[PATCH] D31482: [AMDGPU][GFX9] Set +fp32-denormals for >=gfx900 unless -cl-denorms-are-zero is set

Konstantin Zhuravlyov via Phabricator via cfe-commits cfe-commits at lists.llvm.org
Wed Mar 29 16:06:09 PDT 2017


kzhuravl added inline comments.


================
Comment at: llvm/tools/clang/lib/Basic/Targets.cpp:2114-2116
+  static bool hasFullSpeedFP32Denorms(StringRef GPUName) {
+    return parseAMDGCNName(GPUName) >= GK_GFX9;
+  }
----------------
arsenm wrote:
> This is misleading since it was true on VI as well. I think just FMA rate changed
Yes, GFX8 supports f32 denorms at full speed too.  However, it doesn't have a full speed fma, so we didh't enable it then since it caused too many mad-heavy apps to slow down.


https://reviews.llvm.org/D31482





More information about the cfe-commits mailing list