[DIALECT] Rename triton_gpu to ttg and triton_nvidia_gpu to ttng (#5266)

It may cause changes for downstream tasks but we think it's beneficial
to shorten dialect name and make them consistent. That is, we are using
`tt` to represent the `triton` dialect.

Author: Jokeren

bin/triton-tensor-layout.cpp

@@ -22,16 +22,16 @@ using namespace mlir;
 // clang-format off
 // Example usage:
 //
-// triton-tensor-layout -l "#triton_gpu.nvidia_mma<{versionMajor = 3, versionMinor = 0, warpsPerCTA = [8, 1], CTAsPerCGA = [1, 1], CTASplitNum = [1, 1], CTAOrder = [1, 0], instrShape = [16, 256, 32]}>" -t "tensor<128x256xf16>"
+// triton-tensor-layout -l "#ttg.nvidia_mma<{versionMajor = 3, versionMinor = 0, warpsPerCTA = [8, 1], CTAsPerCGA = [1, 1], CTASplitNum = [1, 1], CTAOrder = [1, 0], instrShape = [16, 256, 32]}>" -t "tensor<128x256xf16>"
 //
 // triton-tensor-layout -i input.mlir -t "tensor<1x128x128xf16>" -o output.txt
 //
 // triton-tensor-layout -i input.mlir -t "tensor<1x128x128xf16>" -o output.txt -alias-names="blocked,mma" -use-hw-view
 //
 // An input file usually looks like:
 // '''
-// #mma = #triton_gpu.amd_mfma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 1, 8], instrShape = [32, 32], isTransposed = false}>
-// #blocked = #triton_gpu.blocked<{sizePerThread = [1, 8, 1], threadsPerWarp = [1, 16, 4], warpsPerCTA = [1, 1, 8], order = [0, 1, 2]}>
+// #mma = #ttg.amd_mfma<{versionMajor = 2, versionMinor = 0, warpsPerCTA = [1, 1, 8], instrShape = [32, 32], isTransposed = false}>
+// #blocked = #ttg.blocked<{sizePerThread = [1, 8, 1], threadsPerWarp = [1, 16, 4], warpsPerCTA = [1, 1, 8], order = [0, 1, 2]}>
 // '''
 // clang-format on
 
@@ -83,7 +83,7 @@ LogicalResult layoutPrint(RankedTensorType tensorType, raw_ostream &os) {
   StringRef dialectName = tensorType.getEncoding().getDialect().getNamespace();
 
   // Dispatch to the corresponding dialect helper function to print the layout.
-  if (dialectName == "triton_gpu") {
+  if (dialectName == "ttg") {
     os << triton::gpu::getLayoutStr(tensorType, UseHWPointOfView);
     return success();
   }

include/triton/Analysis/Allocation.h

@@ -180,8 +180,8 @@ class Allocation {
 private:
   /// A class that represents a shared memory buffer
   struct BufferT {
-    /// Explicit: triton_gpu.local_alloc
-    /// Scratch: triton_gpu.convert_layout
+    /// Explicit: ttg.local_alloc
+    /// Scratch: ttg.convert_layout
     /// Virtual: triton.call
     enum class BufferKind { Explicit, Scratch, Virtual };
 

include/triton/Conversion/TritonGPUToLLVM/Utility.h

@@ -400,7 +400,7 @@ inline Value getGlobalScratchPtr(Location loc, RewriterBase &rewriter,
 
   ModuleOp mod = funcOp.getOperation()->getParentOfType<ModuleOp>();
   auto allocSizeAttr = mod.getOperation()->getAttrOfType<mlir::IntegerAttr>(
-      "triton_gpu.global_scratch_memory_size");
+      "ttg.global_scratch_memory_size");
   if (!allocSizeAttr) {
     return gmemBase;
   }

include/triton/Conversion/TritonToTritonGPU/TritonToTritonGPUPass.h

@@ -12,11 +12,11 @@ template <typename T> class OperationPass;
 
 namespace triton {
 
-constexpr static char AttrNumWarpsName[] = "triton_gpu.num-warps";
-constexpr static char AttrNumCTAsName[] = "triton_gpu.num-ctas";
-constexpr static char AttrTargetName[] = "triton_gpu.target";
+constexpr static char AttrNumWarpsName[] = "ttg.num-warps";
+constexpr static char AttrNumCTAsName[] = "ttg.num-ctas";
+constexpr static char AttrTargetName[] = "ttg.target";
 
