commit e64433b3cc5539a934a650d8f2c346099770601d
author Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> Tue Apr 13 12:48:53 2021 -0700
committer Copybara-Service <copybara-worker@google.com> Fri Apr 16 13:02:04 2021 -0700
tree 8b579a924769b91a4e91862c557740f4963122af
parent f4c1865061a016a48aeb5f62e68561bde3a43b97

[AMDGPU] Factor out predicate FmaakFmamkF32Insts

Differential Revision: https://reviews.llvm.org/D100409

GitOrigin-RevId: 13015ebd6f568f98e9b908564195f03f8f0d02a9

lib/Target/AMDGPU/AMDGPU.td

diff --git a/lib/Target/AMDGPU/AMDGPU.td b/lib/Target/AMDGPU/AMDGPU.td
index 11c1b22..d08efec 100644
--- a/lib/Target/AMDGPU/AMDGPU.td
+++ b/lib/Target/AMDGPU/AMDGPU.td
@@ -1321,6 +1321,10 @@
 def HasPackedFP32Ops : Predicate<"Subtarget->hasPackedFP32Ops()">,
   AssemblerPredicate<(all_of FeaturePackedFP32Ops)>;
 
+def HasFmaakFmamkF32Insts :
+  Predicate<"Subtarget->hasFmaakFmamkF32Insts()">,
+  AssemblerPredicate<(any_of FeatureGFX10Insts)>;
+
 def HasExtendedImageInsts : Predicate<"Subtarget->hasExtendedImageInsts()">,
   AssemblerPredicate<(all_of FeatureExtendedImageInsts)>;
 

lib/Target/AMDGPU/GCNSubtarget.h

diff --git a/lib/Target/AMDGPU/GCNSubtarget.h b/lib/Target/AMDGPU/GCNSubtarget.h
index 415b1cb..00ea653 100644
--- a/lib/Target/AMDGPU/GCNSubtarget.h
+++ b/lib/Target/AMDGPU/GCNSubtarget.h
@@ -838,6 +838,10 @@
     return HasPackedFP32Ops;
   }
 
+  bool hasFmaakFmamkF32Insts() const {
+    return getGeneration() >= GFX10;
+  }
+
   bool hasExtendedImageInsts() const {
     return HasExtendedImageInsts;
   }

lib/Target/AMDGPU/VOP2Instructions.td

diff --git a/lib/Target/AMDGPU/VOP2Instructions.td b/lib/Target/AMDGPU/VOP2Instructions.td
index e24c2c2..76da6dc 100644
--- a/lib/Target/AMDGPU/VOP2Instructions.td
+++ b/lib/Target/AMDGPU/VOP2Instructions.td
@@ -749,17 +749,21 @@
   }
 } // End AddedComplexity = 30
 
+let SubtargetPredicate = HasFmaakFmamkF32Insts in {
+def V_FMAMK_F32 : VOP2_Pseudo<"v_fmamk_f32", VOP_MADMK_F32, [], "">;
+
+let isCommutable = 1 in
+def V_FMAAK_F32 : VOP2_Pseudo<"v_fmaak_f32", VOP_MADAK_F32, [], "">;
+}
+
 let SubtargetPredicate = isGFX10Plus in {
 
-def V_FMAMK_F32 : VOP2_Pseudo<"v_fmamk_f32", VOP_MADMK_F32, [], "">;
-let FPDPRounding = 1 in
+let FPDPRounding = 1 in {
 def V_FMAMK_F16 : VOP2_Pseudo <"v_fmamk_f16", VOP_MADMK_F16, [], "">;
 
-let isCommutable = 1 in {
-def V_FMAAK_F32 : VOP2_Pseudo<"v_fmaak_f32", VOP_MADAK_F32, [], "">;
-let FPDPRounding = 1 in
+let isCommutable = 1 in
 def V_FMAAK_F16 : VOP2_Pseudo <"v_fmaak_f16", VOP_MADAK_F16, [], "">;
-} // End isCommutable = 1
+} // End FPDPRounding  = 1
 
 let Constraints = "$vdst = $src2",
     DisableEncoding="$src2",