Propagate DAG node ordering during type legalization and instruction selection

A node's ordering is only propagated during legalization if (a) the new node does
not have an ordering (is not a CSE'd node), or (b) the new node has an ordering
that is higher than the node being legalized.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177465 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 71 insertions, 8 deletions

diff --git a/test/CodeGen/NVPTX/sched1.ll b/test/CodeGen/NVPTX/sched1.ll
new file mode 100644
index 0000000000..03ab635e73
--- /dev/null
+++ b/test/CodeGen/NVPTX/sched1.ll
@@ -0,0 +1,31 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
+
+; Ensure source scheduling is working
+
+define void @foo(i32* %a) {
+; CHECK: .func foo
+; CHECK: ld.u32
+; CHECK-NEXT: ld.u32
+; CHECK-NEXT: ld.u32
+; CHECK-NEXT: ld.u32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+  %ptr0 = getelementptr i32* %a, i32 0
+  %val0 = load i32* %ptr0
+  %ptr1 = getelementptr i32* %a, i32 1
+  %val1 = load i32* %ptr1
+  %ptr2 = getelementptr i32* %a, i32 2
+  %val2 = load i32* %ptr2
+  %ptr3 = getelementptr i32* %a, i32 3
+  %val3 = load i32* %ptr3
+
+  %t0 = add i32 %val0, %val1
+  %t1 = add i32 %t0, %val2
+  %t2 = add i32 %t1, %val3
+
+  store i32 %t2, i32* %a
+
+  ret void
+}
+
diff --git a/test/CodeGen/NVPTX/sched2.ll b/test/CodeGen/NVPTX/sched2.ll
new file mode 100644
index 0000000000..71a9a4963f
--- /dev/null
+++ b/test/CodeGen/NVPTX/sched2.ll
@@ -0,0 +1,32 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
+
+define void @foo(<2 x i32>* %a) {
+; CHECK: .func foo
+; CHECK: ld.v2.u32
+; CHECK-NEXT: ld.v2.u32
+; CHECK-NEXT: ld.v2.u32
+; CHECK-NEXT: ld.v2.u32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+; CHECK-NEXT: add.s32
+  %ptr0 = getelementptr <2 x i32>* %a, i32 0
+  %val0 = load <2 x i32>* %ptr0
+  %ptr1 = getelementptr <2 x i32>* %a, i32 1
+  %val1 = load <2 x i32>* %ptr1
+  %ptr2 = getelementptr <2 x i32>* %a, i32 2
+  %val2 = load <2 x i32>* %ptr2
+  %ptr3 = getelementptr <2 x i32>* %a, i32 3
+  %val3 = load <2 x i32>* %ptr3
+
+  %t0 = add <2 x i32> %val0, %val1
+  %t1 = add <2 x i32> %t0, %val2
+  %t2 = add <2 x i32> %t1, %val3
+
+  store <2 x i32> %t2, <2 x i32>* %a
+
+  ret void
+}
+
diff --git a/test/CodeGen/NVPTX/vector-loads.ll b/test/CodeGen/NVPTX/vector-loads.ll
index f5a1795e3c..58882bf166 100644
--- a/test/CodeGen/NVPTX/vector-loads.ll
+++ b/test/CodeGen/NVPTX/vector-loads.ll
@@ -9,7 +9,7 @@
 
 define void @foo(<2 x float>* %a) {
 ; CHECK: .func foo
-; CHECK: ld.v2.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}}, [%r{{[0-9]+}}];
+; CHECK: ld.v2.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}}
   %t1 = load <2 x float>* %a
   %t2 = fmul <2 x float> %t1, %t1
   store <2 x float> %t2, <2 x float>* %a
@@ -18,7 +18,7 @@ define void @foo(<2 x float>* %a) {
 
 define void @foo2(<4 x float>* %a) {
 ; CHECK: .func foo2
-; CHECK: ld.v4.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}}, [%r{{[0-9]+}}];
+; CHECK: ld.v4.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}}
   %t1 = load <4 x float>* %a
   %t2 = fmul <4 x float> %t1, %t1
   store <4 x float> %t2, <4 x float>* %a
@@ -27,8 +27,8 @@ define void @foo2(<4 x float>* %a) {
 
 define void @foo3(<8 x float>* %a) {
 ; CHECK: .func foo3
-; CHECK: ld.v4.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}}, [%r{{[0-9]+}}];
-; CHECK-NEXT: ld.v4.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}}, [%r{{[0-9]+}}+16];
+; CHECK: ld.v4.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}}
+; CHECK-NEXT: ld.v4.f32 {%f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}, %f{{[0-9]+}}}
   %t1 = load <8 x float>* %a
   %t2 = fmul <8 x float> %t1, %t1
   store <8 x float> %t2, <8 x float>* %a
@@ -39,7 +39,7 @@ define void @foo3(<8 x float>* %a) {
 
 define void @foo4(<2 x i32>* %a) {
 ; CHECK: .func foo4
-; CHECK: ld.v2.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}}, [%r{{[0-9]+}}];
+; CHECK: ld.v2.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}}
   %t1 = load <2 x i32>* %a
   %t2 = mul <2 x i32> %t1, %t1
   store <2 x i32> %t2, <2 x i32>* %a
@@ -48,7 +48,7 @@ define void @foo4(<2 x i32>* %a) {
 
 define void @foo5(<4 x i32>* %a) {
 ; CHECK: .func foo5
-; CHECK: ld.v4.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}}, [%r{{[0-9]+}}];
+; CHECK: ld.v4.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}}
   %t1 = load <4 x i32>* %a
   %t2 = mul <4 x i32> %t1, %t1
   store <4 x i32> %t2, <4 x i32>* %a
@@ -57,8 +57,8 @@ define void @foo5(<4 x i32>* %a) {
 
 define void @foo6(<8 x i32>* %a) {
 ; CHECK: .func foo6
-; CHECK: ld.v4.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}}, [%r{{[0-9]+}}];
-; CHECK-NEXT: ld.v4.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}}, [%r{{[0-9]+}}+16];
+; CHECK: ld.v4.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}}
+; CHECK-NEXT: ld.v4.u32 {%r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}, %r{{[0-9]+}}}
   %t1 = load <8 x i32>* %a
   %t2 = mul <8 x i32> %t1, %t1
   store <8 x i32> %t2, <8 x i32>* %a