-constexpr static char AttrNumThreadsPerWarp[] = "triton_gpu.threads-per-warp";
+constexpr static char AttrNumThreadsPerWarp[] = "ttg.threads-per-warp";
 
 // Create the pass with numWarps passed from cl::opt.
 std::unique_ptr<OperationPass<ModuleOp>> createConvertTritonToTritonGPUPass();

include/triton/Dialect/TritonGPU/IR/CMakeLists.txt

@@ -1,12 +1,12 @@
 set(MLIR_BINARY_DIR ${CMAKE_BINARY_DIR})
 
 set(LLVM_TARGET_DEFINITIONS TritonGPUOps.td)
-mlir_tablegen(Dialect.h.inc -gen-dialect-decls -dialect=triton_gpu)
-mlir_tablegen(Dialect.cpp.inc -gen-dialect-defs -dialect=triton_gpu)
+mlir_tablegen(Dialect.h.inc -gen-dialect-decls -dialect=ttg)
+mlir_tablegen(Dialect.cpp.inc -gen-dialect-defs -dialect=ttg)
 mlir_tablegen(Ops.h.inc -gen-op-decls)
 mlir_tablegen(Ops.cpp.inc -gen-op-defs)
-mlir_tablegen(Types.h.inc -gen-typedef-decls -typedefs-dialect=triton_gpu)
-mlir_tablegen(Types.cpp.inc -gen-typedef-defs -typedefs-dialect=triton_gpu)
+mlir_tablegen(Types.h.inc -gen-typedef-decls -typedefs-dialect=ttg)
+mlir_tablegen(Types.cpp.inc -gen-typedef-defs -typedefs-dialect=ttg)
 add_mlir_doc(TritonGPUDialect TritonGPUDialect dialects/ -gen-dialect-doc)
 add_mlir_doc(TritonGPUOps TritonGPUOps dialects/ -gen-op-doc)
 add_public_tablegen_target(TritonGPUTableGen)

include/triton/Dialect/TritonGPU/IR/TritonGPUAttrDefs.td

@@ -616,7 +616,7 @@ Example 1, a row-major coalesced layout may partition a 16x16 tensor over 2 warp
 
 for
 
-#triton_gpu.blocked_layout<{
+#ttg.blocked_layout<{
   sizePerThread = {2, 2}
   threadsPerWarp = {8, 4}
   warpsPerCTA = {1, 2}
@@ -642,7 +642,7 @@ Example 2, a row-major coalesced layout may partition a 32x32 tensor over 2 warp
 [ 28 28 29 29 30 30 31 31 ; 60 60 61 61 62 62 63 63  28 28 29 29 30 30 31 31 ; 60 60 61 61 62 62 63 63 ]
 for
 
-#triton_gpu.blocked_layout<{
+#ttg.blocked_layout<{
   sizePerThread = {2, 2}
   threadsPerWarp = {8, 4}
   warpsPerCTA = {1, 2}
@@ -672,7 +672,7 @@ CTA [1,0]                                              CTA [1,1]
 [ 28 28 29 29 30 30 31 31 ; 60 60 61 61 62 62 63 63 ]  [ 28 28 29 29 30 30 31 31 ; 60 60 61 61 62 62 63 63 ]
 for
 
-#triton_gpu.blocked_layout<{
+#ttg.blocked_layout<{
   sizePerThread = {2, 2}
   threadsPerWarp = {8, 4}
   warpsPerCTA = {1, 2}

include/triton/Dialect/TritonGPU/IR/TritonGPUDialect.td

@@ -4,7 +4,7 @@
 include "mlir/IR/OpBase.td"
 
 def TritonGPU_Dialect : Dialect {
-  let name = "triton_gpu";
+  let name = "ttg";
 
   let cppNamespace = "::mlir::triton::gpu";
 
