[SPARC] Switch to the Machine Scheduler.

The (mostly-deprecated) SelectionDAG-based ILPListDAGScheduler scheduler
was making poor scheduling decisions, causing high register pressure and
extraneous register spills.

Switching to the newer machine scheduler generates better code -- even
without there being a machine model defined for SPARC yet.

(Actually committing the test changes too, this time, unlike r247315)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@247343 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/Sparc/SparcSubtarget.cpp b/lib/Target/Sparc/SparcSubtarget.cpp
index 6bf8d41306263449b5e5499b7a0f43a3bc498296..d701594d27af018a702f1df43f37a320c5dd3c7b 100644 (file)
--- a/lib/Target/Sparc/SparcSubtarget.cpp
+++ b/lib/Target/Sparc/SparcSubtarget.cpp
@@ -81,3 +81,7 @@ int SparcSubtarget::getAdjustedFrameSize(int frameSize) const {
   }
   return frameSize;
 }
+
+bool SparcSubtarget::enableMachineScheduler() const {
+  return true;
+}

diff --git a/lib/Target/Sparc/SparcSubtarget.h b/lib/Target/Sparc/SparcSubtarget.h
index 9d21911d88f03abfdf89e399d2503897e4d93ded..e2fd2f04528af31a7b716e7efb8d5d0483d784ca 100644 (file)
--- a/lib/Target/Sparc/SparcSubtarget.h
+++ b/lib/Target/Sparc/SparcSubtarget.h
@@ -60,6 +60,8 @@ public:
     return &TSInfo;
   }
 
+  bool enableMachineScheduler() const override;
+
   bool isV9() const { return IsV9; }
   bool isVIS() const { return IsVIS; }
   bool isVIS2() const { return IsVIS2; }
@@ -85,7 +87,6 @@ public:
   /// returns adjusted framesize which includes space for register window
   /// spills and arguments.
   int getAdjustedFrameSize(int stackSize) const;
-
 };
 
 } // end namespace llvm

diff --git a/test/CodeGen/SPARC/2011-01-22-SRet.ll b/test/CodeGen/SPARC/2011-01-22-SRet.ll
index ae9764e820845b50f44a98327ac538491f5ad471..678544ebf2cb47241fe6130ef0af7be6b5d3bfc8 100644 (file)
--- a/test/CodeGen/SPARC/2011-01-22-SRet.ll
+++ b/test/CodeGen/SPARC/2011-01-22-SRet.ll
@@ -19,8 +19,8 @@ entry:
 define i32 @test() nounwind {
 entry:
 ;CHECK-LABEL: test:
-;CHECK: st {{.+}}, [%sp+64]
 ;CHECK: call make_foo
+;CHECK: st {{.+}}, [%sp+64]
 ;CHECK: unimp 12
   %f = alloca %struct.foo_t, align 8
   call void @make_foo(%struct.foo_t* noalias sret %f, i32 10, i32 20, i32 30) nounwind

diff --git a/test/CodeGen/SPARC/64abi.ll b/test/CodeGen/SPARC/64abi.ll
index 7c08998a142799df145b66c137705c507648dfac..96104ecc3c68ca9c96ba6711fc3d5b3e48261b04 100644 (file)
--- a/test/CodeGen/SPARC/64abi.ll
+++ b/test/CodeGen/SPARC/64abi.ll
@@ -1,19 +1,19 @@
 ; RUN: llc < %s -march=sparcv9 -disable-sparc-delay-filler -disable-sparc-leaf-proc | FileCheck %s
 
