From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Tue, 16 Jul 2019 02:46:05 +0000 (+0000)
Subject: AMDGPU: Avoid code predicates for extload PatFrags
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=48a2462687932fd6f424a4d6c683256cb7235c4d;p=llvm

AMDGPU: Avoid code predicates for extload PatFrags

Use the MemoryVT field. This will be necessary for tablegen to
automatically handle patterns for GlobalISel.

Doesn't handle the d16 lo/hi patterns. Those are a special case since
it involvess the custom node type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@366168 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/AMDGPUInstructions.td b/lib/Target/AMDGPU/AMDGPUInstructions.td
index b29b0e7b17e..9e9510e0fa4 100644
--- a/lib/Target/AMDGPU/AMDGPUInstructions.td
+++ b/lib/Target/AMDGPU/AMDGPUInstructions.td
@@ -379,27 +379,6 @@ class FlatStoreAddress : CodePatPred<[{
          AS == AMDGPUAS::GLOBAL_ADDRESS;
 }]>;
 
-class AZExtLoadBase <SDPatternOperator ld_node>: PatFrag<(ops node:$ptr),
-                                              (ld_node node:$ptr), [{
-  LoadSDNode *L = cast<LoadSDNode>(N);
-  return L->getExtensionType() == ISD::ZEXTLOAD ||
-         L->getExtensionType() == ISD::EXTLOAD;
-}]>;
-
-def az_extload : AZExtLoadBase <unindexedload>;
-
-def az_extloadi8 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
-}]>;
-
-def az_extloadi16 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
-}]>;
-
-def az_extloadi32 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
-  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i32;
-}]>;
-
 class PrivateLoad <SDPatternOperator op> : LoadFrag <op>, PrivateAddress;
 class PrivateStore <SDPatternOperator op> : StoreFrag <op>, PrivateAddress;
 
@@ -419,9 +398,11 @@ class ConstantLoad <SDPatternOperator op> : LoadFrag <op>, ConstantAddress;
 
 
 def load_private : PrivateLoad <load>;
-def az_extloadi8_private : PrivateLoad <az_extloadi8>;
+def extloadi8_private : PrivateLoad <extloadi8>;
+def zextloadi8_private : PrivateLoad <zextloadi8>;
 def sextloadi8_private : PrivateLoad <sextloadi8>;
-def az_extloadi16_private : PrivateLoad <az_extloadi16>;
+def extloadi16_private : PrivateLoad <extloadi16>;
+def zextloadi16_private : PrivateLoad <zextloadi16>;
 def sextloadi16_private : PrivateLoad <sextloadi16>;
 
 def store_private : PrivateStore <store>;
@@ -433,9 +414,11 @@ def truncstorei8_hi16_private : StoreHi16<truncstorei8>, PrivateAddress;
 
 def load_global : GlobalLoad <load>;
 def sextloadi8_global : GlobalLoad <sextloadi8>;
-def az_extloadi8_global : GlobalLoad <az_extloadi8>;
+def extloadi8_global : GlobalLoad <extloadi8>;
+def zextloadi8_global : GlobalLoad <zextloadi8>;
 def sextloadi16_global : GlobalLoad <sextloadi16>;
-def az_extloadi16_global : GlobalLoad <az_extloadi16>;
+def extloadi16_global : GlobalLoad <extloadi16>;
+def zextloadi16_global : GlobalLoad <zextloadi16>;
 def atomic_load_global : GlobalLoad<atomic_load>;
 
 def store_global : GlobalStore <store>;
@@ -479,9 +462,11 @@ def store_align16_local : Aligned16Bytes <
 >;
 
 def load_flat          : FlatLoad <load>;
-def az_extloadi8_flat  : FlatLoad <az_extloadi8>;
+def extloadi8_flat  : FlatLoad <extloadi8>;
+def zextloadi8_flat  : FlatLoad <zextloadi8>;
 def sextloadi8_flat    : FlatLoad <sextloadi8>;
-def az_extloadi16_flat : FlatLoad <az_extloadi16>;
+def extloadi16_flat : FlatLoad <extloadi16>;
+def zextloadi16_flat : FlatLoad <zextloadi16>;
 def sextloadi16_flat   : FlatLoad <sextloadi16>;
 def atomic_load_flat   : FlatLoad<atomic_load>;
 
