AMDGPU: Consolidate SMRD TableGen patterns

[tstellar]

Summary

Merge the SMRD patterns for CI into the same multiclass as the
patterns for other sub-targets.

This removes some duplicate code and will make it easier for some
future GlobalISel changes I would like to do.

[nhaehnle]

Nice. Patch LGTM except for one cosmetic issue. No need to update on Phabricator before commit as far as I'm concerned.

lib/Target/AMDGPU/SMInstructions.td

749–750 ↗

(On Diff #167151)

This patch removes the only place where SM_LOAD_PATTERN.AddedComplexity is used, so the comment should be removed, and the defm name should be removed as well to keep things consistent.

llvm/trunk/lib/Target/AMDGPU/SMInstructions.td

 } // let SubtargetPredicate = HasScalarAtomics
 
 let SubtargetPredicate = isGFX9 in {
 defm S_DCACHE_DISCARD    : SM_Pseudo_Discards <"s_dcache_discard">;
 defm S_DCACHE_DISCARD_X2 : SM_Pseudo_Discards <"s_dcache_discard_x2">;
 }
 
 //===----------------------------------------------------------------------===//
-// Scalar Memory Patterns
-//===----------------------------------------------------------------------===//
-
-
-def smrd_load : PatFrag <(ops node:$ptr), (load node:$ptr), [{ return isUniformLoad(N);}]>;
-
-def SMRDImm         : ComplexPattern<i64, 2, "SelectSMRDImm">;
-def SMRDImm32       : ComplexPattern<i64, 2, "SelectSMRDImm32">;
-def SMRDSgpr        : ComplexPattern<i64, 2, "SelectSMRDSgpr">;
-def SMRDBufferImm   : ComplexPattern<i32, 1, "SelectSMRDBufferImm">;
-def SMRDBufferImm32 : ComplexPattern<i32, 1, "SelectSMRDBufferImm32">;
-
-multiclass SMRD_Pattern <string Instr, ValueType vt> {
-
-  // 1. IMM offset
-  def : GCNPat <
-    (smrd_load (SMRDImm i64:$sbase, i32:$offset)),
-    (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, 0))
-  >;
-
-  // 2. SGPR offset
-  def : GCNPat <
-    (smrd_load (SMRDSgpr i64:$sbase, i32:$offset)),
-    (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, 0))
-  >;
-}
-
-multiclass SMLoad_Pattern <string Instr, ValueType vt> {
-  // 1. Offset as an immediate
-  // name this pattern to reuse AddedComplexity on CI
-  def _IMM : GCNPat <
-    (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm i32:$offset), i1:$glc),
-    (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, (as_i1imm $glc)))
-  >;
-
-  // 2. Offset loaded in an 32bit SGPR
-  def : GCNPat <
-    (SIsbuffer_load v4i32:$sbase, i32:$offset, i1:$glc),
-    (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, (as_i1imm $glc)))
-  >;
-}
-
-
-let OtherPredicates = [isSICI] in {
-def : GCNPat <
-  (i64 (readcyclecounter)),
-  (S_MEMTIME)
->;
-}
-
-// Global and constant loads can be selected to either MUBUF or SMRD
-// instructions, but SMRD instructions are faster so we want the instruction
-// selector to prefer those.
-let AddedComplexity = 100 in {
-
-defm : SMRD_Pattern <"S_LOAD_DWORD",    i32>;
-defm : SMRD_Pattern <"S_LOAD_DWORDX2",  v2i32>;
-defm : SMRD_Pattern <"S_LOAD_DWORDX4",  v4i32>;
-defm : SMRD_Pattern <"S_LOAD_DWORDX8",  v8i32>;
-defm : SMRD_Pattern <"S_LOAD_DWORDX16", v16i32>;
-
-// Name the pattern to reuse AddedComplexity on CI
-defm SM_LOAD_PATTERN : SMLoad_Pattern <"S_BUFFER_LOAD_DWORD",     i32>;
-defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX2",   v2i32>;
-defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX4",   v4i32>;
-defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX8",   v8i32>;
-defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX16",  v16i32>;
-} // End let AddedComplexity = 100
-
-let OtherPredicates = [isVI] in {
-
-def : GCNPat <
-  (i64 (readcyclecounter)),
-  (S_MEMREALTIME)
->;
-
-} // let OtherPredicates = [isVI]
-
-
-//===----------------------------------------------------------------------===//
 // Targets
 //===----------------------------------------------------------------------===//
 
 //===----------------------------------------------------------------------===//
 // SI
 //===----------------------------------------------------------------------===//
 
 class SMRD_Real_si <bits<5> op, SM_Pseudo ps>
class SMRD_Real_ci <bits<5> op, SM_Pseudo ps>
   let Inst{14-9}  = !if(ps.has_sbase, sbase{6-1}, ?);
   let Inst{21-15} = !if(ps.has_sdst, sdst{6-0}, ?);
   let Inst{26-22} = op;
   let Inst{31-27} = 0x18; //encoding
 }
 
 def S_DCACHE_INV_VOL_ci : SMRD_Real_ci <0x1d, S_DCACHE_INV_VOL>;
 