-; CHECK: intarg
+; CHECK-LABEL: intarg:
 ; The save/restore frame is not strictly necessary here, but we would need to
 ; refer to %o registers instead.
 ; CHECK: save %sp, -128, %sp
+; CHECK: ldx [%fp+2231], [[R2:%[gilo][0-7]]]
+; CHECK: ld [%fp+2227], [[R1:%[gilo][0-7]]]
 ; CHECK: stb %i0, [%i4]
 ; CHECK: stb %i1, [%i4]
 ; CHECK: sth %i2, [%i4]
 ; CHECK: st  %i3, [%i4]
 ; CHECK: stx %i4, [%i4]
 ; CHECK: st  %i5, [%i4]
-; CHECK: ld [%fp+2227], [[R:%[gilo][0-7]]]
-; CHECK: st  [[R]], [%i4]
-; CHECK: ldx [%fp+2231], [[R:%[gilo][0-7]]]
-; CHECK: stx [[R]], [%i4]
+; CHECK: st  [[R1]], [%i4]
+; CHECK: stx [[R2]], [%i4]
 ; CHECK: restore
 define void @intarg(i8  %a0,   ; %i0
                     i8  %a1,   ; %i1
@@ -37,14 +37,14 @@ define void @intarg(i8  %a0,   ; %i0
   ret void
 }
 
-; CHECK: call_intarg
+; CHECK-LABEL: call_intarg:
 ; 16 saved + 8 args.
 ; CHECK: save %sp, -192, %sp
 ; Sign-extend and store the full 64 bits.
 ; CHECK: sra %i0, 0, [[R:%[gilo][0-7]]]
-; CHECK: stx [[R]], [%sp+2223]
 ; Use %o0-%o5 for outgoing arguments
 ; CHECK: mov 5, %o5
+; CHECK: stx [[R]], [%sp+2223]
 ; CHECK: call intarg
 ; CHECK-NOT: add %sp
 ; CHECK: restore
@@ -53,13 +53,13 @@ define void @call_intarg(i32 %i0, i8* %i1) {
   ret void
 }
 
-; CHECK: floatarg
+; CHECK-LABEL: floatarg:
 ; CHECK: save %sp, -128, %sp
+; CHECK: ld [%fp+2307], [[F:%f[0-9]+]]
 ; CHECK: fstod %f1,
 ; CHECK: faddd %f2,
 ; CHECK: faddd %f4,
 ; CHECK: faddd %f6,
-; CHECK: ld [%fp+2307], [[F:%f[0-9]+]]
 ; CHECK: fadds %f31, [[F]]
 define double @floatarg(float %a0,    ; %f1
                         double %a1,   ; %d2
@@ -89,12 +89,12 @@ define double @floatarg(float %a0,    ; %f1
   ret double %s17
 }
 
-; CHECK: call_floatarg
+; CHECK-LABEL: call_floatarg:
 ; CHECK: save %sp, -272, %sp
-; Store 4 bytes, right-aligned in slot.
-; CHECK: st %f1, [%sp+2307]
 ; Store 8 bytes in full slot.
 ; CHECK: std %f2, [%sp+2311]
+; Store 4 bytes, right-aligned in slot.
+; CHECK: st %f1, [%sp+2307]
 ; CHECK: fmovd %f2, %f4
 ; CHECK: call floatarg
 ; CHECK-NOT: add %sp
@@ -109,12 +109,12 @@ define void @call_floatarg(float %f1, double %d2, float %f5, double *%p) {
   ret void
 }
 
-; CHECK: mixedarg
+; CHECK-LABEL: mixedarg:
+; CHECK: ldx [%fp+2247]
+; CHECK: ldx [%fp+2231]
 ; CHECK: fstod %f3
 ; CHECK: faddd %f6
 ; CHECK: faddd %f16
-; CHECK: ldx [%fp+2231]
-; CHECK: ldx [%fp+2247]
 define void @mixedarg(i8 %a0,      ; %i0
                       float %a1,   ; %f3
                       i16 %a2,     ; %i2
@@ -133,7 +133,7 @@ define void @mixedarg(i8 %a0,      ; %i0
   ret void
 }
 
-; CHECK: call_mixedarg
+; CHECK-LABEL: call_mixedarg:
 ; CHECK: stx %i2, [%sp+2247]
 ; CHECK: stx %i0, [%sp+2223]
 ; CHECK: fmovd %f2, %f6
@@ -157,7 +157,7 @@ define void @call_mixedarg(i64 %i0, double %f2, i16* %i2) {
 
