[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

[Farzon Lotfi via llvm-commits]

================
@@ -0,0 +1,152 @@
+// RUN: %clang_cc1 -finclude-default-header -fnative-half-type -triple \
+// RUN:   dxil-pc-shadermodel6.3-compute %s -emit-llvm -o - | \
+// RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-DXIL
+// RUN: %clang_cc1 -finclude-default-header -fnative-half-type -triple \
+// RUN:   spirv-pc-vulkan-compute %s -emit-llvm -o - | \
+// RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-SPIRV
+
+// Test basic lowering to runtime function call.
+
+// CHECK-LABEL: define {{.*}}test_default_parameter_type
+float test_default_parameter_type(half2 p1, half2 p2, float p3) {
+  // CHECK-SPIRV:  %[[MUL:.*]] = call reassoc nnan ninf nsz arcp afn half @llvm.spv.fdot.v2f16(<2 x half> %7, <2 x half> %8)
+  // CHECK-SPIRV:  %[[CONV:.*]] = fpext reassoc nnan ninf nsz arcp afn half %[[MUL]] to float
+  // CHECK-SPIRV:  %[[C:.*]] = load float, ptr %c.addr.i, align 4
+  // CHECK-SPIRV:  %[[RES:.*]] = fadd reassoc nnan ninf nsz arcp afn float %[[CONV]], %[[C]]
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f16(<2 x half> %6, <2 x half> %7, float %8)
----------------
farzonl wrote:

We don't typically lile to capture the register arguments because doing so can make the tests brittle to future changes and force updates if a register name changes instead do what sarah is doing here:
https://github.com/llvm/llvm-project/pull/133941/files `{{.*}}`.
```suggestion
  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f16(<2 x half> %{{.*}}, <2 x half> %{{.*}}, float %{{.*}})
```

https://github.com/llvm/llvm-project/pull/131237