From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Fri, 3 Oct 2014 23:54:41 +0000 (+0000)
Subject: R600: Custom lower [s|u]int_to_fp for i64 -> f64
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=ad2f641c33bf41fb7cfec85e9a8e991ddeb110f9;p=oota-llvm.git

R600: Custom lower [s|u]int_to_fp for i64 -> f64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219037 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/R600/AMDGPUISelLowering.cpp b/lib/Target/R600/AMDGPUISelLowering.cpp
index 9a5e39e8e30..c4ae6a9b91d 100644
--- a/lib/Target/R600/AMDGPUISelLowering.cpp
+++ b/lib/Target/R600/AMDGPUISelLowering.cpp
@@ -285,6 +285,7 @@ AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::UDIV, MVT::i32, Expand);
   setOperationAction(ISD::UREM, MVT::i32, Expand);
   setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
+  setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
   setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
 
   if (!Subtarget->hasFFBH())
@@ -555,6 +556,7 @@ SDValue AMDGPUTargetLowering::LowerOperation(SDValue Op,
   case ISD::FRINT: return LowerFRINT(Op, DAG);
   case ISD::FNEARBYINT: return LowerFNEARBYINT(Op, DAG);
   case ISD::FFLOOR: return LowerFFLOOR(Op, DAG);
+  case ISD::SINT_TO_FP: return LowerSINT_TO_FP(Op, DAG);
   case ISD::UINT_TO_FP: return LowerUINT_TO_FP(Op, DAG);
   }
   return Op;
@@ -1805,13 +1807,43 @@ SDValue AMDGPUTargetLowering::LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const {
   return DAG.getNode(ISD::FADD, SL, MVT::f64, Trunc, Add);
 }
 
+SDValue AMDGPUTargetLowering::LowerINT_TO_FP64(SDValue Op, SelectionDAG &DAG,
+                                               bool Signed) const {
+  SDLoc SL(Op);
+  SDValue Src = Op.getOperand(0);
+
+  SDValue BC = DAG.getNode(ISD::BITCAST, SL, MVT::v2i32, Src);
+
+  SDValue Lo = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, MVT::i32, BC,
+                           DAG.getConstant(0, MVT::i32));
+  SDValue Hi = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SL, MVT::i32, BC,
+                           DAG.getConstant(1, MVT::i32));
+
+  SDValue CvtHi = DAG.getNode(Signed ? ISD::SINT_TO_FP : ISD::UINT_TO_FP,
+                              SL, MVT::f64, Hi);
+
+  SDValue CvtLo = DAG.getNode(ISD::UINT_TO_FP, SL, MVT::f64, Lo);
+
+  SDValue LdExp = DAG.getNode(AMDGPUISD::LDEXP, SL, MVT::f64, CvtHi,
+                              DAG.getConstant(32, MVT::i32));
+
+  return DAG.getNode(ISD::FADD, SL, MVT::f64, LdExp, CvtLo);
+}
+
 SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
                                                SelectionDAG &DAG) const {
   SDValue S0 = Op.getOperand(0);
-  SDLoc DL(Op);
-  if (Op.getValueType() != MVT::f32 || S0.getValueType() != MVT::i64)
+  if (S0.getValueType() != MVT::i64)
     return SDValue();
 
+  EVT DestVT = Op.getValueType();
+  if (DestVT == MVT::f64)
+    return LowerINT_TO_FP64(Op, DAG, false);
+
+  assert(DestVT == MVT::f32);
+
+  SDLoc DL(Op);
+
   // f32 uint_to_fp i64
   SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, S0,
                            DAG.getConstant(0, MVT::i32));
@@ -1824,6 +1856,15 @@ SDValue AMDGPUTargetLowering::LowerUINT_TO_FP(SDValue Op,
   return DAG.getNode(ISD::FADD, DL, MVT::f32, FloatLo, FloatHi);
 }
 
+SDValue AMDGPUTargetLowering::LowerSINT_TO_FP(SDValue Op,
+                                              SelectionDAG &DAG) const {
+  SDValue Src = Op.getOperand(0);
+  if (Src.getValueType() == MVT::i64 && Op.getValueType() == MVT::f64)
+    return LowerINT_TO_FP64(Op, DAG, true);
+
+  return SDValue();
+}
+
 SDValue AMDGPUTargetLowering::ExpandSIGN_EXTEND_INREG(SDValue Op,
                                                       unsigned BitsDiff,
                                                       SelectionDAG &DAG) const {
diff --git a/lib/Target/R600/AMDGPUISelLowering.h b/lib/Target/R600/AMDGPUISelLowering.h
index 911576b0afe..84cb52282f8 100644
--- a/lib/Target/R600/AMDGPUISelLowering.h
+++ b/lib/Target/R600/AMDGPUISelLowering.h
@@ -51,7 +51,9 @@ private:
   SDValue LowerFNEARBYINT(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFFLOOR(SDValue Op, SelectionDAG &DAG) const;
 
+  SDValue LowerINT_TO_FP64(SDValue Op, SelectionDAG &DAG, bool Signed) const;
   SDValue LowerUINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
 
