev_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+
+
+; VI-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 
 ; VI-DAG: v_subrev_f16_e32 v[[R_F16_0:[0-9]+]], v[[B_V2_F16]], v[[A_V2_F16]]
 ; VI: v_subrev_f16_e32 v[[R_F16_1:[0-9]+]], v[[B_F16_1]], v[[A_F16_1]]
 
-; SIVI: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; SIVI: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GFX9: v_pk_add_f16 v[[R_V2_F16:[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]] neg_lo:[0,1] neg_hi:[0,1]
 
@@ -112,9 +115,8 @@ entry:
 ; VI:  v_sub_f16_e32 v[[R_F16_0:[0-9]+]], 1.0, v[[B_V2_F16]]
 ; VI:  v_sub_f16_e32 v[[R_F16_1:[0-9]+]], 2.0, v[[B_F16_1]]
 
-; SIVI: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; SIVI: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GFX9: s_mov_b32 [[K:s[0-9]+]], 0x40003c00
 ; GFX9: v_pk_add_f16 v[[R_V2_F16:[0-9]+]], [[K]], v[[B_V2_F16]] neg_lo:[0,1] neg_hi:[0,1]
@@ -144,9 +146,8 @@ entry:
 ; VI:  v_add_f16_e32 v[[R_F16_0:[0-9]+]], -2.0, v[[A_V2_F16]]
 ; VI:  v_add_f16_e32 v[[R_F16_1:[0-9]+]], -1.0, v[[A_F16_1]]
 
-; SIVI: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; SIVI: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; SIVI: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GFX9: s_mov_b32 [[K:s[0-9]+]], 0xbc00c000
 ; GFX9: v_pk_add_f16 v[[R_V2_F16:[0-9]+]], [[K]], v[[A_V2_F16]]{{$}}

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.ceil.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.ceil.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.ceil.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.ceil.f16.ll Wed Mar 15 14:04:26 2017
@@ -4,7 +4,7 @@
 declare half @llvm.ceil.f16(half %a)
 declare <2 x half> @llvm.ceil.v2f16(<2 x half> %a)
 
-; GCN-LABEL: {{^}}ceil_f16
+; GCN-LABEL: {{^}}ceil_f16:
 ; GCN: buffer_load_ushort v[[A_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32:[0-9]+]], v[[A_F16]]
 ; SI:  v_ceil_f32_e32 v[[R_F32:[0-9]+]], v[[A_F32]]
@@ -22,7 +22,7 @@ entry:
   ret void
 }
 
-; GCN-LABEL: {{^}}ceil_v2f16
+; GCN-LABEL: {{^}}ceil_v2f16:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
 ; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
@@ -33,9 +33,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_ceil_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_ceil_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @ceil_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.cos.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.cos.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.cos.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.cos.f16.ll Wed Mar 15 14:04:26 2017
@@ -25,23 +25,32 @@ entry:
 
