[MLIR][ROCDL] Add ops for LDS read transpose and global to LDS intrinsics #123530
Conversation
|
Thank you for submitting a Pull Request (PR) to the LLVM Project! This PR will be automatically labeled and the relevant teams will be notified. If you wish to, you can add reviewers by using the "Reviewers" section on this page. If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers. If you have further questions, they may be answered by the LLVM GitHub User Guide. You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums. |
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-llvm Author: None (plognjen) ChangesThis PR adds missing ds.read.tr4.b64, ds.read.tr8.b64, ds.read.tr6.b96, Full diff: https://github.com/llvm/llvm-project/pull/123530.diff 3 Files Affected:
diff --git a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
index 71dac3ad39b7b1..0e23c183b4d923 100644
--- a/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
+++ b/mlir/include/mlir/Dialect/LLVMIR/ROCDLOps.td
@@ -412,6 +412,36 @@ def ROCDL_wmma_i32_16x16x16_iu4 : ROCDL_Wmma_IntrOp<"wmma.i32.16x16x16.iu4", [1]
def ROCDL_wmma_f32_16x16x16_fp8 : ROCDL_Wmma_IntrOp<"wmma.f32.16x16x16.fp8_fp8", [1]>;
def ROCDL_wmma_f32_16x16x16_bf8 : ROCDL_Wmma_IntrOp<"wmma.f32.16x16x16.bf8_bf8", [1]>;
+//===---------------------------------------------------------------------===//
+// LDS transpose intrinsics
+
+def ROCDLGlobalBuffer : LLVM_PointerInAddressSpace<1>;
+def ROCDLBufferLDS : LLVM_PointerInAddressSpace<3>;
+
+class ROCDL_LDS_Read_Tr_IntrOp<string mnemonic> :
+ ROCDL_IntrOp<mnemonic, [1], [], [], 1>,
+ Arguments<(ins Arg<ROCDLBufferLDS, "", [MemRead]>:$ptr)>{
+ let assemblyFormat = "$ptr attr-dict `:` type($ptr) `->` type($res)";
+ }
+
+def ROCDL_ds_read_tr4_b64 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr4.b64">;
+def ROCDL_ds_read_tr8_b64 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr8.b64">;
+def ROCDL_ds_read_tr6_b96 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr6.b96">;
+def ROCDL_ds_read_tr16_b64 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr16.b64">;
+
+//===---------------------------------------------------------------------===//
+// Global load to LDS intrinsic
+
+def ROCDL_GlobalLoadLDSOp :
+ ROCDL_IntrOp<"global.load.lds", [], [], [], 0>,
+ Arguments<(ins Arg<ROCDLGlobalBuffer, "", [MemRead]>:$globalPtr,
+ Arg<ROCDLBufferLDS, "", [MemWrite]>:$ldsPtr,
+ I32:$size,
+ I32:$offset,
+ I32:$aux)> {
+ let assemblyFormat = "operands attr-dict";
+}
+
//===---------------------------------------------------------------------===//
// Operations on raw buffer resources (stride of 0, bounds checks either off or in
// raw buffer mode).
diff --git a/mlir/test/Dialect/LLVMIR/rocdl.mlir b/mlir/test/Dialect/LLVMIR/rocdl.mlir
index 92789246edb4f3..c80ebebaafe3ad 100644
--- a/mlir/test/Dialect/LLVMIR/rocdl.mlir
+++ b/mlir/test/Dialect/LLVMIR/rocdl.mlir
@@ -227,6 +227,32 @@ func.func @rocdl.xdlops(%arg0 : f32, %arg1 : f32,
llvm.return
}
+llvm.func @rocdl.ds.read.tr(%ptr : !llvm.ptr<3>) -> vector<4xf16> {
+ // CHECK-LABEL: rocdl.ds.read.tr
+ // CHECK: rocdl.ds.read.tr4.b64 {{.*}} : <3> -> vector<2xi32>
+ %r0 = rocdl.ds.read.tr4.b64 %ptr : !llvm.ptr<3> -> vector<2xi32>
+ // CHECK: rocdl.ds.read.tr6.b96 {{.*}} : <3> -> vector<3xi32>
+ %r1 = rocdl.ds.read.tr6.b96 %ptr : !llvm.ptr<3> -> vector<3xi32>
+ // CHECK: rocdl.ds.read.tr8.b64 {{.*}} : <3> -> vector<2xi32>
+ %r2 = rocdl.ds.read.tr8.b64 %ptr : !llvm.ptr<3> -> vector<2xi32>
+ // CHECK: rocdl.ds.read.tr16.b64 {{.*}} : <3> -> vector<4xf16>
+ %r3 = rocdl.ds.read.tr16.b64 %ptr : !llvm.ptr<3> -> vector<4xf16>
+ // CHECK: rocdl.ds.read.tr16.b64 {{.*}} : <3> -> vector<4xbf16>
+ %r4 = rocdl.ds.read.tr16.b64 %ptr : !llvm.ptr<3> -> vector<4xbf16>
+ llvm.return %r3 : vector<4xf16>
+}
+
+llvm.func @rocdl.global.load.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3>) {
