[MachineScheduler] checkResourceLimit boundary condition update

When we call checkResourceLimit in bumpCycle or bumpNode, and we
know the resource count has just reached the limit (the equations
 are equal). We should return true to mark that we are resource
limited for next schedule, or else we might continue to schedule
in favor of latency for 1 more schedule and create a schedule that
 actually overbook the resource.

When we call checkResourceLimit to estimate the resource limite before
scheduling, we don't need to return true even if the equations are
equal, as it shouldn't limit the schedule for it .

Differential Revision: https://reviews.llvm.org/D62345

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@362805 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/MachineScheduler.cpp b/lib/CodeGen/MachineScheduler.cpp
index 92c051d5c467e8d930d565338d49c4cf1a1926bd..ae1170ad1be65d1ac6d3a0fbfe15dc6afa495626 100644 (file)
--- a/lib/CodeGen/MachineScheduler.cpp
+++ b/lib/CodeGen/MachineScheduler.cpp
@@ -1837,9 +1837,15 @@ SchedBoundary::~SchedBoundary() { delete HazardRec; }
 
 /// Given a Count of resource usage and a Latency value, return true if a
 /// SchedBoundary becomes resource limited.
+/// If we are checking after scheduling a node, we should return true when
+/// we just reach the resource limit.
 static bool checkResourceLimit(unsigned LFactor, unsigned Count,
-                               unsigned Latency) {
-  return (int)(Count - (Latency * LFactor)) > (int)LFactor;
+                               unsigned Latency, bool AfterSchedNode) {
+  int ResCntFactor = (int)(Count - (Latency * LFactor));
+  if (AfterSchedNode)
+    return ResCntFactor >= (int)LFactor;
+  else
+    return ResCntFactor > (int)LFactor;
 }
 
 void SchedBoundary::reset() {
@@ -2134,7 +2140,7 @@ void SchedBoundary::bumpCycle(unsigned NextCycle) {
   CheckPending = true;
   IsResourceLimited =
       checkResourceLimit(SchedModel->getLatencyFactor(), getCriticalCount(),
-                         getScheduledLatency());
+                         getScheduledLatency(), true);
 
   LLVM_DEBUG(dbgs() << "Cycle: " << CurrCycle << ' ' << Available.getName()
                     << '\n');
@@ -2302,7 +2308,7 @@ void SchedBoundary::bumpNode(SUnit *SU) {
     // resource limited. If a stall occurred, bumpCycle does this.
     IsResourceLimited =
         checkResourceLimit(SchedModel->getLatencyFactor(), getCriticalCount(),
-                           getScheduledLatency());
+                           getScheduledLatency(), true);
 
   // Update CurrMOps after calling bumpCycle to handle stalls, since bumpCycle
   // resets CurrMOps. Loop to handle instructions with more MOps than issue in
@@ -2521,7 +2527,7 @@ void GenericSchedulerBase::setPolicy(CandPolicy &Policy, bool IsPostRA,
     RemLatency = computeRemLatency(CurrZone);
     RemLatencyComputed = true;
     OtherResLimited = checkResourceLimit(SchedModel->getLatencyFactor(),
-                                         OtherCount, RemLatency);
+                                         OtherCount, RemLatency, false);
   }
 
   // Schedule aggressively for latency in PostRA mode. We don't check for

diff --git a/test/CodeGen/PowerPC/build-vector-tests.ll b/test/CodeGen/PowerPC/build-vector-tests.ll
index e94808f0a2d270165d2989d68550b49a5f550050..adbd7622a80c81983c0f529dcc3bbe16a47fea13 100644 (file)
--- a/test/CodeGen/PowerPC/build-vector-tests.ll
+++ b/test/CodeGen/PowerPC/build-vector-tests.ll
@@ -2024,8 +2024,8 @@ define <4 x i32> @fromDiffMemConsDConvdtoi(double* nocapture readonly %ptr) {
 ; P9LE:       # %bb.0: # %entry
 ; P9LE-NEXT:    lfd f0, 24(r3)
 ; P9LE-NEXT:    lfd f2, 8(r3)
-; P9LE-NEXT:    lfd f1, 16(r3)
 ; P9LE-NEXT:    xxmrghd vs0, vs2, vs0
+; P9LE-NEXT:    lfd f1, 16(r3)
 ; P9LE-NEXT:    lfd f3, 0(r3)
 ; P9LE-NEXT:    xvcvdpsxws v2, vs0
 ; P9LE-NEXT:    xxmrghd vs0, vs3, vs1
@@ -3608,8 +3608,8 @@ define <4 x i32> @fromDiffMemConsDConvdtoui(double* nocapture readonly %ptr) {
 ; P9LE:       # %bb.0: # %entry
 ; P9LE-NEXT:    lfd f0, 24(r3)
 ; P9LE-NEXT:    lfd f2, 8(r3)
-; P9LE-NEXT:    lfd f1, 16(r3)
 ; P9LE-NEXT:    xxmrghd vs0, vs2, vs0
+; P9LE-NEXT:    lfd f1, 16(r3)
 ; P9LE-NEXT:    lfd f3, 0(r3)
 ; P9LE-NEXT:    xvcvdpuxws v2, vs0
 ; P9LE-NEXT:    xxmrghd vs0, vs3, vs1