 ; GCN-LABEL: {{^}}cos_v2f16
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
-; SI:  v_mov_b32_e32 v[[HALF_PIE:[0-9]+]], 0x3e22f983{{$}}
-; GCN: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
-; SI:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], v[[HALF_PIE]], v[[A_F32_0]]
-; VI:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], 0.15915494, v[[A_F32_0]]
-; GCN: v_fract_f32_e32 v[[F_F32_0:[0-9]+]], v[[M_F32_0]]
-; SI:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], v[[HALF_PIE]], v[[A_F32_1]]
-; VI:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], 0.15915494, v[[A_F32_1]]
-; GCN: v_fract_f32_e32 v[[F_F32_1:[0-9]+]], v[[M_F32_1]]
-; GCN: v_cos_f32_e32 v[[R_F32_0:[0-9]+]], v[[F_F32_0]]
-; GCN: v_cos_f32_e32 v[[R_F32_1:[0-9]+]], v[[F_F32_1]]
-; GCN: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; GCN: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; SI-DAG:  v_mov_b32_e32 v[[HALF_PIE:[0-9]+]], 0x3e22f983{{$}}
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI-DAG:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], v[[HALF_PIE]], v[[A_F32_0]]
+; SI-DAG:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], v[[HALF_PIE]], v[[A_F32_1]]
+
+
+; VI-DAG: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; VI-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; VI-DAG:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], 0.15915494, v[[A_F32_0]]
+; VI-DAG:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], 0.15915494, v[[A_F32_1]]
+
+; GCN-DAG: v_fract_f32_e32 v[[F_F32_0:[0-9]+]], v[[M_F32_0]]
+; GCN-DAG: v_fract_f32_e32 v[[F_F32_1:[0-9]+]], v[[M_F32_1]]
+; GCN-DAG: v_cos_f32_e32 v[[R_F32_0:[0-9]+]], v[[F_F32_0]]
+; GCN-DAG: v_cos_f32_e32 v[[R_F32_1:[0-9]+]], v[[F_F32_1]]
+
+; FIXME: Remove and for VI+
+; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; VI: v_and_b32_e32 v[[R_F16_0]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @cos_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.f16.ll Wed Mar 15 14:04:26 2017
@@ -33,9 +33,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_exp_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_exp_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @exp2_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.exp2.ll Wed Mar 15 14:04:26 2017
@@ -1,6 +1,6 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG --check-prefix=FUNC
 ;RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=CM --check-prefix=FUNC
-;RUN: llc < %s -march=amdgcn -mcpu=SI | FileCheck %s --check-prefix=SI --check-prefix=FUNC
+;RUN: llc < %s -march=amdgcn | FileCheck %s --check-prefix=SI --check-prefix=FUNC
 ;RUN: llc < %s -march=amdgcn -mcpu=tonga | FileCheck %s --check-prefix=SI --check-prefix=FUNC
 
 ;FUNC-LABEL: {{^}}test:

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.floor.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.floor.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.floor.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.floor.f16.ll Wed Mar 15 14:04:26 2017
@@ -33,9 +33,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_floor_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_floor_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @floor_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.fma.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.fma.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.fma.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.fma.f16.ll Wed Mar 15 14:04:26 2017
@@ -103,24 +103,31 @@ define void @fma_f16_imm_c(
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[C_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+
+; SI: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI: v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
+; SI: v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
-; SI:  v_fma_f32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]], v[[B_F32_0]], v[[C_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]], v[[B_F32_1]], v[[C_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
-; VI:  v_fma_f16 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]], v[[C_V2_F16]]
-; VI:  v_fma_f16 v[[R_F16_1:[0-9]+]], v[[A_F16_1]], v[[B_F16_1]], v[[C_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; SI-DAG:  v_fma_f32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]], v[[B_F32_0]], v[[C_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]], v[[B_F32_1]], v[[C_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+; VI: v_fma_f16 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]], v[[C_V2_F16]]
+; VI: v_fma_f16 v[[R_F16_1:[0-9]+]], v[[A_F16_1]], v[[B_F16_1]], v[[C_F16_1]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fma_v2f16(
@@ -136,26 +143,26 @@ define void @fma_v2f16(
   ret void
 }
 
-; GCN-LABEL: {{^}}fma_v2f16_imm_a
+; GCN-LABEL: {{^}}fma_v2f16_imm_a:
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[C_V2_F16:[0-9]+]]
 ; SI:  v_mov_b32_e32 v[[A_F32:[0-9]+]], 0x40400000{{$}}
 ; VI:  v_mov_b32_e32 v[[A_F16:[0-9]+]], 0x4200{{$}}
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+; GCN-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; GCN-DAG: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
 ; SI:  v_fma_f32 v[[R_F32_0:[0-9]+]], v[[B_F32_0]], v[[A_F32]], v[[C_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32]], v[[C_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32]], v[[C_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
 ; VI:  v_fma_f16 v[[R_F16_0:[0-9]+]], v[[B_V2_F16]], v[[A_F16]], v[[C_V2_F16]]
 ; VI:  v_fma_f16 v[[R_F16_1:[0-9]+]], v[[B_F16_1]], v[[A_F16]], v[[C_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fma_v2f16_imm_a(
@@ -169,26 +176,31 @@ define void @fma_v2f16_imm_a(
   ret void
 }
 
