[PATCH] D36856: [AMDGPU] Use v_max_f* for fcanonicalize

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added inline comments.


================
Comment at: lib/Target/AMDGPU/AMDGPUInstructions.td:45-47
+def FP16Denormals : Predicate<"Subtarget->hasFP16Denormals()">;
+def FP32Denormals : Predicate<"Subtarget->hasFP32Denormals()">;
+def FP64Denormals : Predicate<"Subtarget->hasFP64Denormals()">;
----------------
arsenm wrote:
> How / why this change?
These predicates were previously unused, thus error went undetected.


================
Comment at: test/CodeGen/AMDGPU/fcanonicalize-denorms.ll:8
+declare i32 @llvm.amdgcn.workitem.id.x() #0
+
+; GCN-LABEL:  {{^}}test_canonicalize_value_f64:
----------------
arsenm wrote:
> Can you merge this with fcanonicalize.ll? That one avoids multiple run lines by using the attributes on the different functions
fcnonicalize.ll defaults to SI, to it cannot lower f16 tests. I can change it to tonga.


https://reviews.llvm.org/D36856