[PowerPC] Enable use of lxvw4x/stxvw4x in VSX code generation

Currently the VSX support enables use of lxvd2x and stxvd2x for 2x64
types, but does not yet use lxvw4x and stxvw4x for 4x32 types.  This
patch adds that support.

As with lxvd2x/stxvd2x, this involves straightforward overriding of
the patterns normally recognized for lvx/stvx, with preference given
to the VSX patterns when VSX is enabled.

In addition, the logic for permitting misaligned memory accesses is
modified so that v4r32 and v4i32 are treated the same as v2f64 and
v2i64 when VSX is enabled.  Finally, the DAG generation for unaligned
loads is changed to just use a normal LOAD (which will become lxvw4x)
on P8 and later hardware, where unaligned loads are preferred over
lvsl/lvx/lvx/vperm.

A number of tests now generate the VSX loads/stores instead of
lvx/stvx, so this patch adds VSX variants to those tests.  I've also
added <4 x float> tests to the vsx.ll test case, and created a
vsx-p8.ll test case to be used for testing code generation for the
P8Vector feature.  For now, that simply tests the unaligned load/store
behavior.

This has been tested along with a temporary patch to enable the VSX
and P8Vector features, with no new regressions encountered with or
without the temporary patch applied.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@220047 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index b29674d992bb5e05e41e281e96315ab660ac0f6e..2808b41671d11f84af94b85142b740a147d77daa 100644 (file)
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -8312,6 +8312,8 @@ SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
     unsigned ABIAlignment = getDataLayout()->getABITypeAlignment(Ty);
     if (ISD::isNON_EXTLoad(N) && VT.isVector() &&
         TM.getSubtarget<PPCSubtarget>().hasAltivec() &&
+        // P8 and later hardware should just use LOAD.
+        !TM.getSubtarget<PPCSubtarget>().hasP8Vector() &&
         (VT == MVT::v16i8 || VT == MVT::v8i16 ||
          VT == MVT::v4i32 || VT == MVT::v4f32) &&
         LD->getAlignment() < ABIAlignment) {
@@ -9204,7 +9206,8 @@ bool PPCTargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
 
   if (VT.getSimpleVT().isVector()) {
     if (Subtarget.hasVSX()) {
-      if (VT != MVT::v2f64 && VT != MVT::v2i64)
+      if (VT != MVT::v2f64 && VT != MVT::v2i64 &&
+          VT != MVT::v4f32 && VT != MVT::v4i32)
         return false;
     } else {
       return false;

diff --git a/lib/Target/PowerPC/PPCInstrVSX.td b/lib/Target/PowerPC/PPCInstrVSX.td
index f30e64f95d1757cb1728d6fa40b6ee5cf0d7abce..4b2f5a321f803d048de37db453e890b284fe93e2 100644 (file)
--- a/lib/Target/PowerPC/PPCInstrVSX.td
+++ b/lib/Target/PowerPC/PPCInstrVSX.td
@@ -63,7 +63,8 @@ let Uses = [RM] in {
 
     def LXVW4X : XX1Form<31, 780,
                          (outs vsrc:$XT), (ins memrr:$src),
-                         "lxvw4x $XT, $src", IIC_LdStLFD, []>;
+                         "lxvw4x $XT, $src", IIC_LdStLFD,
+                         [(set v4i32:$XT, (load xoaddr:$src))]>;
   }
 
   // Store indexed instructions
@@ -80,7 +81,8 @@ let Uses = [RM] in {
 
     def STXVW4X : XX1Form<31, 908,
                          (outs), (ins vsrc:$XT, memrr:$dst),
-                         "stxvw4x $XT, $dst", IIC_LdStSTFD, []>;
+                         "stxvw4x $XT, $dst", IIC_LdStSTFD,
+                         [(store v4i32:$XT, xoaddr:$dst)]>;
   }
 
   // Add/Mul Instructions
@@ -811,6 +813,13 @@ def : Pat<(sext_inreg v2i64:$C, v2i32),
 def : Pat<(v2f64 (sint_to_fp (sext_inreg v2i64:$C, v2i32))),
           (XVCVSXWDP (XXSLDWI $C, $C, 1))>;
 
