[llvm-branch-commits] [llvm] [AMDGPU] Expose buffer resource num_records width in TargetParser (PR #220103)

[Matt Arsenault via llvm-branch-commits]

================
@@ -3209,6 +3209,44 @@ TEST(TargetParserTest, testAMDGPUgetMaxHWAddressableLocalMemorySize) {
             327680u);
 }
 
+TEST(TargetParserTest, testAMDGPUgetBufferResourceNumRecordsWidth) {
+  EXPECT_EQ(AMDGPU::getBufferResourceNumRecordsWidth(AMDGPU::GK_GFX900), 32u);
+  EXPECT_EQ(AMDGPU::getBufferResourceNumRecordsWidth(AMDGPU::GK_GFX1201), 32u);
+  EXPECT_EQ(AMDGPU::getBufferResourceNumRecordsWidth(AMDGPU::GK_GFX1250), 45u);
+
+  // Generic families that agree on resource width behave.
+  EXPECT_EQ(AMDGPU::getBufferResourceNumRecordsWidth(AMDGPU::GK_GFX9_GENERIC),
+            32u);
+
+  EXPECT_EQ(AMDGPU::getBufferResourceNumRecordsWidth(AMDGPU::GK_GENERIC),
+            std::nullopt);
+  EXPECT_EQ(AMDGPU::getBufferResourceNumRecordsWidth(AMDGPU::GK_GENERIC_HSA),
+            std::nullopt);
----------------
arsenm wrote:

Can we equivalently just use 0 instead of optional 

https://github.com/llvm/llvm-project/pull/220103