-; GCN-LABEL: {{^}}fma_v2f16_imm_b
+; GCN-LABEL: {{^}}fma_v2f16_imm_b:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[C_V2_F16:[0-9]+]]
 ; SI:  v_mov_b32_e32 v[[B_F32:[0-9]+]], 0x40400000{{$}}
 ; VI:  v_mov_b32_e32 v[[B_F16:[0-9]+]], 0x4200{{$}}
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
-; SI:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
-; SI:  v_fma_f32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]], v[[B_F32]], v[[C_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]], v[[B_F32]], v[[C_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
+; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
+; SI-DAG:  v_fma_f32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]], v[[B_F32]], v[[C_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]], v[[B_F32]], v[[C_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
 ; VI:  v_fma_f16 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]], v[[B_F16]], v[[C_V2_F16]]
 ; VI:  v_fma_f16 v[[R_F16_1:[0-9]+]], v[[A_F16_1]], v[[B_F16]], v[[C_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fma_v2f16_imm_b(
@@ -202,26 +214,31 @@ define void @fma_v2f16_imm_b(
   ret void
 }
 
-; GCN-LABEL: {{^}}fma_v2f16_imm_c
+; GCN-LABEL: {{^}}fma_v2f16_imm_c:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; SI:  v_mov_b32_e32 v[[C_F32:[0-9]+]], 0x40400000{{$}}
 ; VI:  v_mov_b32_e32 v[[C_F16:[0-9]+]], 0x4200{{$}}
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_fma_f32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]], v[[B_F32_0]], v[[C_F32]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]], v[[B_F32_1]], v[[C_F32]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_fma_f32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]], v[[B_F32_1]], v[[C_F32]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI:  v_fma_f16 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]], v[[B_V2_F16]], v[[C_F16]]
 ; VI:  v_fma_f16 v[[R_F16_1:[0-9]+]], v[[A_F16_1]], v[[B_F16_1]], v[[C_F16]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @fma_v2f16_imm_c(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.fmuladd.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.fmuladd.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.fmuladd.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.fmuladd.f16.ll Wed Mar 15 14:04:26 2017
@@ -98,22 +98,28 @@ define void @fmuladd_f16_imm_b(
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[C_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+; SI-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; SI-DAG: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
 ; SI:  v_mac_f32_e32 v[[C_F32_0]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[C_F32_0]]
 ; SI:  v_mac_f32_e32 v[[C_F32_1]], v[[B_F32_1]], v[[A_F32_1]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[C_F32_1]]
-; SI:  v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; SI:  v_cvt_f16_f32_e32 v[[R_F16_LO:[0-9]+]], v[[C_F32_0]]
 ; SI:  v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
 
 ; FIXME: and should be unnecessary
 ; VI-FLUSH: v_mac_f16_e32 v[[C_V2_F16]], v[[B_V2_F16]], v[[A_V2_F16]]

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.log2.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.log2.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.log2.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.log2.f16.ll Wed Mar 15 14:04:26 2017
@@ -25,17 +25,16 @@ entry:
 ; GCN-LABEL: {{^}}log2_v2f16
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
-; SI:  v_log_f32_e32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]]
-; SI:  v_log_f32_e32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI-DAG:  v_log_f32_e32 v[[R_F32_0:[0-9]+]], v[[A_F32_0]]
+; SI-DAG:  v_log_f32_e32 v[[R_F32_1:[0-9]+]], v[[A_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_log_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_log_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @log2_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.maxnum.f16.ll Wed Mar 15 14:04:26 2017
@@ -65,21 +65,26 @@ entry:
 ; GCN-LABEL: {{^}}maxnum_v2f16:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_max_f32_e32 v[[R_F32_0:[0-9]+]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_max_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_max_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI:  v_max_f16_e32 v[[R_F16_0:[0-9]+]], v[[B_V2_F16]], v[[A_V2_F16]]
 ; VI:  v_max_f16_e32 v[[R_F16_1:[0-9]+]], v[[B_F16_1]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @maxnum_v2f16(
@@ -105,9 +110,9 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_max_f16_e32 v[[R_F16_0:[0-9]+]], 0x4200, v[[B_V2_F16]]
 ; VI:  v_max_f16_e32 v[[R_F16_1:[0-9]+]], 4.0, v[[B_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @maxnum_v2f16_imm_a(
@@ -131,9 +136,9 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_max_f16_e32 v[[R_F16_0:[0-9]+]], 4.0, v[[A_V2_F16]]
 ; VI:  v_max_f16_e32 v[[R_F16_1:[0-9]+]], 0x4200, v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @maxnum_v2f16_imm_b(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.minnum.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.minnum.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.minnum.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.minnum.f16.ll Wed Mar 15 14:04:26 2017
@@ -65,21 +65,25 @@ entry:
 ; GCN-LABEL: {{^}}minnum_v2f16:
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_min_f32_e32 v[[R_F32_0:[0-9]+]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_min_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_min_f32_e32 v[[R_F32_1:[0-9]+]], v[[B_F32_1]], v[[A_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI:  v_min_f16_e32 v[[R_F16_0:[0-9]+]], v[[B_V2_F16]], v[[A_V2_F16]]
 ; VI:  v_min_f16_e32 v[[R_F16_1:[0-9]+]], v[[B_F16_1]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @minnum_v2f16(
@@ -96,18 +100,21 @@ entry:
 