@@ -495,9 +480,11 @@ def truncstorei16_hi16_flat : StoreHi16<truncstorei16>, FlatStoreAddress;
 
 def constant_load : ConstantLoad<load>;
 def sextloadi8_constant : ConstantLoad <sextloadi8>;
-def az_extloadi8_constant : ConstantLoad <az_extloadi8>;
+def extloadi8_constant : ConstantLoad <extloadi8>;
+def zextloadi8_constant : ConstantLoad <zextloadi8>;
 def sextloadi16_constant : ConstantLoad <sextloadi16>;
-def az_extloadi16_constant : ConstantLoad <az_extloadi16>;
+def extloadi16_constant : ConstantLoad <extloadi16>;
+def zextloadi16_constant : ConstantLoad <zextloadi16>;
 
 
 class local_binary_atomic_op<SDNode atomic_op> :
diff --git a/lib/Target/AMDGPU/BUFInstructions.td b/lib/Target/AMDGPU/BUFInstructions.td
index 7d9ca59c6d0..4ff9aeb2e31 100644
--- a/lib/Target/AMDGPU/BUFInstructions.td
+++ b/lib/Target/AMDGPU/BUFInstructions.td
@@ -851,9 +851,11 @@ defm BUFFER_LOAD_DWORDX4 : MUBUF_Pseudo_Loads <
   "buffer_load_dwordx4", VReg_128, v4i32
 >;
 
-defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_UBYTE", i32, az_extloadi8_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_UBYTE", i32, extloadi8_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_UBYTE", i32, zextloadi8_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_SBYTE", i32, sextloadi8_global>;
-defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_USHORT", i32, az_extloadi16_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_USHORT", i32, extloadi16_global>;
+defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_USHORT", i32, zextloadi16_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_SSHORT", i32, sextloadi16_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_DWORD", i32, load_global>;
 defm : MUBUF_Pseudo_Load_Pats<"BUFFER_LOAD_DWORDX2", v2i32, load_global>;
@@ -1437,9 +1439,11 @@ multiclass MUBUFLoad_Atomic_Pattern <MUBUF_Pseudo Instr_ADDR64, MUBUF_Pseudo Ins
 
 let SubtargetPredicate = isGFX6GFX7 in {
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_SBYTE_ADDR64, i32, sextloadi8_constant>;
-def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_UBYTE_ADDR64, i32, az_extloadi8_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_UBYTE_ADDR64, i32, extloadi8_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_UBYTE_ADDR64, i32, zextloadi8_constant>;
 def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_SSHORT_ADDR64, i32, sextloadi16_constant>;
-def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, az_extloadi16_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, extloadi16_constant>;
+def : MUBUFLoad_PatternADDR64 <BUFFER_LOAD_USHORT_ADDR64, i32, zextloadi16_constant>;
 
 defm : MUBUFLoad_Atomic_Pattern <BUFFER_LOAD_DWORD_ADDR64, BUFFER_LOAD_DWORD_OFFSET, i32, atomic_load_global>;
 defm : MUBUFLoad_Atomic_Pattern <BUFFER_LOAD_DWORDX2_ADDR64, BUFFER_LOAD_DWORDX2_OFFSET, i64, atomic_load_global>;
@@ -1458,9 +1462,11 @@ multiclass MUBUFLoad_Pattern <MUBUF_Pseudo Instr_OFFSET, ValueType vt,
 let OtherPredicates = [Has16BitInsts] in {
 
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_OFFSET, i16, sextloadi8_constant>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, az_extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, extloadi8_constant>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, zextloadi8_constant>;
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_SBYTE_OFFSET, i16, sextloadi8_global>;
-defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, az_extloadi8_global>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, extloadi8_global>;
+defm : MUBUFLoad_Pattern <BUFFER_LOAD_UBYTE_OFFSET, i16, zextloadi8_global>;
 
 defm : MUBUFLoad_Pattern <BUFFER_LOAD_USHORT_OFFSET, i16, load_global>;
 
@@ -1497,11 +1503,14 @@ multiclass MUBUFScratchLoadPat_D16 <MUBUF_Pseudo InstrOffen,
 }
 
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_SBYTE_OFFEN, BUFFER_LOAD_SBYTE_OFFSET, i32, sextloadi8_private>;
-defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i32, az_extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i32, extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i32, zextloadi8_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_SBYTE_OFFEN, BUFFER_LOAD_SBYTE_OFFSET, i16, sextloadi8_private>;
-defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i16, az_extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i16, extloadi8_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_UBYTE_OFFEN, BUFFER_LOAD_UBYTE_OFFSET, i16, zextloadi8_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_SSHORT_OFFEN, BUFFER_LOAD_SSHORT_OFFSET, i32, sextloadi16_private>;
-defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i32, az_extloadi16_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i32, extloadi16_private>;
+defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i32, zextloadi16_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_USHORT_OFFEN, BUFFER_LOAD_USHORT_OFFSET, i16, load_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_DWORD_OFFEN, BUFFER_LOAD_DWORD_OFFSET, i32, load_private>;
 defm : MUBUFScratchLoadPat <BUFFER_LOAD_DWORDX2_OFFEN, BUFFER_LOAD_DWORDX2_OFFSET, v2i32, load_private>;
diff --git a/lib/Target/AMDGPU/EvergreenInstructions.td b/lib/Target/AMDGPU/EvergreenInstructions.td
index fbe7d757f35..0550092ce1d 100644
--- a/lib/Target/AMDGPU/EvergreenInstructions.td
+++ b/lib/Target/AMDGPU/EvergreenInstructions.td
@@ -30,10 +30,6 @@ class EGOrCaymanPat<dag pattern, dag result> : AMDGPUPat<pattern, result> {
   let SubtargetPredicate = isEGorCayman;
 }
 