@@ -21,24 +21,24 @@ def TritonGPU_Dialect : Dialect {
   ];
 
   let extraClassDeclaration = [{
-    static std::string getNumWarpsAttrName() { return "triton_gpu.num-warps"; }
+    static std::string getNumWarpsAttrName() { return "ttg.num-warps"; }
     static int getNumWarps(ModuleOp mod) {
-      if (!mod->hasAttr("triton_gpu.num-warps"))
+      if (!mod->hasAttr("ttg.num-warps"))
         llvm::report_fatal_error(
-            "TritonGPU module should contain a triton_gpu.num-warps attribute");
-      return cast<IntegerAttr>(mod->getAttr("triton_gpu.num-warps")).getInt();
+            "TritonGPU module should contain a ttg.num-warps attribute");
+      return cast<IntegerAttr>(mod->getAttr("ttg.num-warps")).getInt();
     }
     static int getNumCTAs(ModuleOp mod) {
-      if (!mod->hasAttr("triton_gpu.num-ctas"))
+      if (!mod->hasAttr("ttg.num-ctas"))
         return 1;
-      return cast<IntegerAttr>(mod->getAttr("triton_gpu.num-ctas")).getInt();
+      return cast<IntegerAttr>(mod->getAttr("ttg.num-ctas")).getInt();
     }
     void registerTypes();
 
-    static std::string getThreadsPerWarpAttrName() { return "triton_gpu.threads-per-warp"; }
+    static std::string getThreadsPerWarpAttrName() { return "ttg.threads-per-warp"; }
 
     static int getThreadsPerWarp(ModuleOp mod) {
-      Attribute threadsPerWarp = mod->getDiscardableAttr("triton_gpu.threads-per-warp");
+      Attribute threadsPerWarp = mod->getDiscardableAttr("ttg.threads-per-warp");
       if(!threadsPerWarp) {
         return 32;
       }

include/triton/Dialect/TritonGPU/IR/TritonGPUOps.td

@@ -188,13 +188,13 @@ def TTG_LocalDeallocOp : TTG_Op<"local_dealloc", [MemoryEffects<[MemFree<SharedM
 
     Because we assume a memdesc is dead at the first point that post-dominates
     its uses, ops that wait for an async operation on a memdesc to complete
-    (such as triton_nvidia_gpu.warp_group_dot_wait) should also take the memdesc as an
+    (such as ttng.warp_group_dot_wait) should also take the memdesc as an
     operand.
   }];
 
   let arguments = (ins TTG_MemDescType:$src);
 
-  // Use qualified() otherwise "!triton_gpu.memdesc<X>" is printed as "<X>".
+  // Use qualified() otherwise "!ttg.memdesc<X>" is printed as "<X>".
   let assemblyFormat = [{$src attr-dict `:` qualified(type($src))}];
 }
 
@@ -215,7 +215,7 @@ def TTG_MemDescSubviewOp : TTG_Op<"memdesc_subview", [Pure]> {
   let arguments = (
     ins TTG_MemDescType:$src, Variadic<I32>:$offsets);
 
-  // Use qualified() otherwise "!triton_gpu.memdesc<X>" is printed as "<X>".
+  // Use qualified() otherwise "!ttg.memdesc<X>" is printed as "<X>".
   let assemblyFormat = [{$src `[` $offsets `]` attr-dict `:` qualified(type($src)) `->` qualified(type($result))}];
 
   let results = (outs TTG_MemDescType:$result);
@@ -262,7 +262,7 @@ def TTG_LocalLoadOp : TTG_Op<"local_load", [DeclareOpInterfaceMethods<MemoryEffe
       build($_builder, $_state, retType, src, /*token=*/static_cast<mlir::Value>(nullptr));
       }]>];
 