   SDValue ExpandSIGN_EXTEND_INREG(SDValue Op,
                                   unsigned BitsDiff,
diff --git a/lib/Target/R600/SIISelLowering.h b/lib/Target/R600/SIISelLowering.h
index 9cf4dbcb2fd..9e50f0b2d69 100644
--- a/lib/Target/R600/SIISelLowering.h
+++ b/lib/Target/R600/SIISelLowering.h
@@ -37,6 +37,7 @@ class SITargetLowering : public AMDGPUTargetLowering {
   SDValue LowerFDIV32(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFDIV64(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerFDIV(SDValue Op, SelectionDAG &DAG) const;
+  SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG, bool Signed) const;
   SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerTrig(SDValue Op, SelectionDAG &DAG) const;
   SDValue LowerBRCOND(SDValue Op, SelectionDAG &DAG) const;
diff --git a/test/CodeGen/R600/sint_to_fp.f64.ll b/test/CodeGen/R600/sint_to_fp.f64.ll
new file mode 100644
index 00000000000..d2b3f12c1f3
--- /dev/null
+++ b/test/CodeGen/R600/sint_to_fp.f64.ll
@@ -0,0 +1,60 @@
+; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
+
+declare i32 @llvm.r600.read.tidig.x() nounwind readnone
+
+; SI-LABEL: {{^}}sint_to_fp_i32_to_f64
+; SI: V_CVT_F64_I32_e32
+define void @sint_to_fp_i32_to_f64(double addrspace(1)* %out, i32 %in) {
+  %result = sitofp i32 %in to double
+  store double %result, double addrspace(1)* %out
+  ret void
+}
+
+; SI-LABEL: {{^}}sint_to_fp_i1_f64:
+; SI: V_CMP_EQ_I32_e64 [[CMP:s\[[0-9]+:[0-9]\]]],
+; FIXME: We should the VGPR sources for V_CNDMASK are copied from SGPRs,
+; we should be able to fold the SGPRs into the V_CNDMASK instructions.
+; SI: V_CNDMASK_B32_e64 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[CMP]]
+; SI: V_CNDMASK_B32_e64 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[CMP]]
+; SI: BUFFER_STORE_DWORDX2
+; SI: S_ENDPGM
+define void @sint_to_fp_i1_f64(double addrspace(1)* %out, i32 %in) {
+  %cmp = icmp eq i32 %in, 0
+  %fp = sitofp i1 %cmp to double
+  store double %fp, double addrspace(1)* %out, align 4
+  ret void
+}
+
+; SI-LABEL: {{^}}sint_to_fp_i1_f64_load:
+; SI: V_CNDMASK_B32_e64 [[IRESULT:v[0-9]]], 0, -1
+; SI-NEXT: V_CVT_F64_I32_e32 [[RESULT:v\[[0-9]+:[0-9]\]]], [[IRESULT]]
+; SI: BUFFER_STORE_DWORDX2 [[RESULT]]
+; SI: S_ENDPGM
+define void @sint_to_fp_i1_f64_load(double addrspace(1)* %out, i1 %in) {
+  %fp = sitofp i1 %in to double
+  store double %fp, double addrspace(1)* %out, align 8
+  ret void
+}
+
+; SI-LABEL: @s_sint_to_fp_i64_to_f64
+define void @s_sint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 %in) {
+  %result = sitofp i64 %in to double
+  store double %result, double addrspace(1)* %out
+  ret void
+}
+
+; SI-LABEL: @v_sint_to_fp_i64_to_f64
+; SI: BUFFER_LOAD_DWORDX2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
+; SI-DAG: V_CVT_F64_U32_e32 [[LO_CONV:v\[[0-9]+:[0-9]+\]]], v[[LO]]
+; SI-DAG: V_CVT_F64_I32_e32 [[HI_CONV:v\[[0-9]+:[0-9]+\]]], v[[HI]]
+; SI: V_LDEXP_F64 [[LDEXP:v\[[0-9]+:[0-9]+\]]], [[HI_CONV]], 32
+; SI: V_ADD_F64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[LDEXP]], [[LO_CONV]]
+; SI: BUFFER_STORE_DWORDX2 [[RESULT]]
+define void @v_sint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 addrspace(1)* %in) {
+  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
+  %gep = getelementptr i64 addrspace(1)* %in, i32 %tid
+  %val = load i64 addrspace(1)* %gep, align 8
+  %result = sitofp i64 %val to double
+  store double %result, double addrspace(1)* %out
+  ret void
+}
diff --git a/test/CodeGen/R600/sint_to_fp64.ll b/test/CodeGen/R600/sint_to_fp64.ll
deleted file mode 100644
index 43889b616da..00000000000
--- a/test/CodeGen/R600/sint_to_fp64.ll
+++ /dev/null
@@ -1,35 +0,0 @@
-; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
-
-; SI: {{^}}sint_to_fp64:
-; SI: V_CVT_F64_I32_e32
-define void @sint_to_fp64(double addrspace(1)* %out, i32 %in) {
-  %result = sitofp i32 %in to double
-  store double %result, double addrspace(1)* %out
-  ret void
-}
-
-; SI-LABEL: {{^}}sint_to_fp_i1_f64:
-; SI: V_CMP_EQ_I32_e64 [[CMP:s\[[0-9]+:[0-9]\]]],
-; FIXME: We should the VGPR sources for V_CNDMASK are copied from SGPRs,
-; we should be able to fold the SGPRs into the V_CNDMASK instructions.
-; SI: V_CNDMASK_B32_e64 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[CMP]]
-; SI: V_CNDMASK_B32_e64 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}, [[CMP]]
-; SI: BUFFER_STORE_DWORDX2
-; SI: S_ENDPGM
-define void @sint_to_fp_i1_f64(double addrspace(1)* %out, i32 %in) {
-  %cmp = icmp eq i32 %in, 0
-  %fp = sitofp i1 %cmp to double
-  store double %fp, double addrspace(1)* %out, align 4
-  ret void
-}
-
-; SI-LABEL: {{^}}sint_to_fp_i1_f64_load:
-; SI: V_CNDMASK_B32_e64 [[IRESULT:v[0-9]]], 0, -1
-; SI-NEXT: V_CVT_F64_I32_e32 [[RESULT:v\[[0-9]+:[0-9]\]]], [[IRESULT]]
-; SI: BUFFER_STORE_DWORDX2 [[RESULT]]
-; SI: S_ENDPGM
-define void @sint_to_fp_i1_f64_load(double addrspace(1)* %out, i1 %in) {
-  %fp = sitofp i1 %in to double
-  store double %fp, double addrspace(1)* %out, align 8
-  ret void
-}
diff --git a/test/CodeGen/R600/uint_to_fp.f64.ll b/test/CodeGen/R600/uint_to_fp.f64.ll
index 19dbbee8354..2e926d6999a 100644
--- a/test/CodeGen/R600/uint_to_fp.f64.ll
+++ b/test/CodeGen/R600/uint_to_fp.f64.ll
@@ -1,6 +1,8 @@
 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
 