+// Loads.
+def : Pat<(v4i32 (load xoaddr:$src)), (LXVW4X xoaddr:$src)>;
+
+// Stores.
+def : Pat<(store v4i32:$rS, xoaddr:$dst),
+          (STXVW4X $rS, xoaddr:$dst)>;
+
 } // AddedComplexity
 } // HasVSX
 

diff --git a/test/CodeGen/PowerPC/unal4-std.ll b/test/CodeGen/PowerPC/unal4-std.ll
index 9f29e31cb9024deec16b18b248bb8763df6ea6dc..e911099111618b7e1394e3aa482ad12532269a5d 100644 (file)
--- a/test/CodeGen/PowerPC/unal4-std.ll
+++ b/test/CodeGen/PowerPC/unal4-std.ll
@@ -1,4 +1,5 @@
-; RUN: llc < %s -mcpu=pwr7 | FileCheck %s
+; RUN: llc < %s -mcpu=pwr7 -mattr=-vsx| FileCheck %s
+; RUN: llc < %s -mcpu=pwr7 -mattr=+vsx | FileCheck -check-prefix=CHECK-VSX %s
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
 target triple = "powerpc64-unknown-linux-gnu"
 
@@ -22,6 +23,9 @@ if.end210:                                        ; preds = %entry
 ; a multiple of 4).
 ; CHECK: @copy_to_conceal
 ; CHECK: stdx {{[0-9]+}}, 0,
+
+; CHECK-VSX: @copy_to_conceal
+; CHECK-VSX: stxvw4x {{[0-9]+}}, 0,
 }
 
 attributes #0 = { nounwind "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }

diff --git a/test/CodeGen/PowerPC/unaligned.ll b/test/CodeGen/PowerPC/unaligned.ll
index 0c59516f1186728d3bba44776bf63895030b7967..64c03cdda35e70f240c5a3f7795ab19557e0296f 100644 (file)
--- a/test/CodeGen/PowerPC/unaligned.ll
+++ b/test/CodeGen/PowerPC/unaligned.ll
@@ -92,10 +92,14 @@ entry:
 ; CHECK-DAG: stdx
 ; CHECK: stdx
 
+; For VSX on P7, unaligned loads and stores are preferable to aligned
+; stack slots, but lvsl/vperm is better still.  (On P8 lxvw4x is preferable.)
+; Using unaligned stxvw4x is preferable on both machines.
 ; CHECK-VSX: @foo6
-; CHECK-VSX-DAG: ld
-; CHECK-VSX-DAG: ld
-; CHECK-VSX-DAG: stdx
-; CHECK-VSX: stdx
+; CHECK-VSX-DAG: lvsl
+; CHECK-VSX-DAG: lvx
+; CHECK-VSX-DAG: lvx
+; CHECK-VSX: vperm
+; CHECK-VSX: stxvw4x
 }
 

diff --git a/test/CodeGen/PowerPC/vec-abi-align.ll b/test/CodeGen/PowerPC/vec-abi-align.ll
index 3239cf6c06ab649bf7c42c1c1277302795c2fdd9..5075ff2b8c07964c319f798edd34d5b4ab3ffb71 100644 (file)
--- a/test/CodeGen/PowerPC/vec-abi-align.ll
+++ b/test/CodeGen/PowerPC/vec-abi-align.ll
@@ -1,4 +1,5 @@
-; RUN: llc -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 < %s | FileCheck %s
+; RUN: llc -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 -mattr=-vsx < %s | FileCheck %s
+; RUN: llc -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 -mattr=+vsx < %s | FileCheck -check-prefix=CHECK-VSX %s
 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
 target triple = "powerpc64-unknown-linux-gnu"
 
@@ -16,6 +17,10 @@ entry:
 ; CHECK-LABEL: @test1
 ; CHECK: stvx 2,
 ; CHECK: blr
+
+; CHECK-VSX-LABEL: @test1
+; CHECK-VSX: stxvw4x 34,
+; CHECK-VSX: blr
 }
 
