[clang] [Matrix] Copy Row data from padded cbuffer offsets before swizzle (PR #185346)

[Justin Bogner via cfe-commits]

================
@@ -0,0 +1,19 @@
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.7-library -disable-llvm-passes -emit-llvm -finclude-default-header -fmatrix-memory-layout=row-major -o - %s | FileCheck %s
+
+cbuffer CB {
+  float3x2 Mat;
+}
+
+// CHECK: @Mat = external hidden addrspace(2) global <{ [2 x <{ <2 x float>, target("dx.Padding", 8) }>], <2 x float> }>, align 4
+
+// CHECK-LABEL: define hidden noundef nofpclass(nan inf) <2 x float> @_Z15get_row_swizzlev()
+// CHECK: %matrix.buf.copy = alloca [3 x <2 x float>], align 8
+// CHECK: %cbuf.load = load <2 x float>, ptr addrspace(2) @Mat, align 8
+// CHECK: %cbuf.load2 = load <2 x float>, ptr addrspace(2) getelementptr inbounds nuw (i8, ptr addrspace(2) @Mat, i32 16), align 8
+// CHECK: %cbuf.load4 = load <2 x float>, ptr addrspace(2) getelementptr inbounds nuw (i8, ptr addrspace(2) @Mat, i32 32), align 8
+// CHECK: %0 = load <6 x float>, ptr %matrix.buf.copy, align 8
+// CHECK: %1 = shufflevector <6 x float> %0, <6 x float> poison, <2 x i32> <i32 4, i32 1>
----------------
bogner wrote:

Better to use regex for the value names (both the numbered and the named ones). Hardcoding them like this makes the test susceptible to needing spurious changes.

https://github.com/llvm/llvm-project/pull/185346