R600/SI - Add new CI arithmetic instructions.

Does not yet include larger part required
to match v_mad_i64_i32 / v_mad_u64_u32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202077 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/R600/AMDGPUISelLowering.cpp

@@ -45,6 +45,8 @@ static bool allocateStack(unsigned ValNo, MVT ValVT, MVT LocVT,
 AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   TargetLowering(TM, new TargetLoweringObjectFileELF()) {
 
+  Subtarget = &TM.getSubtarget<AMDGPUSubtarget>();
+
   // Initialize target lowering borrowed from AMDIL
   InitAMDILLowering();
 

lib/Target/R600/AMDGPUISelLowering.h

@@ -21,9 +21,13 @@
 namespace llvm {
 
 class AMDGPUMachineFunction;
+class AMDGPUSubtarget;
 class MachineRegisterInfo;
 
 class AMDGPUTargetLowering : public TargetLowering {
+protected:
+  const AMDGPUSubtarget *Subtarget;
+
 private:
   void ExtractVectorElements(SDValue Op, SelectionDAG &DAG,
                              SmallVectorImpl<SDValue> &Args,

lib/Target/R600/SIISelLowering.cpp

@@ -14,6 +14,7 @@
 
 #include "SIISelLowering.h"
 #include "AMDGPU.h"
+#include "AMDGPUSubtarget.h"
 #include "AMDILIntrinsicInfo.h"
 #include "SIInstrInfo.h"
 #include "SIMachineFunctionInfo.h"
@@ -30,7 +31,6 @@ using namespace llvm;
 
 SITargetLowering::SITargetLowering(TargetMachine &TM) :
     AMDGPUTargetLowering(TM) {
-
   addRegisterClass(MVT::i1, &AMDGPU::SReg_64RegClass);
   addRegisterClass(MVT::i64, &AMDGPU::VSrc_64RegClass);
 
@@ -175,8 +175,20 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
     }
   }
 
-  setTargetDAGCombine(ISD::SELECT_CC);
+  for (int I = MVT::v1f64; I <= MVT::v8f64; ++I) {
+    MVT::SimpleValueType VT = static_cast<MVT::SimpleValueType>(I);
+    setOperationAction(ISD::FTRUNC, MVT::f64, Expand);
+    setOperationAction(ISD::FCEIL, MVT::f64, Expand);
+    setOperationAction(ISD::FFLOOR, MVT::f64, Expand);
+  }
 
+  if (Subtarget->getGeneration() >= AMDGPUSubtarget::SEA_ISLANDS) {
+    setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
+    setOperationAction(ISD::FCEIL, MVT::f64, Legal);
+    setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
+  }
+
+  setTargetDAGCombine(ISD::SELECT_CC);
   setTargetDAGCombine(ISD::SETCC);
 
   setSchedulingPreference(Sched::RegPressure);

lib/Target/R600/SIInstructions.td

@@ -29,6 +29,9 @@ def SendMsgImm : Operand<i32> {
 def isSI : Predicate<"Subtarget.getGeneration() "
                       ">= AMDGPUSubtarget::SOUTHERN_ISLANDS">;
 
+def isCI : Predicate<"Subtarget.getGeneration() "
+                      ">= AMDGPUSubtarget::SEA_ISLANDS">;
+
 def WAIT_FLAG : InstFlag<"printWaitFlag">;
 
 let Predicates = [isSI] in {
@@ -2104,6 +2107,55 @@ def : MTBUF_StoreResource <v2i32, 2, TBUFFER_STORE_FORMAT_XY>;
 def : MTBUF_StoreResource <v4i32, 3, TBUFFER_STORE_FORMAT_XYZ>;
 def : MTBUF_StoreResource <v4i32, 4, TBUFFER_STORE_FORMAT_XYZW>;
 