 ; The inreg attribute is used to indicate 32-bit sized struct elements that
 ; share an 8-byte slot.
-; CHECK: inreg_fi
+; CHECK-LABEL: inreg_fi:
 ; CHECK: fstoi %f1
 ; CHECK: srlx %i0, 32, [[R:%[gilo][0-7]]]
 ; CHECK: sub [[R]],
@@ -168,7 +168,7 @@ define i32 @inreg_fi(i32 inreg %a0,     ; high bits of %i0
   ret i32 %rv
 }
 
-; CHECK: call_inreg_fi
+; CHECK-LABEL: call_inreg_fi:
 ; Allocate space for 6 arguments, even when only 2 are used.
 ; CHECK: save %sp, -176, %sp
 ; CHECK: sllx %i1, 32, %o0
@@ -179,7 +179,7 @@ define void @call_inreg_fi(i32* %p, i32 %i1, float %f5) {
   ret void
 }
 
-; CHECK: inreg_ff
+; CHECK-LABEL: inreg_ff:
 ; CHECK: fsubs %f0, %f1, %f0
 define float @inreg_ff(float inreg %a0,   ; %f0
                        float inreg %a1) { ; %f1
@@ -187,7 +187,7 @@ define float @inreg_ff(float inreg %a0,   ; %f0
   ret float %rv
 }
 
-; CHECK: call_inreg_ff
+; CHECK-LABEL: call_inreg_ff:
 ; CHECK: fmovs %f3, %f0
 ; CHECK: fmovs %f5, %f1
 ; CHECK: call inreg_ff
@@ -196,7 +196,7 @@ define void @call_inreg_ff(i32* %p, float %f3, float %f5) {
   ret void
 }
 
-; CHECK: inreg_if
+; CHECK-LABEL: inreg_if:
 ; CHECK: fstoi %f0
 ; CHECK: sub %i0
 define i32 @inreg_if(float inreg %a0, ; %f0
@@ -206,7 +206,7 @@ define i32 @inreg_if(float inreg %a0, ; %f0
   ret i32 %rv
 }
 
-; CHECK: call_inreg_if
+; CHECK-LABEL: call_inreg_if:
 ; CHECK: fmovs %f3, %f0
 ; CHECK: mov %i2, %o0
 ; CHECK: call inreg_if
@@ -216,7 +216,7 @@ define void @call_inreg_if(i32* %p, float %f3, i32 %i2) {
 }
 
 ; The frontend shouldn't do this. Just pass i64 instead.
-; CHECK: inreg_ii
+; CHECK-LABEL: inreg_ii:
 ; CHECK: srlx %i0, 32, [[R:%[gilo][0-7]]]
 ; CHECK: sub %i0, [[R]], %i0
 define i32 @inreg_ii(i32 inreg %a0,   ; high bits of %i0
@@ -225,7 +225,7 @@ define i32 @inreg_ii(i32 inreg %a0,   ; high bits of %i0
   ret i32 %rv
 }
 
-; CHECK: call_inreg_ii
+; CHECK-LABEL: call_inreg_ii:
 ; CHECK: srl %i2, 0, [[R2:%[gilo][0-7]]]
 ; CHECK: sllx %i1, 32, [[R1:%[gilo][0-7]]]
 ; CHECK: or [[R1]], [[R2]], %o0
@@ -236,7 +236,7 @@ define void @call_inreg_ii(i32* %p, i32 %i1, i32 %i2) {
 }
 
 ; Structs up to 32 bytes in size can be returned in registers.
-; CHECK: ret_i64_pair
+; CHECK-LABEL: ret_i64_pair:
 ; CHECK: ldx [%i2], %i0
 ; CHECK: ldx [%i3], %i1
 define { i64, i64 } @ret_i64_pair(i32 %a0, i32 %a1, i64* %p, i64* %q) {
@@ -248,7 +248,7 @@ define { i64, i64 } @ret_i64_pair(i32 %a0, i32 %a1, i64* %p, i64* %q) {
   ret { i64, i64 } %rv2
 }
 
-; CHECK: call_ret_i64_pair
+; CHECK-LABEL: call_ret_i64_pair:
 ; CHECK: call ret_i64_pair
 ; CHECK: stx %o0, [%i0]
 ; CHECK: stx %o1, [%i0]
@@ -263,7 +263,7 @@ define void @call_ret_i64_pair(i64* %i0) {
 }
 