 ; GCN-LABEL: {{^}}minnum_v2f16_imm_a:
 ; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_min_f32_e32 v[[R_F32_0:[0-9]+]], 0x40400000, v[[B_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; SI:  v_min_f32_e32 v[[R_F32_1:[0-9]+]], 4.0, v[[B_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+; SI-DAG:  v_min_f32_e32 v[[R_F32_1:[0-9]+]], 4.0, v[[B_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
+
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; VI:  v_min_f16_e32 v[[R_F16_0:[0-9]+]], 0x4200, v[[B_V2_F16]]
 ; VI:  v_min_f16_e32 v[[R_F16_1:[0-9]+]], 4.0, v[[B_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @minnum_v2f16_imm_a(
@@ -131,9 +138,9 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_min_f16_e32 v[[R_F16_0:[0-9]+]], 4.0, v[[A_V2_F16]]
 ; VI:  v_min_f16_e32 v[[R_F16_1:[0-9]+]], 0x4200, v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @minnum_v2f16_imm_b(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.rint.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.rint.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.rint.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.rint.f16.ll Wed Mar 15 14:04:26 2017
@@ -33,9 +33,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_rndne_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_rndne_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @rint_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.sin.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.sin.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.sin.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.sin.f16.ll Wed Mar 15 14:04:26 2017
@@ -26,22 +26,30 @@ entry:
 ; GCN-LABEL: {{^}}sin_v2f16
 ; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; SI:  v_mov_b32_e32 v[[HALF_PIE:[0-9]+]], 0x3e22f983{{$}}
-; GCN: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
-; SI:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], v[[HALF_PIE]], v[[A_F32_0]]
-; VI:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], 0.15915494, v[[A_F32_0]]
-; GCN: v_fract_f32_e32 v[[F_F32_0:[0-9]+]], v[[M_F32_0]]
-; SI:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], v[[HALF_PIE]], v[[A_F32_1]]
-; VI:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], 0.15915494, v[[A_F32_1]]
-; GCN: v_fract_f32_e32 v[[F_F32_1:[0-9]+]], v[[M_F32_1]]
-; GCN: v_sin_f32_e32 v[[R_F32_0:[0-9]+]], v[[F_F32_0]]
-; GCN: v_sin_f32_e32 v[[R_F32_1:[0-9]+]], v[[F_F32_1]]
-; GCN: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
-; GCN: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI-DAG: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI-DAG: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+
+
+; VI: v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+
+
+; SI-DAG:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], v[[HALF_PIE]], v[[A_F32_0]]
+; VI-DAG:  v_mul_f32_e32 v[[M_F32_0:[0-9]+]], 0.15915494, v[[A_F32_0]]
+; GCN-DAG: v_fract_f32_e32 v[[F_F32_0:[0-9]+]], v[[M_F32_0]]
+
+; SI-DAG:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], v[[HALF_PIE]], v[[A_F32_1]]
+; VI-DAG:  v_mul_f32_e32 v[[M_F32_1:[0-9]+]], 0.15915494, v[[A_F32_1]]
+; GCN-DAG: v_fract_f32_e32 v[[F_F32_1:[0-9]+]], v[[M_F32_1]]
+
+; GCN-DAG: v_sin_f32_e32 v[[R_F32_0:[0-9]+]], v[[F_F32_0]]
+; GCN-DAG: v_sin_f32_e32 v[[R_F32_1:[0-9]+]], v[[F_F32_1]]
+; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
+; GCN-DAG: v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @sin_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.sqrt.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.sqrt.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.sqrt.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.sqrt.f16.ll Wed Mar 15 14:04:26 2017
@@ -33,9 +33,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_sqrt_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_sqrt_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @sqrt_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/llvm.trunc.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/llvm.trunc.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/llvm.trunc.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/llvm.trunc.f16.ll Wed Mar 15 14:04:26 2017
@@ -33,9 +33,8 @@ entry:
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
 ; VI:  v_trunc_f16_e32 v[[R_F16_0:[0-9]+]], v[[A_V2_F16]]
 ; VI:  v_trunc_f16_e32 v[[R_F16_1:[0-9]+]], v[[A_F16_1]]
-; GCN: v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
 ; GCN: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
-; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_LO]]
+; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
 define void @trunc_v2f16(