diff --git a/test/CodeGen/PowerPC/csr-save-restore-order.ll b/test/CodeGen/PowerPC/csr-save-restore-order.ll
index dc858953e8120b1b27ef02b2e9685d04d3e43971..ad9b781e852bae8d8d3745fb9a353e8f88f1c7f0 100644 (file)
--- a/test/CodeGen/PowerPC/csr-save-restore-order.ll
+++ b/test/CodeGen/PowerPC/csr-save-restore-order.ll
@@ -57,9 +57,9 @@ define i64 @test(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
 ; CHECK-PWR9-NEXT:    .cfi_offset v31, -304
 ; CHECK-PWR9-NEXT:    std r14, 240(r1) # 8-byte Folded Spill
 ; CHECK-PWR9-NEXT:    std r15, 248(r1) # 8-byte Folded Spill
-; CHECK-PWR9-NEXT:    std r16, 256(r1) # 8-byte Folded Spill
 ; CHECK-PWR9-NEXT:    stxv v20, 48(r1) # 16-byte Folded Spill
 ; CHECK-PWR9-NEXT:    stxv v21, 64(r1) # 16-byte Folded Spill
+; CHECK-PWR9-NEXT:    std r16, 256(r1) # 8-byte Folded Spill
 ; CHECK-PWR9-NEXT:    stxv v22, 80(r1) # 16-byte Folded Spill
 ; CHECK-PWR9-NEXT:    std r17, 264(r1) # 8-byte Folded Spill
 ; CHECK-PWR9-NEXT:    stxv v23, 96(r1) # 16-byte Folded Spill
@@ -112,8 +112,8 @@ define i64 @test(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
 ; CHECK-PWR9-NEXT:    ld r4, 32(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lxv v31, 224(r1) # 16-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lxv v30, 208(r1) # 16-byte Folded Reload
-; CHECK-PWR9-NEXT:    lxv v29, 192(r1) # 16-byte Folded Reload
 ; CHECK-PWR9-NEXT:    add r3, r4, r3
+; CHECK-PWR9-NEXT:    lxv v29, 192(r1) # 16-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lxv v28, 176(r1) # 16-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lxv v27, 160(r1) # 16-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lxv v26, 144(r1) # 16-byte Folded Reload
@@ -128,11 +128,11 @@ define i64 @test(i64 %p1, i64 %p2, i64 %p3, i64 %p4) {
 ; CHECK-PWR9-NEXT:    lfd f29, 504(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lfd f28, 496(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    lfd f27, 488(r1) # 8-byte Folded Reload
-; CHECK-PWR9-NEXT:    lfd f26, 480(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r31, 376(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r30, 368(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r29, 360(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r28, 352(r1) # 8-byte Folded Reload
+; CHECK-PWR9-NEXT:    lfd f26, 480(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r27, 344(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r26, 336(r1) # 8-byte Folded Reload
 ; CHECK-PWR9-NEXT:    ld r25, 328(r1) # 8-byte Folded Reload

diff --git a/test/CodeGen/PowerPC/f128-passByValue.ll b/test/CodeGen/PowerPC/f128-passByValue.ll
index 7d400b85310abb8f2ef64f0a7703c928c4c8eafb..cbccaea3bce189f833142fa20bf2caf882f99dd2 100644 (file)
--- a/test/CodeGen/PowerPC/f128-passByValue.ll
+++ b/test/CodeGen/PowerPC/f128-passByValue.ll
@@ -85,6 +85,7 @@ define fp128 @maxVecParam(fp128 %p1, fp128 %p2, fp128 %p3, fp128 %p4, fp128 %p5,
 ; CHECK-LABEL: maxVecParam:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    xsaddqp v2, v2, v3
+; CHECK-NEXT:    lxv v[[REG0:[0-9]+]], 224(r1)
 ; CHECK-NEXT:    xsaddqp v2, v2, v4
 ; CHECK-NEXT:    xsaddqp v2, v2, v5
 ; CHECK-NEXT:    xsaddqp v2, v2, v6
@@ -95,7 +96,6 @@ define fp128 @maxVecParam(fp128 %p1, fp128 %p2, fp128 %p3, fp128 %p4, fp128 %p5,
 ; CHECK-NEXT:    xsaddqp v2, v2, v11
 ; CHECK-NEXT:    xsaddqp v2, v2, v12
 ; CHECK-NEXT:    xsaddqp v2, v2, v13
-; CHECK-NEXT:    lxv v[[REG0:[0-9]+]], 224(r1)
 ; CHECK-NEXT:    xssubqp v2, v2, v[[REG0]]
 ; CHECK-NEXT:    blr
                           fp128 %p6, fp128 %p7, fp128 %p8, fp128 %p9, fp128 %p10,