-; SI-LABEL: {{^}}uint_to_fp_f64_i32:
+declare i32 @llvm.r600.read.tidig.x() nounwind readnone
+
+; SI-LABEL: {{$}}uint_to_fp_f64_i32
 ; SI: V_CVT_F64_U32_e32
 ; SI: S_ENDPGM
 define void @uint_to_fp_f64_i32(double addrspace(1)* %out, i32 %in) {
@@ -34,3 +36,40 @@ define void @uint_to_fp_i1_f64_load(double addrspace(1)* %out, i1 %in) {
   store double %fp, double addrspace(1)* %out, align 8
   ret void
 }
+
+; SI-LABEL: {{$}}v_uint_to_fp_i64_to_f64
+; SI: BUFFER_LOAD_DWORDX2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
+; SI-DAG: V_CVT_F64_U32_e32 [[LO_CONV:v\[[0-9]+:[0-9]+\]]], v[[LO]]
+; SI-DAG: V_CVT_F64_U32_e32 [[HI_CONV:v\[[0-9]+:[0-9]+\]]], v[[HI]]
+; SI: V_LDEXP_F64 [[LDEXP:v\[[0-9]+:[0-9]+\]]], [[HI_CONV]], 32
+; SI: V_ADD_F64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[LDEXP]], [[LO_CONV]]
+; SI: BUFFER_STORE_DWORDX2 [[RESULT]]
+define void @v_uint_to_fp_i64_to_f64(double addrspace(1)* %out, i64 addrspace(1)* %in) {
+  %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
+  %gep = getelementptr i64 addrspace(1)* %in, i32 %tid
+  %val = load i64 addrspace(1)* %gep, align 8
+  %result = uitofp i64 %val to double
+  store double %result, double addrspace(1)* %out
+  ret void
+}
+
+; SI-LABEL: {{$}}s_uint_to_fp_f64_i64
+define void @s_uint_to_fp_f64_i64(double addrspace(1)* %out, i64 %in) {
+  %cast = uitofp i64 %in to double
+  store double %cast, double addrspace(1)* %out, align 8
+  ret void
+}
+
+; SI-LABEL: {{$}}s_uint_to_fp_v2f64_v2i64
+define void @s_uint_to_fp_v2f64_v2i64(<2 x double> addrspace(1)* %out, <2 x i64> %in) {
+  %cast = uitofp <2 x i64> %in to <2 x double>
+  store <2 x double> %cast, <2 x double> addrspace(1)* %out, align 16
+  ret void
+}
+
+; SI-LABEL: {{$}}s_uint_to_fp_v4f64_v4i64
+define void @s_uint_to_fp_v4f64_v4i64(<4 x double> addrspace(1)* %out, <4 x i64> %in) {
+  %cast = uitofp <4 x i64> %in to <4 x double>
+  store <4 x double> %cast, <4 x double> addrspace(1)* %out, align 16
+  ret void
+}