-// FIXME: These are deprecated
-def az_extloadi8_local : LocalLoad <az_extloadi8>;
-def az_extloadi16_local : LocalLoad <az_extloadi16>;
-
 //===----------------------------------------------------------------------===//
 // Evergreen / Cayman store instructions
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/AMDGPU/FLATInstructions.td b/lib/Target/AMDGPU/FLATInstructions.td
index df334790b85..4070d94dd4a 100644
--- a/lib/Target/AMDGPU/FLATInstructions.td
+++ b/lib/Target/AMDGPU/FLATInstructions.td
@@ -767,11 +767,14 @@ class FlatSignedAtomicPat <FLAT_Pseudo inst, SDPatternOperator node, ValueType v
 
 let OtherPredicates = [HasFlatAddressSpace] in {
 
-def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, extloadi8_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, zextloadi8_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_flat, i32>;
-def : FlatLoadPat <FLAT_LOAD_UBYTE, az_extloadi8_flat, i16>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, extloadi8_flat, i16>;
+def : FlatLoadPat <FLAT_LOAD_UBYTE, zextloadi8_flat, i16>;
 def : FlatLoadPat <FLAT_LOAD_SBYTE, sextloadi8_flat, i16>;
-def : FlatLoadPat <FLAT_LOAD_USHORT, az_extloadi16_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, extloadi16_flat, i32>;
+def : FlatLoadPat <FLAT_LOAD_USHORT, zextloadi16_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_USHORT, load_flat, i16>;
 def : FlatLoadPat <FLAT_LOAD_SSHORT, sextloadi16_flat, i32>;
 def : FlatLoadPat <FLAT_LOAD_DWORD, load_flat, i32>;
@@ -849,11 +852,14 @@ def atomic_pk_fadd_global : global_binary_atomic_op_frag<SIglobal_atomic_pk_fadd
 
 let OtherPredicates = [HasFlatGlobalInsts], AddedComplexity = 10 in {
 
-def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, az_extloadi8_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, extloadi8_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, zextloadi8_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_SBYTE, sextloadi8_global, i32>;
-def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, az_extloadi8_global, i16>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, extloadi8_global, i16>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_UBYTE, zextloadi8_global, i16>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_SBYTE, sextloadi8_global, i16>;
-def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, az_extloadi16_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, extloadi16_global, i32>;
+def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, zextloadi16_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_SSHORT, sextloadi16_global, i32>;
 def : FlatLoadSignedPat <GLOBAL_LOAD_USHORT, load_global, i16>;
 
diff --git a/lib/Target/AMDGPU/R600Instructions.td b/lib/Target/AMDGPU/R600Instructions.td
index dcbedbd8cfd..d3ce7ffd673 100644
--- a/lib/Target/AMDGPU/R600Instructions.td
+++ b/lib/Target/AMDGPU/R600Instructions.td
@@ -295,6 +295,32 @@ class VTX_READ <string name, dag outs, list<dag> pattern>
   let VTXInst = 1;
 }
 
+// FIXME: Deprecated.
+class AZExtLoadBase <SDPatternOperator ld_node>: PatFrag<(ops node:$ptr),
+                                              (ld_node node:$ptr), [{
+  LoadSDNode *L = cast<LoadSDNode>(N);
+  return L->getExtensionType() == ISD::ZEXTLOAD ||
+         L->getExtensionType() == ISD::EXTLOAD;
+}]>;
+
+def az_extload : AZExtLoadBase <unindexedload>;
+
+def az_extloadi8 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
+  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i8;
+}]>;
+
+def az_extloadi16 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
+  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i16;
+}]>;
+
+def az_extloadi32 : PatFrag<(ops node:$ptr), (az_extload node:$ptr), [{
+  return cast<LoadSDNode>(N)->getMemoryVT() == MVT::i32;
+}]>;
+
+// FIXME: These are deprecated
+def az_extloadi8_local : LocalLoad <az_extloadi8>;
+def az_extloadi16_local : LocalLoad <az_extloadi16>;
+
 class LoadParamFrag <PatFrag load_type> : PatFrag <
   (ops node:$ptr), (load_type node:$ptr),
   [{ return isConstantLoad(cast<LoadSDNode>(N), 0) ||