diff --git a/test/CodeGen/PowerPC/vec_conv_fp_to_i_8byte_elts.ll b/test/CodeGen/PowerPC/vec_conv_fp_to_i_8byte_elts.ll
index a059e56bc3f66db41a0cc575d5bc666fa79bca26..dfca814cab23ebccc76cdba0800cb0ef50d7436d 100644 (file)
--- a/test/CodeGen/PowerPC/vec_conv_fp_to_i_8byte_elts.ll
+++ b/test/CodeGen/PowerPC/vec_conv_fp_to_i_8byte_elts.ll
@@ -167,15 +167,15 @@ define void @test16elt(<16 x i64>* noalias nocapture sret %agg.result, <16 x dou
 ; CHECK-P9-NEXT:    lxv vs1, 96(r4)
 ; CHECK-P9-NEXT:    lxv vs2, 80(r4)
 ; CHECK-P9-NEXT:    lxv vs3, 64(r4)
+; CHECK-P9-NEXT:    xvcvdpuxds vs3, vs3
 ; CHECK-P9-NEXT:    lxv vs4, 48(r4)
-; CHECK-P9-NEXT:    xvcvdpuxds vs4, vs4
 ; CHECK-P9-NEXT:    lxv vs5, 32(r4)
 ; CHECK-P9-NEXT:    lxv vs6, 16(r4)
 ; CHECK-P9-NEXT:    lxv vs7, 0(r4)
 ; CHECK-P9-NEXT:    xvcvdpuxds vs7, vs7
 ; CHECK-P9-NEXT:    xvcvdpuxds vs6, vs6
 ; CHECK-P9-NEXT:    xvcvdpuxds vs5, vs5
-; CHECK-P9-NEXT:    xvcvdpuxds vs3, vs3
+; CHECK-P9-NEXT:    xvcvdpuxds vs4, vs4
 ; CHECK-P9-NEXT:    xvcvdpuxds vs2, vs2
 ; CHECK-P9-NEXT:    xvcvdpuxds vs1, vs1
 ; CHECK-P9-NEXT:    xvcvdpuxds vs0, vs0
@@ -195,15 +195,15 @@ define void @test16elt(<16 x i64>* noalias nocapture sret %agg.result, <16 x dou
 ; CHECK-BE-NEXT:    lxv vs1, 96(r4)
 ; CHECK-BE-NEXT:    lxv vs2, 80(r4)
 ; CHECK-BE-NEXT:    lxv vs3, 64(r4)
+; CHECK-BE-NEXT:    xvcvdpuxds vs3, vs3
 ; CHECK-BE-NEXT:    lxv vs4, 48(r4)
-; CHECK-BE-NEXT:    xvcvdpuxds vs4, vs4
 ; CHECK-BE-NEXT:    lxv vs5, 32(r4)
 ; CHECK-BE-NEXT:    lxv vs6, 16(r4)
 ; CHECK-BE-NEXT:    lxv vs7, 0(r4)
 ; CHECK-BE-NEXT:    xvcvdpuxds vs7, vs7
 ; CHECK-BE-NEXT:    xvcvdpuxds vs6, vs6
 ; CHECK-BE-NEXT:    xvcvdpuxds vs5, vs5
-; CHECK-BE-NEXT:    xvcvdpuxds vs3, vs3
