[PATCH] D124757: [X86] Replace avx512f integer add reduction builtins with generic builtin

[Simon Pilgrim via Phabricator via cfe-commits]

RKSimon created this revision.
RKSimon added reviewers: pengfei, craig.topper.
Herald added a subscriber: StephenFan.
Herald added a project: All.
RKSimon requested review of this revision.
Herald added a project: clang.

D124741 <https://reviews.llvm.org/D124741> added the generic "__builtin_reduce_add" which we can use to replace the x86 specific integer add reduction builtins - internally these were mapping to the same intrinsic already so there are no test changes required.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D124757

Files:
  clang/include/clang/Basic/BuiltinsX86.def
  clang/lib/CodeGen/CGBuiltin.cpp
  clang/lib/Headers/avx512fintrin.h


Index: clang/lib/Headers/avx512fintrin.h
===================================================================
--- clang/lib/Headers/avx512fintrin.h
+++ clang/lib/Headers/avx512fintrin.h
@@ -9319,7 +9319,7 @@
  */
 
 static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_add_epi64(__m512i __W) {
-  return __builtin_ia32_reduce_add_q512(__W);
+  return __builtin_reduce_add((__v8di)__W);
 }
 
 static __inline__ long long __DEFAULT_FN_ATTRS512 _mm512_reduce_mul_epi64(__m512i __W) {
@@ -9337,7 +9337,7 @@
 static __inline__ long long __DEFAULT_FN_ATTRS512
 _mm512_mask_reduce_add_epi64(__mmask8 __M, __m512i __W) {
   __W = _mm512_maskz_mov_epi64(__M, __W);
-  return __builtin_ia32_reduce_add_q512(__W);
+  return __builtin_reduce_add((__v8di)__W);
 }
 
 static __inline__ long long __DEFAULT_FN_ATTRS512
@@ -9383,7 +9383,7 @@
 
 static __inline__ int __DEFAULT_FN_ATTRS512
 _mm512_reduce_add_epi32(__m512i __W) {
-  return __builtin_ia32_reduce_add_d512((__v16si)__W);
+  return __builtin_reduce_add((__v16si)__W);
 }
 
 static __inline__ int __DEFAULT_FN_ATTRS512
@@ -9404,7 +9404,7 @@
 static __inline__ int __DEFAULT_FN_ATTRS512
 _mm512_mask_reduce_add_epi32( __mmask16 __M, __m512i __W) {
   __W = _mm512_maskz_mov_epi32(__M, __W);
-  return __builtin_ia32_reduce_add_d512((__v16si)__W);
+  return __builtin_reduce_add((__v16si)__W);
 }
 
 static __inline__ int __DEFAULT_FN_ATTRS512
Index: clang/lib/CodeGen/CGBuiltin.cpp
===================================================================
--- clang/lib/CodeGen/CGBuiltin.cpp
+++ clang/lib/CodeGen/CGBuiltin.cpp
@@ -14496,12 +14496,6 @@
     return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true);
 
   // Reductions
-  case X86::BI__builtin_ia32_reduce_add_d512:
-  case X86::BI__builtin_ia32_reduce_add_q512: {
-    Function *F =
-        CGM.getIntrinsic(Intrinsic::vector_reduce_add, Ops[0]->getType());
-    return Builder.CreateCall(F, {Ops[0]});
-  }
   case X86::BI__builtin_ia32_reduce_fadd_pd512:
   case X86::BI__builtin_ia32_reduce_fadd_ps512:
   case X86::BI__builtin_ia32_reduce_fadd_ph512:
Index: clang/include/clang/Basic/BuiltinsX86.def
===================================================================
--- clang/include/clang/Basic/BuiltinsX86.def
+++ clang/include/clang/Basic/BuiltinsX86.def
@@ -1990,8 +1990,6 @@
 TARGET_BUILTIN(__builtin_ia32_selectsd_128, "V2dUcV2dV2d", "ncV:128:", "avx512f")
 
 // generic reduction intrinsics
-TARGET_BUILTIN(__builtin_ia32_reduce_add_d512, "iV16i", "ncV:512:", "avx512f")
-TARGET_BUILTIN(__builtin_ia32_reduce_add_q512, "OiV8Oi", "ncV:512:", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_reduce_fadd_pd512, "ddV8d", "ncV:512:", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_reduce_fadd_ps512, "ffV16f", "ncV:512:", "avx512f")
 TARGET_BUILTIN(__builtin_ia32_reduce_fadd_ph512, "xxV32x", "ncV:512:", "avx512fp16")


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D124757.426371.patch
Type: text/x-patch
Size: 2839 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20220502/4b2961a2/attachment.bin>