+let Predicates = [isCI] in {
+
+// Sea island new arithmetic instructinos
+let neverHasSideEffects = 1 in {
+defm V_TRUNC_F64 : VOP1_64 <0x00000017, "V_TRUNC_F64",
+  [(set f64:$dst, (ftrunc f64:$src0))]
+>;
+defm V_CEIL_F64 : VOP1_64 <0x00000018, "V_CEIL_F64",
+  [(set f64:$dst, (fceil f64:$src0))]
+>;
+defm V_FLOOR_F64 : VOP1_64 <0x0000001A, "V_FLOOR_F64",
+  [(set f64:$dst, (ffloor f64:$src0))]
+>;
+
+defm V_RNDNE_F64 : VOP1_64 <0x00000019, "V_RNDNE_F64", []>;
+
+def V_QSAD_PK_U16_U8 : VOP3_32 <0x00000173, "V_QSAD_PK_U16_U8", []>;
+def V_MQSAD_U16_U8 : VOP3_32 <0x000000172, "V_MQSAD_U16_U8", []>;
+def V_MQSAD_U32_U8 : VOP3_32 <0x00000175, "V_MQSAD_U32_U8", []>;
+def V_MAD_U64_U32 : VOP3_64 <0x00000176, "V_MAD_U64_U32", []>;
+
+// XXX - Does this set VCC?
+def V_MAD_I64_I32 : VOP3_64 <0x00000177, "V_MAD_I64_I32", []>;
+} // End neverHasSideEffects = 1
+
+// Remaining instructions:
+// FLAT_*
+// S_CBRANCH_CDBGUSER
+// S_CBRANCH_CDBGSYS
+// S_CBRANCH_CDBGSYS_OR_USER
+// S_CBRANCH_CDBGSYS_AND_USER
+// S_DCACHE_INV_VOL
+// V_EXP_LEGACY_F32
+// V_LOG_LEGACY_F32
+// DS_NOP
+// DS_GWS_SEMA_RELEASE_ALL
+// DS_WRAP_RTN_B32
+// DS_CNDXCHG32_RTN_B64
+// DS_WRITE_B96
+// DS_WRITE_B128
+// DS_CONDXCHG32_RTN_B128
+// DS_READ_B96
+// DS_READ_B128
+// BUFFER_LOAD_DWORDX3
+// BUFFER_STORE_DWORDX3
+
+} // End Predicates = [isCI]
+
+
 /********** ====================== **********/
 /**********   Indirect adressing   **********/
 /********** ====================== **********/

test/CodeGen/R600/fceil.ll

@@ -0,0 +1,84 @@
+; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI %s
+
+declare double @llvm.ceil.f64(double) nounwind readnone
+declare <2 x double> @llvm.ceil.v2f64(<2 x double>) nounwind readnone
+declare <3 x double> @llvm.ceil.v3f64(<3 x double>) nounwind readnone
+declare <4 x double> @llvm.ceil.v4f64(<4 x double>) nounwind readnone
+declare <8 x double> @llvm.ceil.v8f64(<8 x double>) nounwind readnone
+declare <16 x double> @llvm.ceil.v16f64(<16 x double>) nounwind readnone
+
+; CI-LABEL: @fceil_f64:
+; CI: V_CEIL_F64_e32
+define void @fceil_f64(double addrspace(1)* %out, double %x) {
+  %y = call double @llvm.ceil.f64(double %x) nounwind readnone
+  store double %y, double addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @fceil_v2f64:
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+define void @fceil_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
+  %y = call <2 x double> @llvm.ceil.v2f64(<2 x double> %x) nounwind readnone
+  store <2 x double> %y, <2 x double> addrspace(1)* %out
+  ret void
+}
+
+; FIXME-CI-LABEL: @fceil_v3f64:
+; FIXME-CI: V_CEIL_F64_e32
+; FIXME-CI: V_CEIL_F64_e32
+; FIXME-CI: V_CEIL_F64_e32
+; define void @fceil_v3f64(<3 x double> addrspace(1)* %out, <3 x double> %x) {
+;   %y = call <3 x double> @llvm.ceil.v3f64(<3 x double> %x) nounwind readnone
+;   store <3 x double> %y, <3 x double> addrspace(1)* %out
+;   ret void
+; }
+
+; CI-LABEL: @fceil_v4f64:
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+define void @fceil_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
+  %y = call <4 x double> @llvm.ceil.v4f64(<4 x double> %x) nounwind readnone
+  store <4 x double> %y, <4 x double> addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @fceil_v8f64:
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+define void @fceil_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
+  %y = call <8 x double> @llvm.ceil.v8f64(<8 x double> %x) nounwind readnone
+  store <8 x double> %y, <8 x double> addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @fceil_v16f64:
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+; CI: V_CEIL_F64_e32
+define void @fceil_v16f64(<16 x double> addrspace(1)* %out, <16 x double> %x) {
+  %y = call <16 x double> @llvm.ceil.v16f64(<16 x double> %x) nounwind readnone
+  store <16 x double> %y, <16 x double> addrspace(1)* %out
+  ret void
+}