+; CHECK-BE-NEXT:    xvcvdpuxds vs4, vs4
 ; CHECK-BE-NEXT:    xvcvdpuxds vs2, vs2
 ; CHECK-BE-NEXT:    xvcvdpuxds vs1, vs1
 ; CHECK-BE-NEXT:    xvcvdpuxds vs0, vs0
@@ -381,15 +381,15 @@ define void @test16elt_signed(<16 x i64>* noalias nocapture sret %agg.result, <1
 ; CHECK-P9-NEXT:    lxv vs1, 96(r4)
 ; CHECK-P9-NEXT:    lxv vs2, 80(r4)
 ; CHECK-P9-NEXT:    lxv vs3, 64(r4)
+; CHECK-P9-NEXT:    xvcvdpsxds vs3, vs3
 ; CHECK-P9-NEXT:    lxv vs4, 48(r4)
-; CHECK-P9-NEXT:    xvcvdpsxds vs4, vs4
 ; CHECK-P9-NEXT:    lxv vs5, 32(r4)
 ; CHECK-P9-NEXT:    lxv vs6, 16(r4)
 ; CHECK-P9-NEXT:    lxv vs7, 0(r4)
 ; CHECK-P9-NEXT:    xvcvdpsxds vs7, vs7
 ; CHECK-P9-NEXT:    xvcvdpsxds vs6, vs6
 ; CHECK-P9-NEXT:    xvcvdpsxds vs5, vs5
-; CHECK-P9-NEXT:    xvcvdpsxds vs3, vs3
+; CHECK-P9-NEXT:    xvcvdpsxds vs4, vs4
 ; CHECK-P9-NEXT:    xvcvdpsxds vs2, vs2
 ; CHECK-P9-NEXT:    xvcvdpsxds vs1, vs1
 ; CHECK-P9-NEXT:    xvcvdpsxds vs0, vs0
@@ -409,15 +409,15 @@ define void @test16elt_signed(<16 x i64>* noalias nocapture sret %agg.result, <1
 ; CHECK-BE-NEXT:    lxv vs1, 96(r4)
 ; CHECK-BE-NEXT:    lxv vs2, 80(r4)
 ; CHECK-BE-NEXT:    lxv vs3, 64(r4)
+; CHECK-BE-NEXT:    xvcvdpsxds vs3, vs3
 ; CHECK-BE-NEXT:    lxv vs4, 48(r4)
-; CHECK-BE-NEXT:    xvcvdpsxds vs4, vs4
 ; CHECK-BE-NEXT:    lxv vs5, 32(r4)
 ; CHECK-BE-NEXT:    lxv vs6, 16(r4)
 ; CHECK-BE-NEXT:    lxv vs7, 0(r4)
 ; CHECK-BE-NEXT:    xvcvdpsxds vs7, vs7
 ; CHECK-BE-NEXT:    xvcvdpsxds vs6, vs6
 ; CHECK-BE-NEXT:    xvcvdpsxds vs5, vs5
-; CHECK-BE-NEXT:    xvcvdpsxds vs3, vs3
+; CHECK-BE-NEXT:    xvcvdpsxds vs4, vs4
 ; CHECK-BE-NEXT:    xvcvdpsxds vs2, vs2
 ; CHECK-BE-NEXT:    xvcvdpsxds vs1, vs1
 ; CHECK-BE-NEXT:    xvcvdpsxds vs0, vs0