-  // Use qualified() otherwise "!triton_gpu.memdesc<X>" is printed as "<X>".
+  // Use qualified() otherwise "!ttg.memdesc<X>" is printed as "<X>".
   let assemblyFormat = [{$src (`token` $token^)? attr-dict `:` qualified(type($src)) `->` type($result)}];
 
   let results = (outs TT_Tensor:$result);
@@ -277,7 +277,7 @@ def TTG_LocalStoreOp : TTG_Op<"local_store", [DeclareOpInterfaceMethods<MemoryEf
   let arguments = (ins TT_Tensor:$src, TTG_MemDescType:$dst);
 
   let hasVerifier = 1;
-  // Use qualified() otherwise "!triton_gpu.memdesc<X>" is printed as "<X>".
+  // Use qualified() otherwise "!ttg.memdesc<X>" is printed as "<X>".
   let assemblyFormat = [{
     $src `,` $dst attr-dict `:` type($src) `->` qualified(type($dst))
   }];

include/triton/Dialect/TritonNvidiaGPU/IR/CMakeLists.txt

@@ -1,12 +1,12 @@
 set(MLIR_BINARY_DIR ${CMAKE_BINARY_DIR})
 
 set(LLVM_TARGET_DEFINITIONS TritonNvidiaGPUOps.td)
-mlir_tablegen(Dialect.h.inc -gen-dialect-decls -dialect=triton_nvidia_gpu)
-mlir_tablegen(Dialect.cpp.inc -gen-dialect-defs -dialect=triton_nvidia_gpu)
+mlir_tablegen(Dialect.h.inc -gen-dialect-decls -dialect=ttng)
+mlir_tablegen(Dialect.cpp.inc -gen-dialect-defs -dialect=ttng)
 mlir_tablegen(Ops.h.inc -gen-op-decls)
 mlir_tablegen(Ops.cpp.inc -gen-op-defs)
-mlir_tablegen(Types.h.inc -gen-typedef-decls -typedefs-dialect=triton_nvidia_gpu)
-mlir_tablegen(Types.cpp.inc -gen-typedef-defs -typedefs-dialect=triton_nvidia_gpu)
+mlir_tablegen(Types.h.inc -gen-typedef-decls -typedefs-dialect=ttng)
+mlir_tablegen(Types.cpp.inc -gen-typedef-defs -typedefs-dialect=ttng)
 add_mlir_doc(TritonNvidiaGPUDialect TritonNvidiaGPUDialect dialects/ -gen-dialect-doc)
 add_mlir_doc(TritonNvidiaGPUOps TritonNvidiaGPUOps dialects/ -gen-op-doc)
 add_public_tablegen_target(TritonNvidiaGPUTableGen)

include/triton/Dialect/TritonNvidiaGPU/IR/TritonNvidiaGPUDialect.td

@@ -25,7 +25,7 @@
 include "mlir/IR/OpBase.td"
 
 def TritonNvidiaGPU_Dialect : Dialect {
-  let name = "triton_nvidia_gpu";
+  let name = "ttng";
 
   let cppNamespace = "::mlir::triton::nvidia_gpu";
 
@@ -43,18 +43,18 @@ def TritonNvidiaGPU_Dialect : Dialect {
   ];
 
   let extraClassDeclaration = [{
-    static std::string getNumWarpsAttrName() { return "triton_gpu.num-warps"; }
+    static std::string getNumWarpsAttrName() { return "ttg.num-warps"; }
     static int getNumWarps(ModuleOp mod) {
-      if(!mod->hasAttr("triton_gpu.num-warps"))
+      if(!mod->hasAttr("ttg.num-warps"))
         llvm::report_fatal_error(
-            "TritonGPU module should contain a triton_gpu.num-warps attribute");
-      return cast<IntegerAttr>(mod->getAttr("triton_gpu.num-warps")).getInt();
+            "TritonGPU module should contain a ttg.num-warps attribute");
+      return cast<IntegerAttr>(mod->getAttr("ttg.num-warps")).getInt();
     }
     static int getNumCTAs(ModuleOp mod) {
-      if(!mod->hasAttr("triton_gpu.num-ctas"))
+      if(!mod->hasAttr("ttg.num-ctas"))
         llvm::report_fatal_error(
-            "TritonGPU module should contain a triton_gpu.num-ctas attribute");
-      return cast<IntegerAttr>(mod->getAttr("triton_gpu.num-ctas")).getInt();
+            "TritonGPU module should contain a ttg.num-ctas attribute");
+      return cast<IntegerAttr>(mod->getAttr("ttg.num-ctas")).getInt();
     }
     void registerTypes();
   }];