[PATCH] D111907: [AMDGPU] Divergence driven selection for fused bitlogic

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rG7cdb1df8c704: [AMDGPU] Divergence driven selection for fused bitlogic (authored by rampitec).

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D111907/new/

https://reviews.llvm.org/D111907

Files:
  llvm/lib/Target/AMDGPU/SOPInstructions.td
  llvm/lib/Target/AMDGPU/VOP3Instructions.td
  llvm/test/CodeGen/AMDGPU/fused-bitlogic.ll


Index: llvm/test/CodeGen/AMDGPU/fused-bitlogic.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/fused-bitlogic.ll
+++ llvm/test/CodeGen/AMDGPU/fused-bitlogic.ll
@@ -9,8 +9,7 @@
 ; GCN-NEXT:    s_waitcnt lgkmcnt(0)
 ; GCN-NEXT:    global_load_dwordx3 v[0:2], v3, s[0:1]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_or_b32_e32 v0, v1, v0
-; GCN-NEXT:    v_or_b32_e32 v0, v0, v2
+; GCN-NEXT:    v_or3_b32 v0, v1, v0, v2
 ; GCN-NEXT:    v_not_b32_e32 v0, v0
 ; GCN-NEXT:    global_store_dword v3, v0, s[0:1]
 ; GCN-NEXT:    s_endpgm
@@ -39,10 +38,8 @@
 ; GCN-NEXT:    global_load_dwordx2 v[4:5], v6, s[0:1] offset:16
 ; GCN-NEXT:    global_load_dwordx4 v[0:3], v6, s[0:1]
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
-; GCN-NEXT:    v_or_b32_e32 v1, v3, v1
-; GCN-NEXT:    v_or_b32_e32 v0, v2, v0
-; GCN-NEXT:    v_or_b32_e32 v0, v0, v4
-; GCN-NEXT:    v_or_b32_e32 v1, v1, v5
+; GCN-NEXT:    v_or3_b32 v1, v3, v1, v5
+; GCN-NEXT:    v_or3_b32 v0, v2, v0, v4
 ; GCN-NEXT:    v_not_b32_e32 v0, v0
 ; GCN-NEXT:    v_not_b32_e32 v1, v1
 ; GCN-NEXT:    global_store_dwordx2 v6, v[0:1], s[0:1]
@@ -103,8 +100,8 @@
 ; GCN-NEXT:    s_waitcnt vmcnt(0)
 ; GCN-NEXT:    v_and_b32_e32 v1, v3, v1
 ; GCN-NEXT:    v_and_b32_e32 v0, v2, v0
-; GCN-NEXT:    v_and_b32_e32 v0, v0, v4
 ; GCN-NEXT:    v_and_b32_e32 v1, v1, v5
+; GCN-NEXT:    v_and_b32_e32 v0, v0, v4
 ; GCN-NEXT:    v_not_b32_e32 v0, v0
 ; GCN-NEXT:    v_not_b32_e32 v1, v1
 ; GCN-NEXT:    global_store_dwordx2 v6, v[0:1], s[0:1]
Index: llvm/lib/Target/AMDGPU/VOP3Instructions.td
===================================================================
--- llvm/lib/Target/AMDGPU/VOP3Instructions.td
+++ llvm/lib/Target/AMDGPU/VOP3Instructions.td
@@ -667,6 +667,14 @@
 def : VOPBinOpClampPat<saddsat, V_ADD_I32_e64, i32>;
 def : VOPBinOpClampPat<ssubsat, V_SUB_I32_e64, i32>;
 
+def : GCNPat<(getDivergentFrag<or>.ret (or_oneuse i64:$src0, i64:$src1), i64:$src2),
+             (REG_SEQUENCE VReg_64,
+               (V_OR3_B32_e64 (i32 (EXTRACT_SUBREG $src0, sub0)),
+                              (i32 (EXTRACT_SUBREG $src1, sub0)),
+                              (i32 (EXTRACT_SUBREG $src2, sub0))), sub0,
+               (V_OR3_B32_e64 (i32 (EXTRACT_SUBREG $src0, sub1)),
+                              (i32 (EXTRACT_SUBREG $src1, sub1)),
+                              (i32 (EXTRACT_SUBREG $src2, sub1))), sub1)>;
 
 // FIXME: Probably should hardcode clamp bit in pseudo and avoid this.
 class OpSelBinOpClampPat<SDPatternOperator node,
Index: llvm/lib/Target/AMDGPU/SOPInstructions.td
===================================================================
--- llvm/lib/Target/AMDGPU/SOPInstructions.td
+++ llvm/lib/Target/AMDGPU/SOPInstructions.td
@@ -558,19 +558,19 @@
 >;
 
 def S_NAND_B32 : SOP2_32 <"s_nand_b32",
-  [(set i32:$sdst, (not (and_oneuse i32:$src0, i32:$src1)))]
+  [(set i32:$sdst, (UniformUnaryFrag<not> (and_oneuse i32:$src0, i32:$src1)))]
 >;
 
 def S_NAND_B64 : SOP2_64 <"s_nand_b64",
-  [(set i64:$sdst, (not (and_oneuse i64:$src0, i64:$src1)))]
+  [(set i64:$sdst, (UniformUnaryFrag<not> (and_oneuse i64:$src0, i64:$src1)))]
 >;
 
 def S_NOR_B32 : SOP2_32 <"s_nor_b32",
-  [(set i32:$sdst, (not (or_oneuse i32:$src0, i32:$src1)))]
+  [(set i32:$sdst, (UniformUnaryFrag<not> (or_oneuse i32:$src0, i32:$src1)))]
 >;
 
 def S_NOR_B64 : SOP2_64 <"s_nor_b64",
-  [(set i64:$sdst, (not (or_oneuse i64:$src0, i64:$src1)))]
+  [(set i64:$sdst, (UniformUnaryFrag<not> (or_oneuse i64:$src0, i64:$src1)))]
 >;
 } // End isCommutable = 1
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D111907.380308.patch
Type: text/x-patch
Size: 3544 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20211018/9d91e41a/attachment-0001.bin>