[llvm] r354346 - [X86][AVX] Update VBROADCAST folds to always use v2i64 X86vzload

[Simon Pilgrim via llvm-commits]

Author: rksimon
Date: Tue Feb 19 08:33:17 2019
New Revision: 354346

URL: http://llvm.org/viewvc/llvm-project?rev=354346&view=rev
Log:
[X86][AVX] Update VBROADCAST folds to always use v2i64 X86vzload

The VBROADCAST combines and SimplifyDemandedVectorElts improvements mean that we now more consistently use shorter (128-bit) X86vzload input operands.

Follow up to D58053

Modified:
    llvm/trunk/lib/Target/X86/X86InstrAVX512.td
    llvm/trunk/lib/Target/X86/X86InstrSSE.td
    llvm/trunk/test/CodeGen/X86/insertelement-shuffle.ll
    llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
    llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll

Modified: llvm/trunk/lib/Target/X86/X86InstrAVX512.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrAVX512.td?rev=354346&r1=354345&r2=354346&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrAVX512.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrAVX512.td Tue Feb 19 08:33:17 2019
@@ -1378,7 +1378,7 @@ multiclass avx512_subvec_broadcast_rm_dq
 
 let Predicates = [HasAVX512] in {
   // 32-bit targets will fail to load a i64 directly but can use ZEXT_LOAD.
-  def : Pat<(v8i64 (X86VBroadcast (v8i64 (X86vzload addr:$src)))),
+  def : Pat<(v8i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQZm addr:$src)>;
 }
 
@@ -1386,7 +1386,7 @@ let Predicates = [HasVLX] in {
   // 32-bit targets will fail to load a i64 directly but can use ZEXT_LOAD.
   def : Pat<(v2i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQZ128m addr:$src)>;
-  def : Pat<(v4i64 (X86VBroadcast (v4i64 (X86vzload addr:$src)))),
+  def : Pat<(v4i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQZ256m addr:$src)>;
 }
 let Predicates = [HasVLX, HasBWI] in {

Modified: llvm/trunk/lib/Target/X86/X86InstrSSE.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrSSE.td?rev=354346&r1=354345&r2=354346&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrSSE.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrSSE.td Tue Feb 19 08:33:17 2019
@@ -7850,7 +7850,7 @@ let Predicates = [HasAVX2, NoVLX] in {
   // 32-bit targets will fail to load a i64 directly but can use ZEXT_LOAD.
   def : Pat<(v2i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQrm addr:$src)>;
-  def : Pat<(v4i64 (X86VBroadcast (v4i64 (X86vzload addr:$src)))),
+  def : Pat<(v4i64 (X86VBroadcast (v2i64 (X86vzload addr:$src)))),
             (VPBROADCASTQYrm addr:$src)>;
 
   def : Pat<(v4i32 (X86VBroadcast (v4i32 (scalar_to_vector (loadi32 addr:$src))))),

Modified: llvm/trunk/test/CodeGen/X86/insertelement-shuffle.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/insertelement-shuffle.ll?rev=354346&r1=354345&r2=354346&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/insertelement-shuffle.ll (original)
+++ llvm/trunk/test/CodeGen/X86/insertelement-shuffle.ll Tue Feb 19 08:33:17 2019
@@ -95,8 +95,7 @@ define <8 x i64> @insert_subvector_into_
 ;
 ; X86_AVX512-LABEL: insert_subvector_into_undef:
 ; X86_AVX512:       # %bb.0:
-; X86_AVX512-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X86_AVX512-NEXT:    vbroadcastsd %xmm0, %zmm0
+; X86_AVX512-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %zmm0
 ; X86_AVX512-NEXT:    retl
 ;
 ; X64_AVX512-LABEL: insert_subvector_into_undef:

Modified: llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx2.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx2.ll?rev=354346&r1=354345&r2=354346&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx2.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx2.ll Tue Feb 19 08:33:17 2019
@@ -655,8 +655,7 @@ define <16 x i8> @combine_pshufb_inserti
 define <8 x i32> @combine_permd_insertion_as_broadcast_v4i64(i64 %a0) {
 ; X86-LABEL: combine_permd_insertion_as_broadcast_v4i64:
 ; X86:       # %bb.0:
-; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X86-NEXT:    vbroadcastsd %xmm0, %ymm0
+; X86-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %ymm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_permd_insertion_as_broadcast_v4i64:

Modified: llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll?rev=354346&r1=354345&r2=354346&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll Tue Feb 19 08:33:17 2019
@@ -975,8 +975,7 @@ define <16 x float> @combine_vpermi2var_
 define <8 x i64> @combine_vpermvar_insertion_as_broadcast_v8i64(i64 %a0) {
 ; X86-LABEL: combine_vpermvar_insertion_as_broadcast_v8i64:
 ; X86:       # %bb.0:
-; X86-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
-; X86-NEXT:    vbroadcastsd %xmm0, %zmm0
+; X86-NEXT:    vbroadcastsd {{[0-9]+}}(%esp), %zmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: combine_vpermvar_insertion_as_broadcast_v8i64: