[Mlir-commits] [mlir] [MLIR][XeGPU] Add unroll patterns for scatter ops (PR #143602)

[Jianhui Li]

================
@@ -158,4 +158,147 @@ gpu.module @test {
     %c = xegpu.dpas %a, %b : vector<32x32xf16>, vector<32x32xf16> -> vector<32x32xf32>
     gpu.return %c : vector<32x32xf32>
   }
+
+//-----
+
+  // CHECK-LABEL: test_create_tdesc_vec
+  // CHECK-SAME: [[arg0:%.+]]: ui64
+  // CHECK-COUNT-2: xegpu.create_tdesc [[arg0]], {{.*}} : ui64, vector<16xindex> -> !xegpu.tensor_desc<16xf32, #xegpu.scatter_tdesc_attr<>>
+  gpu.func @test_create_tdesc_vec(%src: ui64) -> !xegpu.tensor_desc<32xf32, #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>> {
+    %cst = arith.constant dense<[
+    0,   8,  16,  24,  32,  40,  48,  56,
+    64,  72,  80,  88,  96, 104, 112, 120,
+    128, 136, 144, 152, 160, 168, 176, 184,
+    192, 200, 208, 216, 224, 232, 240, 248 
+    ]> : vector<32xindex>
+    %tdesc = xegpu.create_tdesc %src, %cst : ui64, vector<32xindex> -> !xegpu.tensor_desc<32xf32,  #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>>
+    gpu.return %tdesc : !xegpu.tensor_desc<32xf32,  #xegpu.scatter_tdesc_attr<>,  #xegpu.layout<inst_data = [16]>>
+  }
+
+//-----
+
+  // CHECK-LABEL: test_create_tdesc_step
+  // CHECK-SAME: [[arg0:%.+]]: ui64
+  // CHECK-COUNT-2: xegpu.create_tdesc [[arg0]], {{.*}} : ui64, vector<16xindex> -> !xegpu.tensor_desc<16xf32, #xegpu.scatter_tdesc_attr<>>
+  gpu.func @test_create_tdesc_step(%src: ui64) -> !xegpu.tensor_desc<32xf32, #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>> {
+    %step = arith.constant dense<8> : vector<32xindex>
+    %seq = vector.step  : vector<32xindex>
+    %cst = arith.muli %seq, %step : vector<32xindex>
+    %tdesc = xegpu.create_tdesc %src, %cst : ui64, vector<32xindex> -> !xegpu.tensor_desc<32xf32,  #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>>
+    gpu.return %tdesc : !xegpu.tensor_desc<32xf32,  #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>>
+  }
+
+//-----
+
+    // CHECK-LABEL: test_load
+    // CHECK-SAME: [[arg0:%.+]]: ui64
+    // CHECK-COUNT-2: xegpu.create_tdesc [[arg0]], {{.*}} : ui64, vector<16xindex> -> !xegpu.tensor_desc<16xf32, #xegpu.scatter_tdesc_attr<>>
+    // CHECK-COUNT-2: xegpu.load  {{.*}} : !xegpu.tensor_desc<16xf32, #xegpu.scatter_tdesc_attr<>>, vector<16xi1> -> vector<16xf32>
+    gpu.func @test_load(%src: ui64) -> vector<32xf32> {
+      %cst = arith.constant dense<[
+      0,   8,  16,  24,  32,  40,  48,  56,
+      64,  72,  80,  88,  96, 104, 112, 120,
+      128, 136, 144, 152, 160, 168, 176, 184,
+      192, 200, 208, 216, 224, 232, 240, 248 
+      ]> : vector<32xindex>
+      
+      %c17 = arith.constant 17: index
+      %mask = vector.create_mask %c17: vector<32xi1>
+
+      %tdesc = xegpu.create_tdesc %src, %cst : ui64, vector<32xindex> -> !xegpu.tensor_desc<32xf32,  #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>>
+      %ld = xegpu.load %tdesc, %mask: !xegpu.tensor_desc<32xf32, #xegpu.scatter_tdesc_attr<>, #xegpu.layout<inst_data = [16]>>, vector<32xi1> -> vector<32xf32>
+      
+      gpu.return %ld : vector<32xf32> 
+    }
+
+//-----
+
+  // CHECK-LABEL: test_prefetch
+  // CHECK-SAME: [[arg0:%.+]]: ui64
+  // CHECK-COUNT-2: xegpu.create_tdesc [[arg0]], {{.*}} : ui64, vector<16xindex> -> !xegpu.tensor_desc<16xf32, #xegpu.scatter_tdesc_attr<>>
+  // CHECK-COUNT-2: xegpu.prefetch {{.*}} : !xegpu.tensor_desc<16xf32, #xegpu.scatter_tdesc_attr<>>
+  gpu.func @test_prefetch(%src: ui64)  {
+
+      %cst = arith.constant dense<[
----------------
Jianhui-Li wrote:

resolved. 

https://github.com/llvm/llvm-project/pull/143602