 ; Function Attrs: nounwind
@@ -35,6 +40,13 @@ entry:
 ; CHECK: addi [[REGB:[0-9]+]], 1, 112
 ; CHECK: lvx 2, [[REGB]], [[REG16]]
 ; CHECK: blr
+
+; CHECK-VSX-LABEL: @test2
+; CHECK-VSX: ld {{[0-9]+}}, 112(1)
+; CHECK-VSX: li [[REG16:[0-9]+]], 16
+; CHECK-VSX: addi [[REGB:[0-9]+]], 1, 112
+; CHECK-VSX: lxvw4x {{[0-9]+}}, [[REGB]], [[REG16]]
+; CHECK-VSX: blr
 }
 
 ; Function Attrs: nounwind
@@ -54,6 +66,13 @@ entry:
 ; CHECK: addi [[REGB:[0-9]+]], 1, 128
 ; CHECK: lvx 2, [[REGB]], [[REG16]]
 ; CHECK: blr
+
+; CHECK-VSX-LABEL: @test3
+; CHECK-VSX: ld {{[0-9]+}}, 128(1)
+; CHECK-VSX: li [[REG16:[0-9]+]], 16
+; CHECK-VSX: addi [[REGB:[0-9]+]], 1, 128
+; CHECK-VSX: lxvw4x {{[0-9]+}}, [[REGB]], [[REG16]]
+; CHECK-VSX: blr
 }
 
 attributes #0 = { nounwind }

diff --git a/test/CodeGen/PowerPC/vec_misaligned.ll b/test/CodeGen/PowerPC/vec_misaligned.ll
index 304a84d49a9da82dc3cddace2b7172ba05cc5558..73a4a4d395daf4227871ae6af2d099af0ecdfa18 100644 (file)
--- a/test/CodeGen/PowerPC/vec_misaligned.ll
+++ b/test/CodeGen/PowerPC/vec_misaligned.ll
@@ -1,5 +1,5 @@
 ; RUN: llc < %s -march=ppc32 -mcpu=g5 | FileCheck %s
-; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mattr=+altivec | FileCheck %s
+; RUN: llc < %s -mtriple=powerpc64-unknown-linux-gnu -mattr=+altivec -mattr=-vsx -mattr=-power8-vector | FileCheck %s
 ; RUN: llc < %s -mtriple=powerpc64le-unknown-linux-gnu -mattr=+altivec | FileCheck %s -check-prefix=CHECK-LE
 
 target datalayout = "E-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f128:64:128"

diff --git a/test/CodeGen/PowerPC/vrspill.ll b/test/CodeGen/PowerPC/vrspill.ll
index c3d1bf8f1ead2ad201dac41b6f0df2cc430110d3..b990442aed87545dbc59596188a41c4dc436205c 100644 (file)
--- a/test/CodeGen/PowerPC/vrspill.ll
+++ b/test/CodeGen/PowerPC/vrspill.ll
@@ -1,5 +1,7 @@
-; RUN: llc -O0 -mtriple=powerpc-unknown-linux-gnu -mattr=+altivec -verify-machineinstrs < %s | FileCheck %s
-; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -mattr=+altivec -verify-machineinstrs -fast-isel=false < %s | FileCheck %s
+; RUN: llc -O0 -mtriple=powerpc-unknown-linux-gnu -mattr=+altivec -mattr=-vsx -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -mattr=+altivec -mattr=-vsx -verify-machineinstrs -fast-isel=false < %s | FileCheck %s
+; RUN: llc -O0 -mtriple=powerpc-unknown-linux-gnu -mattr=+altivec -mattr=+vsx -verify-machineinstrs < %s | FileCheck -check-prefix=CHECK-VSX %s
+; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -mattr=+altivec -mattr=+vsx -verify-machineinstrs -fast-isel=false < %s | FileCheck -check-prefix=CHECK-VSX %s
 
 ; This verifies that we generate correct spill/reload code for vector regs.
 