test/CodeGen/R600/ffloor.ll

@@ -0,0 +1,84 @@
+; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI %s
+
+declare double @llvm.floor.f64(double) nounwind readnone
+declare <2 x double> @llvm.floor.v2f64(<2 x double>) nounwind readnone
+declare <3 x double> @llvm.floor.v3f64(<3 x double>) nounwind readnone
+declare <4 x double> @llvm.floor.v4f64(<4 x double>) nounwind readnone
+declare <8 x double> @llvm.floor.v8f64(<8 x double>) nounwind readnone
+declare <16 x double> @llvm.floor.v16f64(<16 x double>) nounwind readnone
+
+; CI-LABEL: @ffloor_f64:
+; CI: V_FLOOR_F64_e32
+define void @ffloor_f64(double addrspace(1)* %out, double %x) {
+  %y = call double @llvm.floor.f64(double %x) nounwind readnone
+  store double %y, double addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @ffloor_v2f64:
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+define void @ffloor_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
+  %y = call <2 x double> @llvm.floor.v2f64(<2 x double> %x) nounwind readnone
+  store <2 x double> %y, <2 x double> addrspace(1)* %out
+  ret void
+}
+
+; FIXME-CI-LABEL: @ffloor_v3f64:
+; FIXME-CI: V_FLOOR_F64_e32
+; FIXME-CI: V_FLOOR_F64_e32
+; FIXME-CI: V_FLOOR_F64_e32
+; define void @ffloor_v3f64(<3 x double> addrspace(1)* %out, <3 x double> %x) {
+;   %y = call <3 x double> @llvm.floor.v3f64(<3 x double> %x) nounwind readnone
+;   store <3 x double> %y, <3 x double> addrspace(1)* %out
+;   ret void
+; }
+
+; CI-LABEL: @ffloor_v4f64:
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+define void @ffloor_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
+  %y = call <4 x double> @llvm.floor.v4f64(<4 x double> %x) nounwind readnone
+  store <4 x double> %y, <4 x double> addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @ffloor_v8f64:
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+define void @ffloor_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
+  %y = call <8 x double> @llvm.floor.v8f64(<8 x double> %x) nounwind readnone
+  store <8 x double> %y, <8 x double> addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @ffloor_v16f64:
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+; CI: V_FLOOR_F64_e32
+define void @ffloor_v16f64(<16 x double> addrspace(1)* %out, <16 x double> %x) {
+  %y = call <16 x double> @llvm.floor.v16f64(<16 x double> %x) nounwind readnone
+  store <16 x double> %y, <16 x double> addrspace(1)* %out
+  ret void
+}

test/CodeGen/R600/ftrunc.ll

@@ -0,0 +1,84 @@
+; RUN: llc -march=r600 -mcpu=bonaire < %s | FileCheck -check-prefix=CI %s
+
+declare double @llvm.trunc.f64(double) nounwind readnone
+declare <2 x double> @llvm.trunc.v2f64(<2 x double>) nounwind readnone
+declare <3 x double> @llvm.trunc.v3f64(<3 x double>) nounwind readnone
+declare <4 x double> @llvm.trunc.v4f64(<4 x double>) nounwind readnone
+declare <8 x double> @llvm.trunc.v8f64(<8 x double>) nounwind readnone
+declare <16 x double> @llvm.trunc.v16f64(<16 x double>) nounwind readnone
+
+; CI-LABEL: @ftrunc_f64:
+; CI: V_TRUNC_F64_e32
+define void @ftrunc_f64(double addrspace(1)* %out, double %x) {
+  %y = call double @llvm.trunc.f64(double %x) nounwind readnone
+  store double %y, double addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @ftrunc_v2f64:
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+define void @ftrunc_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) {
+  %y = call <2 x double> @llvm.trunc.v2f64(<2 x double> %x) nounwind readnone
+  store <2 x double> %y, <2 x double> addrspace(1)* %out
+  ret void
+}
+
+; FIXME-CI-LABEL: @ftrunc_v3f64:
+; FIXME-CI: V_TRUNC_F64_e32
+; FIXME-CI: V_TRUNC_F64_e32
+; FIXME-CI: V_TRUNC_F64_e32
+; define void @ftrunc_v3f64(<3 x double> addrspace(1)* %out, <3 x double> %x) {
+;   %y = call <3 x double> @llvm.trunc.v3f64(<3 x double> %x) nounwind readnone
+;   store <3 x double> %y, <3 x double> addrspace(1)* %out
+;   ret void
+; }
+
+; CI-LABEL: @ftrunc_v4f64:
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+define void @ftrunc_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) {
+  %y = call <4 x double> @llvm.trunc.v4f64(<4 x double> %x) nounwind readnone
+  store <4 x double> %y, <4 x double> addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @ftrunc_v8f64:
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+define void @ftrunc_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) {
+  %y = call <8 x double> @llvm.trunc.v8f64(<8 x double> %x) nounwind readnone
+  store <8 x double> %y, <8 x double> addrspace(1)* %out
+  ret void
+}
+
+; CI-LABEL: @ftrunc_v16f64:
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+; CI: V_TRUNC_F64_e32
+define void @ftrunc_v16f64(<16 x double> addrspace(1)* %out, <16 x double> %x) {
+  %y = call <16 x double> @llvm.trunc.v16f64(<16 x double> %x) nounwind readnone
+  store <16 x double> %y, <16 x double> addrspace(1)* %out
+  ret void
+}