+ %aux = llvm.mlir.constant(0 : i32) : i32
+ %offset = llvm.mlir.constant(0 : i32) : i32
+ %size = llvm.mlir.constant(10 : i32) : i32
+
+ //CHECK: rocdl.global.load.lds %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}
+ rocdl.global.load.lds %src, %dst, %size, %offset, %aux
+
+ llvm.return
+}
+
llvm.func @rocdl.make.buffer.rsrc(%ptr : !llvm.ptr,
%stride : i16,
%numRecords : i32,
diff --git a/mlir/test/Target/LLVMIR/rocdl.mlir b/mlir/test/Target/LLVMIR/rocdl.mlir
index 0620c23b5fdad7..996e0e34c790c9 100644
--- a/mlir/test/Target/LLVMIR/rocdl.mlir
+++ b/mlir/test/Target/LLVMIR/rocdl.mlir
@@ -424,6 +424,30 @@ llvm.func @rocdl.wmma(%arg0 : vector<8xf32>, %arg1 : vector<16 x f16>, %arg2 : v
llvm.return %r0 : vector<8xf32>
}
+llvm.func @rocdl.ds.read.tr(%ptr : !llvm.ptr<3>) -> vector<4xf16> {
+ // CHECK-LABEL: rocdl.ds.read.tr
+ // CHECK: call <2 x i32> @llvm.amdgcn.ds.read.tr4.b64.v2i32(ptr addrspace(3) %0)
+ %r0 = rocdl.ds.read.tr4.b64 %ptr : !llvm.ptr<3> -> vector<2xi32>
+ // CHECK: call <3 x i32> @llvm.amdgcn.ds.read.tr6.b96.v3i32(ptr addrspace(3) %0)
+ %r1 = rocdl.ds.read.tr6.b96 %ptr : !llvm.ptr<3> -> vector<3xi32>
+ // CHECK: call <2 x i32> @llvm.amdgcn.ds.read.tr8.b64.v2i32(ptr addrspace(3) %0)
+ %r2 = rocdl.ds.read.tr8.b64 %ptr : !llvm.ptr<3> -> vector<2xi32>
+ // CHECK: call <4 x half> @llvm.amdgcn.ds.read.tr16.b64.v4f16(ptr addrspace(3) %0)
+ %r3 = rocdl.ds.read.tr16.b64 %ptr : !llvm.ptr<3> -> vector<4xf16>
+ // CHECK: call <4 x bfloat> @llvm.amdgcn.ds.read.tr16.b64.v4bf16(ptr addrspace(3) %0)
+ %r4 = rocdl.ds.read.tr16.b64 %ptr : !llvm.ptr<3> -> vector<4xbf16>
+ llvm.return %r3 : vector<4xf16>
+}
+
+llvm.func @rocdl.global.load.lds(%src : !llvm.ptr<1>, %dst: !llvm.ptr<3>) {
+ %aux = llvm.mlir.constant(0 : i32) : i32
+ %offset = llvm.mlir.constant(0 : i32) : i32
+ %size = llvm.mlir.constant(10 : i32) : i32
+ //CHECK: call void @llvm.amdgcn.global.load.lds
+ rocdl.global.load.lds %src, %dst, %size, %offset, %aux
+ llvm.return
+}
+
llvm.func @rocdl.make.buffer.rsrc(%ptr : !llvm.ptr,
%stride : i16,
%numRecords : i32,
|
| //===---------------------------------------------------------------------===// | ||
| // LDS transpose intrinsics | ||
|
|
||
| def ROCDLGlobalBuffer : LLVM_PointerInAddressSpace<1>; | ||
| def ROCDLBufferLDS : LLVM_PointerInAddressSpace<3>; | ||
|
|
||
| class ROCDL_LDS_Read_Tr_IntrOp<string mnemonic> : | ||
| ROCDL_IntrOp<mnemonic, [1], [], [], 1>, | ||
| Arguments<(ins Arg<ROCDLBufferLDS, "", [MemRead]>:$ptr)>{ | ||
| let assemblyFormat = "$ptr attr-dict `:` type($ptr) `->` type($res)"; | ||
| } | ||
|
|
||
| def ROCDL_ds_read_tr4_b64 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr4.b64">; | ||
| def ROCDL_ds_read_tr8_b64 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr8.b64">; | ||
| def ROCDL_ds_read_tr6_b96 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr6.b96">; | ||
| def ROCDL_ds_read_tr16_b64 : ROCDL_LDS_Read_Tr_IntrOp<"ds.read.tr16.b64">; | ||
|
|
||
| //===---------------------------------------------------------------------===// | ||
| // Global load to LDS intrinsic | ||
|
|
||
| def ROCDL_GlobalLoadLDSOp : | ||
| ROCDL_IntrOp<"global.load.lds", [], [], [], 0>, | ||
| Arguments<(ins Arg<ROCDLGlobalBuffer, "", [MemRead]>:$globalPtr, | ||
| Arg<ROCDLBufferLDS, "", [MemWrite]>:$ldsPtr, | ||
| I32:$size, | ||
| I32:$offset, | ||
| I32:$aux)> { | ||
| let assemblyFormat = "operands attr-dict"; | ||
| } |
There was a problem hiding this comment.
Can you mention what architecture these are available from? I don't see them in the gfx940 ISA manual.
|
@kuhar @Groverkss Thanks for the review! |
|
@plognjen Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
This PR adds missing ds.read.tr4.b64, ds.read.tr8.b64, ds.read.tr6.b96,
ds.read.tr16.b64 and global.load.lds ops to
the ROCDL dialect.
The ops are converted to the corresponding intrinsic calls during the
translation from MLIR to LLVM IRs.