Modified: llvm/trunk/test/CodeGen/AMDGPU/select.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/select.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/select.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/select.f16.ll Wed Mar 15 14:04:26 2017
@@ -188,19 +188,19 @@ entry:
 ; GCN-LABEL: {{^}}select_v2f16_imm_a:
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
-; SI:  v_cmp_lt_f32_e32 vcc, 0.5
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cmp_lt_f32_e64
+; SI:  v_cmp_lt_f32_e32 vcc, 0.5
 
 ; VI:  v_cmp_lt_f16_e32
 ; VI:  v_cmp_lt_f16_e64
 ; GCN: v_cndmask_b32_e32
-; SI:  v_cvt_f16_f32_e32
 ; GCN: v_cndmask_b32_e64
 ; SI:  v_cvt_f16_f32_e32
+; SI:  v_cvt_f16_f32_e32
 ; GCN: s_endpgm
 define void @select_v2f16_imm_a(
     <2 x half> addrspace(1)* %r,
@@ -220,18 +220,19 @@ entry:
 ; GCN-LABEL: {{^}}select_v2f16_imm_b:
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
-; SI:  v_cmp_gt_f32_e32 vcc, 0.5
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cmp_gt_f32_e64
+; SI:  v_cmp_gt_f32_e32 vcc, 0.5
 
 ; VI:  v_cmp_gt_f16_e32
 ; VI:  v_cmp_gt_f16_e64
 ; GCN: v_cndmask_b32_e32
-; SI:  v_cvt_f16_f32_e32
 ; GCN: v_cndmask_b32_e64
+
+; SI:  v_cvt_f16_f32_e32
 ; SI:  v_cvt_f16_f32_e32
 ; GCN: s_endpgm
 define void @select_v2f16_imm_b(
@@ -258,8 +259,8 @@ entry:
 ; SI:  v_cvt_f32_f16_e32
 
 ; SI: v_cmp_nlt_f32_e32
-; SI: v_cndmask_b32_e32
-; SI: v_cmp_nlt_f32_e32
+; SI: v_cmp_nlt_f32_e64
+; SI: v_cndmask_b32_e64
 ; SI: v_cndmask_b32_e32
 
 ; VI: v_cmp_nlt_f16_e32
@@ -293,12 +294,13 @@ entry:
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
 ; SI:  v_cvt_f32_f16_e32
-; SI:  v_cmp_lt_f32_e32
 ; SI:  v_cmp_lt_f32_e64
+; SI:  v_cmp_lt_f32_e32
+
 ; VI:  v_cmp_lt_f16_e32
 ; VI:  v_cmp_lt_f16_e64
-; GCN: v_cndmask_b32_e32
-; GCN: v_cndmask_b32_e64
+; GCN: v_cndmask_b32
+; GCN: v_cndmask_b32
 ; SI:  v_cvt_f16_f32_e32
 ; SI:  v_cvt_f16_f32_e32
 ; GCN: s_endpgm