-let AddedComplexity = SM_LOAD_PATTERN_IMM.AddedComplexity in {
+//===----------------------------------------------------------------------===//
+// Scalar Memory Patterns
+//===----------------------------------------------------------------------===//
+
+def smrd_load : PatFrag <(ops node:$ptr), (load node:$ptr), [{ return isUniformLoad(N);}]>;
+
+def SMRDImm         : ComplexPattern<i64, 2, "SelectSMRDImm">;
+def SMRDImm32       : ComplexPattern<i64, 2, "SelectSMRDImm32">;
+def SMRDSgpr        : ComplexPattern<i64, 2, "SelectSMRDSgpr">;
+def SMRDBufferImm   : ComplexPattern<i32, 1, "SelectSMRDBufferImm">;
+def SMRDBufferImm32 : ComplexPattern<i32, 1, "SelectSMRDBufferImm32">;
+
+multiclass SMRD_Pattern <string Instr, ValueType vt> {
+
+  // 1. IMM offset
+  def : GCNPat <
+    (smrd_load (SMRDImm i64:$sbase, i32:$offset)),
+    (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, 0))
+  >;
 
-class SMRD_Pattern_ci <string Instr, ValueType vt> : GCNPat <
+  // 2. 32-bit IMM offset on CI
+  def : GCNPat <
     (smrd_load (SMRDImm32 i64:$sbase, i32:$offset)),
     (vt (!cast<InstSI>(Instr#"_IMM_ci") $sbase, $offset, 0))> {
     let OtherPredicates = [isCIOnly];
   }
 
-def : SMRD_Pattern_ci <"S_LOAD_DWORD",    i32>;
-def : SMRD_Pattern_ci <"S_LOAD_DWORDX2",  v2i32>;
-def : SMRD_Pattern_ci <"S_LOAD_DWORDX4",  v4i32>;
-def : SMRD_Pattern_ci <"S_LOAD_DWORDX8",  v8i32>;
-def : SMRD_Pattern_ci <"S_LOAD_DWORDX16", v16i32>;
+  // 3. SGPR offset
+  def : GCNPat <
+    (smrd_load (SMRDSgpr i64:$sbase, i32:$offset)),
+    (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, 0))
+  >;
+}
+
+multiclass SMLoad_Pattern <string Instr, ValueType vt> {
+  // 1. Offset as an immediate
+  def : GCNPat <
+    (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm i32:$offset), i1:$glc),
+    (vt (!cast<SM_Pseudo>(Instr#"_IMM") $sbase, $offset, (as_i1imm $glc)))
+  >;
 
-class SMLoad_Pattern_ci <string Instr, ValueType vt> : GCNPat <
+  // 2. 32-bit IMM offset on CI
+  def : GCNPat <
     (vt (SIsbuffer_load v4i32:$sbase, (SMRDBufferImm32 i32:$offset), i1:$glc)),
-  (!cast<InstSI>(Instr) $sbase, $offset, (as_i1imm $glc))> {
+    (!cast<InstSI>(Instr#"_IMM_ci") $sbase, $offset, (as_i1imm $glc))> {
     let OtherPredicates = [isCIOnly];
   }
 
-def : SMLoad_Pattern_ci <"S_BUFFER_LOAD_DWORD_IMM_ci", i32>;
-def : SMLoad_Pattern_ci <"S_BUFFER_LOAD_DWORDX2_IMM_ci", v2i32>;
-def : SMLoad_Pattern_ci <"S_BUFFER_LOAD_DWORDX4_IMM_ci", v4i32>;
-def : SMLoad_Pattern_ci <"S_BUFFER_LOAD_DWORDX8_IMM_ci", v8i32>;
-def : SMLoad_Pattern_ci <"S_BUFFER_LOAD_DWORDX16_IMM_ci", v16i32>;
+  // 3. Offset loaded in an 32bit SGPR
+  def : GCNPat <
+    (SIsbuffer_load v4i32:$sbase, i32:$offset, i1:$glc),
+    (vt (!cast<SM_Pseudo>(Instr#"_SGPR") $sbase, $offset, (as_i1imm $glc)))
+  >;
+}
+
+// Global and constant loads can be selected to either MUBUF or SMRD
+// instructions, but SMRD instructions are faster so we want the instruction
+// selector to prefer those.
+let AddedComplexity = 100 in {
 
-} // End let AddedComplexity = SM_LOAD_PATTERN.AddedComplexity
+defm : SMRD_Pattern <"S_LOAD_DWORD",    i32>;
+defm : SMRD_Pattern <"S_LOAD_DWORDX2",  v2i32>;
+defm : SMRD_Pattern <"S_LOAD_DWORDX4",  v4i32>;
+defm : SMRD_Pattern <"S_LOAD_DWORDX8",  v8i32>;
+defm : SMRD_Pattern <"S_LOAD_DWORDX16", v16i32>;
 
+defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORD",     i32>;
+defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX2",   v2i32>;
+defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX4",   v4i32>;
+defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX8",   v8i32>;
+defm : SMLoad_Pattern <"S_BUFFER_LOAD_DWORDX16",  v16i32>;
+} // End let AddedComplexity = 100
+
+let OtherPredicates = [isSICI] in {
+def : GCNPat <
+  (i64 (readcyclecounter)),
+  (S_MEMTIME)
+>;
+}
+
+let OtherPredicates = [isVI] in {
+
+def : GCNPat <
+  (i64 (readcyclecounter)),
+  (S_MEMREALTIME)
+>;
+
+} // let OtherPredicates = [isVI]