diff --git a/test/CodeGen/PowerPC/vec_conv_i_to_fp_8byte_elts.ll b/test/CodeGen/PowerPC/vec_conv_i_to_fp_8byte_elts.ll
index 45e306ec6bf4f5b81ebc9c6df3e196ae86abd27d..a18eab406657dee7bd8b512c5edb99f529b1fdf5 100644 (file)
--- a/test/CodeGen/PowerPC/vec_conv_i_to_fp_8byte_elts.ll
+++ b/test/CodeGen/PowerPC/vec_conv_i_to_fp_8byte_elts.ll
@@ -167,15 +167,15 @@ define void @test16elt(<16 x double>* noalias nocapture sret %agg.result, <16 x
 ; CHECK-P9-NEXT:    lxv v3, 96(r4)
 ; CHECK-P9-NEXT:    lxv v4, 80(r4)
 ; CHECK-P9-NEXT:    lxv v5, 64(r4)
+; CHECK-P9-NEXT:    xvcvuxddp vs4, v5
 ; CHECK-P9-NEXT:    lxv v0, 48(r4)
-; CHECK-P9-NEXT:    xvcvuxddp vs3, v0
 ; CHECK-P9-NEXT:    lxv v1, 32(r4)
 ; CHECK-P9-NEXT:    lxv v6, 16(r4)
 ; CHECK-P9-NEXT:    lxv v7, 0(r4)
 ; CHECK-P9-NEXT:    xvcvuxddp vs0, v7
 ; CHECK-P9-NEXT:    xvcvuxddp vs1, v6
 ; CHECK-P9-NEXT:    xvcvuxddp vs2, v1
-; CHECK-P9-NEXT:    xvcvuxddp vs4, v5
+; CHECK-P9-NEXT:    xvcvuxddp vs3, v0
 ; CHECK-P9-NEXT:    xvcvuxddp vs5, v4
 ; CHECK-P9-NEXT:    xvcvuxddp vs6, v3
 ; CHECK-P9-NEXT:    xvcvuxddp vs7, v2
@@ -195,15 +195,15 @@ define void @test16elt(<16 x double>* noalias nocapture sret %agg.result, <16 x
 ; CHECK-BE-NEXT:    lxv v3, 96(r4)
 ; CHECK-BE-NEXT:    lxv v4, 80(r4)
 ; CHECK-BE-NEXT:    lxv v5, 64(r4)
+; CHECK-BE-NEXT:    xvcvuxddp vs4, v5
 ; CHECK-BE-NEXT:    lxv v0, 48(r4)
-; CHECK-BE-NEXT:    xvcvuxddp vs3, v0
 ; CHECK-BE-NEXT:    lxv v1, 32(r4)
 ; CHECK-BE-NEXT:    lxv v6, 16(r4)
 ; CHECK-BE-NEXT:    lxv v7, 0(r4)
 ; CHECK-BE-NEXT:    xvcvuxddp vs0, v7
 ; CHECK-BE-NEXT:    xvcvuxddp vs1, v6
 ; CHECK-BE-NEXT:    xvcvuxddp vs2, v1
-; CHECK-BE-NEXT:    xvcvuxddp vs4, v5
+; CHECK-BE-NEXT:    xvcvuxddp vs3, v0
 ; CHECK-BE-NEXT:    xvcvuxddp vs5, v4
 ; CHECK-BE-NEXT:    xvcvuxddp vs6, v3
 ; CHECK-BE-NEXT:    xvcvuxddp vs7, v2
@@ -381,15 +381,15 @@ define void @test16elt_signed(<16 x double>* noalias nocapture sret %agg.result,
 ; CHECK-P9-NEXT:    lxv v3, 96(r4)
 ; CHECK-P9-NEXT:    lxv v4, 80(r4)
 ; CHECK-P9-NEXT:    lxv v5, 64(r4)
+; CHECK-P9-NEXT:    xvcvsxddp vs4, v5
 ; CHECK-P9-NEXT:    lxv v0, 48(r4)
-; CHECK-P9-NEXT:    xvcvsxddp vs3, v0
 ; CHECK-P9-NEXT:    lxv v1, 32(r4)
 ; CHECK-P9-NEXT:    lxv v6, 16(r4)
 ; CHECK-P9-NEXT:    lxv v7, 0(r4)
 ; CHECK-P9-NEXT:    xvcvsxddp vs0, v7
 ; CHECK-P9-NEXT:    xvcvsxddp vs1, v6
 ; CHECK-P9-NEXT:    xvcvsxddp vs2, v1
-; CHECK-P9-NEXT:    xvcvsxddp vs4, v5
+; CHECK-P9-NEXT:    xvcvsxddp vs3, v0
 ; CHECK-P9-NEXT:    xvcvsxddp vs5, v4
 ; CHECK-P9-NEXT:    xvcvsxddp vs6, v3
 ; CHECK-P9-NEXT:    xvcvsxddp vs7, v2
@@ -409,15 +409,15 @@ define void @test16elt_signed(<16 x double>* noalias nocapture sret %agg.result,
 ; CHECK-BE-NEXT:    lxv v3, 96(r4)
 ; CHECK-BE-NEXT:    lxv v4, 80(r4)
 ; CHECK-BE-NEXT:    lxv v5, 64(r4)
+; CHECK-BE-NEXT:    xvcvsxddp vs4, v5
 ; CHECK-BE-NEXT:    lxv v0, 48(r4)
-; CHECK-BE-NEXT:    xvcvsxddp vs3, v0
 ; CHECK-BE-NEXT:    lxv v1, 32(r4)
 ; CHECK-BE-NEXT:    lxv v6, 16(r4)
 ; CHECK-BE-NEXT:    lxv v7, 0(r4)
 ; CHECK-BE-NEXT:    xvcvsxddp vs0, v7
 ; CHECK-BE-NEXT:    xvcvsxddp vs1, v6
 ; CHECK-BE-NEXT:    xvcvsxddp vs2, v1
-; CHECK-BE-NEXT:    xvcvsxddp vs4, v5
+; CHECK-BE-NEXT:    xvcvsxddp vs3, v0
 ; CHECK-BE-NEXT:    xvcvsxddp vs5, v4
 ; CHECK-BE-NEXT:    xvcvsxddp vs6, v3
 ; CHECK-BE-NEXT:    xvcvsxddp vs7, v2