Modified: llvm/trunk/test/CodeGen/AMDGPU/sitofp.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/sitofp.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/sitofp.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/sitofp.f16.ll Wed Mar 15 14:04:26 2017
@@ -61,7 +61,6 @@ entry:
 ; GCN:     v_cvt_f32_i32_e32
 ; GCN:     v_cvt_f16_f32_e32
 ; GCN:     v_cvt_f16_f32_e32
-; GCN-DAG: v_and_b32_e32
 ; GCN-DAG: v_lshlrev_b32_e32
 ; GCN-DAG: v_or_b32_e32
 ; GCN:     buffer_store_dword

Modified: llvm/trunk/test/CodeGen/AMDGPU/uitofp.f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/uitofp.f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/uitofp.f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/uitofp.f16.ll Wed Mar 15 14:04:26 2017
@@ -44,7 +44,6 @@ entry:
 ; VI:      v_cvt_f32_i32_e32
 ; GCN:     v_cvt_f16_f32_e32
 ; GCN:     v_cvt_f16_f32_e32
-; GCN-DAG: v_and_b32_e32
 ; GCN-DAG: v_lshlrev_b32_e32
 ; GCN-DAG: v_or_b32_e32
 ; GCN:     buffer_store_dword
@@ -65,7 +64,6 @@ entry:
 ; GCN:     v_cvt_f32_u32_e32
 ; GCN:     v_cvt_f16_f32_e32
 ; GCN:     v_cvt_f16_f32_e32
-; GCN-DAG: v_and_b32_e32
 ; GCN-DAG: v_lshlrev_b32_e32
 ; GCN-DAG: v_or_b32_e32
 ; GCN:     buffer_store_dword

Modified: llvm/trunk/test/CodeGen/AMDGPU/v_mac_f16.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AMDGPU/v_mac_f16.ll?rev=297873&r1=297872&r2=297873&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AMDGPU/v_mac_f16.ll (original)
+++ llvm/trunk/test/CodeGen/AMDGPU/v_mac_f16.ll Wed Mar 15 14:04:26 2017
@@ -302,21 +302,25 @@ entry:
 ; GCN: {{buffer|flat}}_load_dword v[[A_V2_F16:[0-9]+]]
 ; GCN: {{buffer|flat}}_load_dword v[[B_V2_F16:[0-9]+]]
 ; GCN: {{buffer|flat}}_load_dword v[[C_V2_F16:[0-9]+]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
+; SI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
-; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
+; SI-DAG: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; SI-DAG: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[C_F32_0:[0-9]+]], v[[C_V2_F16]]
+; SI-DAG:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_cvt_f32_f16_e32 v[[C_F32_1:[0-9]+]], v[[C_F16_1]]
-; SI:  v_mac_f32_e32 v[[C_F32_0]], v[[B_F32_0]], v[[A_F32_0]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[C_F32_0]]
-; SI:  v_mac_f32_e32 v[[C_F32_1]], v[[B_F32_1]], v[[A_F32_1]]
-; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[C_F32_1]]
-; SI:  v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[R_F16_0]]
+; SI-DAG:  v_mac_f32_e32 v[[C_F32_0]], v[[B_F32_0]], v[[A_F32_0]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_LO:[0-9]+]], v[[C_F32_0]]
+; SI-DAG:  v_mac_f32_e32 v[[C_F32_1]], v[[B_F32_1]], v[[A_F32_1]]
+; SI-DAG:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[C_F32_1]]
 ; SI:  v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
+
+; VI: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; VI: v_lshrrev_b32_e32 v[[C_F16_1:[0-9]+]], 16, v[[C_V2_F16]]
 ; VI:  v_mac_f16_e32 v[[C_V2_F16]], v[[B_V2_F16]], v[[A_V2_F16]]
 ; VI:  v_mac_f16_e32 v[[C_F16_1]], v[[B_F16_1]], v[[A_F16_1]]
 ; VI:  v_and_b32_e32 v[[R_F16_LO:[0-9]+]], 0xffff, v[[C_V2_F16]]
@@ -342,10 +346,10 @@ entry:
 }
 
