AMDGPU: Define HasExpOrExportInsts for export instruction definitions.

[changpeng]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Changpeng Fang (changpeng)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/84083.diff

4 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPU.td (+3)
• (modified) llvm/lib/Target/AMDGPU/EXPInstructions.td (+3-3)
• (modified) llvm/lib/Target/AMDGPU/GCNSubtarget.h (+4)
• (modified) llvm/lib/Target/AMDGPU/SOPInstructions.td (+2)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPU.td b/llvm/lib/Target/AMDGPU/AMDGPU.td
index 7c278fd574ede5..814ac0b93fcf01 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPU.td
+++ b/llvm/lib/Target/AMDGPU/AMDGPU.td
@@ -1874,6 +1874,9 @@ def D16PreservesUnusedBits :
 def LDSRequiresM0Init : Predicate<"Subtarget->ldsRequiresM0Init()">;
 def NotLDSRequiresM0Init : Predicate<"!Subtarget->ldsRequiresM0Init()">;
 
+def HasExpOrExportInsts : Predicate<"Subtarget->hasExpOrExportInsts()">,
+  AssemblerPredicate<(all_of (not FeatureGFX90AInsts))>;
+
 def HasDSAddTid : Predicate<"Subtarget->getGeneration() >= AMDGPUSubtarget::GFX9">,
   AssemblerPredicate<(all_of FeatureGFX9Insts)>;
 
diff --git a/llvm/lib/Target/AMDGPU/EXPInstructions.td b/llvm/lib/Target/AMDGPU/EXPInstructions.td
index 0a1e544949104a..5e3f555e441168 100644
--- a/llvm/lib/Target/AMDGPU/EXPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/EXPInstructions.td
@@ -58,12 +58,12 @@ class EXP_Real_Row<string pseudo, int subtarget, string name = "exp", EXP_Pseudo
 
 // DONE variants have mayLoad = 1.
 // ROW variants have an implicit use of M0.
-let SubtargetPredicate = isNotGFX90APlus in {
+let SubtargetPredicate = HasExpOrExportInsts in {
 def EXP          : EXP_Pseudo<0, 0>;
 def EXP_DONE     : EXP_Pseudo<0, 1>;
 def EXP_ROW      : EXP_Pseudo<1, 0>;
 def EXP_ROW_DONE : EXP_Pseudo<1, 1>;
-} // let SubtargetPredicate = isNotGFX90APlus
+} // let SubtargetPredicate = HasExpOrExportInsts
 
 //===----------------------------------------------------------------------===//
 // SI, VI, GFX10.
@@ -117,7 +117,7 @@ multiclass EXP_Real_gfx11 {
 multiclass VEXPORT_Real_gfx12 {
   defvar ps = !cast<EXP_Pseudo>(NAME);
   def _gfx12 : EXP_Real_Row<NAME, SIEncodingFamily.GFX12, "export">,
-    EXPe_Row, MnemonicAlias<"exp", "export">, Requires<[isGFX12Plus]> {
+    EXPe_Row, MnemonicAlias<"exp", "export">, Requires<[isGFX12Plus, HasExpOrExportInsts]> {
     let AssemblerPredicate = isGFX12Only;
     let DecoderNamespace = "GFX12";
     let row = ps.row;
diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index a933c16b6ed516..bb0ccfce324fd8 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -647,6 +647,10 @@ class GCNSubtarget final : public AMDGPUGenSubtargetInfo,
   // BUFFER/FLAT/GLOBAL_ATOMIC_ADD/MIN/MAX_F64
   bool hasBufferFlatGlobalAtomicsF64() const { return hasGFX90AInsts(); }
 
+  bool hasExpOrExportInsts() const {
+    return !hasGFX940Insts();
+  }
+
   // DS_ADD_F64/DS_ADD_RTN_F64
   bool hasLdsAtomicAddF64() const { return hasGFX90AInsts(); }
 
diff --git a/llvm/lib/Target/AMDGPU/SOPInstructions.td b/llvm/lib/Target/AMDGPU/SOPInstructions.td
index ff79538a86ae18..e14f7f95d467a1 100644
--- a/llvm/lib/Target/AMDGPU/SOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SOPInstructions.td
@@ -1705,6 +1705,7 @@ let SubtargetPredicate = isGFX10Plus in {
 } // End SubtargetPredicate = isGFX10Plus
 
 let SubtargetPredicate = isGFX11Plus in {
+let OtherPredicates = [HasExpOrExportInsts] in
   def S_WAIT_EVENT : SOPP_Pseudo<"s_wait_event", (ins s16imm:$simm16),
                                  "$simm16"> {
                                    let hasSideEffects = 1;
@@ -1737,6 +1738,7 @@ let OtherPredicates = [HasImageInsts] in {
     SOPP_Pseudo<"s_wait_bvhcnt", (ins s16imm:$simm16), "$simm16",
                 [(int_amdgcn_s_wait_bvhcnt timm:$simm16)]>;
 } // End OtherPredicates = [HasImageInsts].
+let OtherPredicates = [HasExpOrExportInsts] in
   def S_WAIT_EXPCNT :
     SOPP_Pseudo<"s_wait_expcnt", (ins s16imm:$simm16), "$simm16",
                 [(int_amdgcn_s_wait_expcnt timm:$simm16)]>;

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff e96c0c1d5e0a9916098b1a31acb006ea6c1108fb b726f08ab3489689799108312dd20ecd2546d80c -- llvm/lib/Target/AMDGPU/GCNSubtarget.h

View the diff from clang-format here.

diff --git a/llvm/lib/Target/AMDGPU/GCNSubtarget.h b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
index bb0ccfce32..36e38b9954 100644
--- a/llvm/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/GCNSubtarget.h
@@ -647,9 +647,7 @@ public:
   // BUFFER/FLAT/GLOBAL_ATOMIC_ADD/MIN/MAX_F64
   bool hasBufferFlatGlobalAtomicsF64() const { return hasGFX90AInsts(); }
 
-  bool hasExpOrExportInsts() const {
-    return !hasGFX940Insts();
-  }
+  bool hasExpOrExportInsts() const { return !hasGFX940Insts(); }
 
   // DS_ADD_F64/DS_ADD_RTN_F64
   bool hasLdsAtomicAddF64() const { return hasGFX90AInsts(); }

[jayfoad]

Please let's just call it HasExportInsts. "exp" is just short for "export" so the longer name doesn't add any meaning.

[rampitec]

Please let's just call it HasExportInsts. "exp" is just short for "export" so the longer name doesn't add any meaning.

No objections.