@@ -15,4 +17,9 @@ entry:
 
 ; CHECK: stvx 2,
 
+; We would prefer to test for "stxvw4x 34," but current -O0 code
+; needlessly generates "vor 3,2,2 / stxvw4x 35,0,3", so we'll settle for
+; the opcode.
+; CHECK-VSX: stxvw4x
+
 declare void @foo(i32*)

diff --git a/test/CodeGen/PowerPC/vsx-p8.ll b/test/CodeGen/PowerPC/vsx-p8.ll
new file mode 100644 (file)
index 0000000..81406b6
--- /dev/null
+++ b/test/CodeGen/PowerPC/vsx-p8.ll
@@ -0,0 +1,42 @@
+; RUN: llc -mcpu=pwr8 -mattr=+power8-vector < %s | FileCheck %s
+target datalayout = "E-m:e-i64:64-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+; Unaligned loads/stores on P8 and later should use VSX where possible.
+
+define <2 x double> @test28u(<2 x double>* %a) {
+  %v = load <2 x double>* %a, align 8
+  ret <2 x double> %v
+
+; CHECK-LABEL: @test28u
+; CHECK: lxvd2x 34, 0, 3
+; CHECK: blr
+}
+
+define void @test29u(<2 x double>* %a, <2 x double> %b) {
+  store <2 x double> %b, <2 x double>* %a, align 8
+  ret void
+
+; CHECK-LABEL: @test29u
+; CHECK: stxvd2x 34, 0, 3
+; CHECK: blr
+}
+
+define <4 x float> @test32u(<4 x float>* %a) {
+  %v = load <4 x float>* %a, align 8
+  ret <4 x float> %v
+
+; CHECK-LABEL: @test32u
+; CHECK: lxvw4x 34, 0, 3
+; CHECK: blr
+}
+
+define void @test33u(<4 x float>* %a, <4 x float> %b) {
+  store <4 x float> %b, <4 x float>* %a, align 8
+  ret void
+
+; CHECK-LABEL: @test33u
+; CHECK: stxvw4x 34, 0, 3
+; CHECK: blr
+}
+

diff --git a/test/CodeGen/PowerPC/vsx.ll b/test/CodeGen/PowerPC/vsx.ll
index 2f226e1f614cc523f31a6f61fa2d30f06be08d92..65343f4a9ba66b00fec03b335b9fb4bc333bb7df 100644 (file)
--- a/test/CodeGen/PowerPC/vsx.ll
+++ b/test/CodeGen/PowerPC/vsx.ll
@@ -356,6 +356,63 @@ define void @test31(<2 x i64>* %a, <2 x i64> %b) {
 ; CHECK: blr
 }
 
+define <4 x float> @test32(<4 x float>* %a) {
+  %v = load <4 x float>* %a, align 16
+  ret <4 x float> %v
+
+; CHECK-LABEL: @test32
+; CHECK: lxvw4x 34, 0, 3
+; CHECK: blr
+}
+
+define void @test33(<4 x float>* %a, <4 x float> %b) {
+  store <4 x float> %b, <4 x float>* %a, align 16
+  ret void
+
+; CHECK-LABEL: @test33
+; CHECK: stxvw4x 34, 0, 3
+; CHECK: blr
+}
+
+define <4 x float> @test32u(<4 x float>* %a) {
+  %v = load <4 x float>* %a, align 8
+  ret <4 x float> %v
+
+; CHECK-LABEL: @test32u
+; CHECK-DAG: lvsl
+; CHECK-DAG: lvx
+; CHECK-DAG: lvx
+; CHECK: vperm 2,
+; CHECK: blr
+}
+
+define void @test33u(<4 x float>* %a, <4 x float> %b) {
+  store <4 x float> %b, <4 x float>* %a, align 8
+  ret void
+
+; CHECK-LABEL: @test33u
+; CHECK: stxvw4x 34, 0, 3
+; CHECK: blr
+}
+
+define <4 x i32> @test34(<4 x i32>* %a) {
+  %v = load <4 x i32>* %a, align 16
+  ret <4 x i32> %v
+
+; CHECK-LABEL: @test34
+; CHECK: lxvw4x 34, 0, 3
+; CHECK: blr
+}
+
+define void @test35(<4 x i32>* %a, <4 x i32> %b) {
+  store <4 x i32> %b, <4 x i32>* %a, align 16
+  ret void
+
+; CHECK-LABEL: @test35
+; CHECK: stxvw4x 34, 0, 3
+; CHECK: blr
+}
+
 define <2 x double> @test40(<2 x i64> %a) {
   %v = uitofp <2 x i64> %a to <2 x double>
   ret <2 x double> %v