 ; GCN-LABEL: {{^}}mac_v2f16_same_add:
-; SI:  v_mad_f32 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, [[ADD0:v[0-9]+]]
-; SI:  v_mad_f32 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, [[ADD1:v[0-9]+]]
-; SI:  v_mac_f32_e32 [[ADD0]], v{{[0-9]+}}, v{{[0-9]+}}
-; SI:  v_mac_f32_e32 [[ADD1]], v{{[0-9]+}}, v{{[0-9]+}}
+; SI:  v_mad_f32 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+; SI:  v_mad_f32 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+; SI:  v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
+; SI:  v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 ; VI:  v_mad_f16 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, [[ADD0:v[0-9]+]]
 ; VI:  v_mad_f16 v{{[0-9]}}, v{{[0-9]+}}, v{{[0-9]+}}, [[ADD1:v[0-9]+]]
 ; VI:  v_mac_f16_e32 [[ADD0]], v{{[0-9]+}}, v{{[0-9]+}}
@@ -471,8 +475,8 @@ entry:
 ; GCN-LABEL: {{^}}mac_v2f16_neg_a_safe_fp_math:
 ; SI:  v_sub_f32_e32 v[[NEG_A0:[0-9]+]], 0, v{{[0-9]+}}
 ; SI:  v_sub_f32_e32 v[[NEG_A1:[0-9]+]], 0, v{{[0-9]+}}
-; SI:  v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[NEG_A0]]
-; SI:  v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[NEG_A1]]
+; SI-DAG:  v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[NEG_A0]]
+; SI-DAG:  v_mac_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[NEG_A1]]
 ; VI:  v_sub_f16_e32 v[[NEG_A0:[0-9]+]], 0, v{{[0-9]+}}
 ; VI:  v_sub_f16_e32 v[[NEG_A1:[0-9]+]], 0, v{{[0-9]+}}
 ; VI:  v_mac_f16_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[NEG_A0]]
@@ -499,8 +503,8 @@ entry:
 ; GCN-LABEL: {{^}}mac_v2f16_neg_b_safe_fp_math:
 ; SI:  v_sub_f32_e32 v[[NEG_A0:[0-9]+]], 0, v{{[0-9]+}}
 ; SI:  v_sub_f32_e32 v[[NEG_A1:[0-9]+]], 0, v{{[0-9]+}}
-; SI:  v_mac_f32_e32 v{{[0-9]+}}, v[[NEG_A0]], v{{[0-9]+}}
-; SI:  v_mac_f32_e32 v{{[0-9]+}}, v[[NEG_A1]], v{{[0-9]+}}
+; SI-DAG:  v_mac_f32_e32 v{{[0-9]+}}, v[[NEG_A0]], v{{[0-9]+}}
+; SI-DAG:  v_mac_f32_e32 v{{[0-9]+}}, v[[NEG_A1]], v{{[0-9]+}}
 ; VI:  v_sub_f16_e32 v[[NEG_A0:[0-9]+]], 0, v{{[0-9]+}}
 ; VI:  v_sub_f16_e32 v[[NEG_A1:[0-9]+]], 0, v{{[0-9]+}}
 ; VI:  v_mac_f16_e32 v{{[0-9]+}}, v[[NEG_A0]], v{{[0-9]+}}
@@ -527,8 +531,8 @@ entry:
 ; GCN-LABEL: {{^}}mac_v2f16_neg_c_safe_fp_math:
 ; SI:  v_sub_f32_e32 v[[NEG_A0:[0-9]+]], 0, v{{[0-9]+}}
 ; SI:  v_sub_f32_e32 v[[NEG_A1:[0-9]+]], 0, v{{[0-9]+}}
-; SI:  v_mac_f32_e32 v[[NEG_A0]], v{{[0-9]+}}, v{{[0-9]+}}
-; SI:  v_mac_f32_e32 v[[NEG_A1]], v{{[0-9]+}}, v{{[0-9]+}}
+; SI-DAG:  v_mac_f32_e32 v[[NEG_A0]], v{{[0-9]+}}, v{{[0-9]+}}
+; SI-DAG:  v_mac_f32_e32 v[[NEG_A1]], v{{[0-9]+}}, v{{[0-9]+}}
 ; VI:  v_sub_f16_e32 v[[NEG_A0:[0-9]+]], 0, v{{[0-9]+}}
 ; VI:  v_sub_f16_e32 v[[NEG_A1:[0-9]+]], 0, v{{[0-9]+}}
 ; VI:  v_mac_f16_e32 v[[NEG_A0]], v{{[0-9]+}}, v{{[0-9]+}}




More information about the llvm-commits mailing list