 ; This is not a C struct, the i32 member uses 8 bytes, but the float only 4.
-; CHECK: ret_i32_float_pair
+; CHECK-LABEL: ret_i32_float_pair:
 ; CHECK: ld [%i2], %i0
 ; CHECK: ld [%i3], %f2
 define { i32, float } @ret_i32_float_pair(i32 %a0, i32 %a1,
@@ -276,7 +276,7 @@ define { i32, float } @ret_i32_float_pair(i32 %a0, i32 %a1,
   ret { i32, float } %rv2
 }
 
-; CHECK: call_ret_i32_float_pair
+; CHECK-LABEL: call_ret_i32_float_pair:
 ; CHECK: call ret_i32_float_pair
 ; CHECK: st %o0, [%i0]
 ; CHECK: st %f2, [%i1]
@@ -291,10 +291,10 @@ define void @call_ret_i32_float_pair(i32* %i0, float* %i1) {
 }
 
 ; This is a C struct, each member uses 4 bytes.
-; CHECK: ret_i32_float_packed
+; CHECK-LABEL: ret_i32_float_packed:
 ; CHECK: ld [%i2], [[R:%[gilo][0-7]]]
-; CHECK: sllx [[R]], 32, %i0
 ; CHECK: ld [%i3], %f1
+; CHECK: sllx [[R]], 32, %i0
 define inreg { i32, float } @ret_i32_float_packed(i32 %a0, i32 %a1,
                                                   i32* %p, float* %q) {
   %r1 = load i32, i32* %p
@@ -305,7 +305,7 @@ define inreg { i32, float } @ret_i32_float_packed(i32 %a0, i32 %a1,
   ret { i32, float } %rv2
 }
 
-; CHECK: call_ret_i32_float_packed
+; CHECK-LABEL: call_ret_i32_float_packed:
 ; CHECK: call ret_i32_float_packed
 ; CHECK: srlx %o0, 32, [[R:%[gilo][0-7]]]
 ; CHECK: st [[R]], [%i0]
@@ -322,7 +322,7 @@ define void @call_ret_i32_float_packed(i32* %i0, float* %i1) {
 
 ; The C frontend should use i64 to return { i32, i32 } structs, but verify that
 ; we don't miscompile thi case where both struct elements are placed in %i0.
-; CHECK: ret_i32_packed
+; CHECK-LABEL: ret_i32_packed:
 ; CHECK: ld [%i2], [[R1:%[gilo][0-7]]]
 ; CHECK: ld [%i3], [[R2:%[gilo][0-7]]]
 ; CHECK: sllx [[R2]], 32, [[R3:%[gilo][0-7]]]
@@ -337,7 +337,7 @@ define inreg { i32, i32 } @ret_i32_packed(i32 %a0, i32 %a1,
   ret { i32, i32 } %rv2
 }
 
-; CHECK: call_ret_i32_packed
+; CHECK-LABEL: call_ret_i32_packed:
 ; CHECK: call ret_i32_packed
 ; CHECK: srlx %o0, 32, [[R:%[gilo][0-7]]]
 ; CHECK: st [[R]], [%i0]
@@ -353,31 +353,31 @@ define void @call_ret_i32_packed(i32* %i0, i32* %i1) {
 }
 
 ; The return value must be sign-extended to 64 bits.
-; CHECK: ret_sext
+; CHECK-LABEL: ret_sext:
 ; CHECK: sra %i0, 0, %i0
 define signext i32 @ret_sext(i32 %a0) {
   ret i32 %a0
 }
 
-; CHECK: ret_zext
+; CHECK-LABEL: ret_zext:
 ; CHECK: srl %i0, 0, %i0
 define zeroext i32 @ret_zext(i32 %a0) {
   ret i32 %a0
 }
 
-; CHECK: ret_nosext
+; CHECK-LABEL: ret_nosext:
 ; CHECK-NOT: sra
 define signext i32 @ret_nosext(i32 signext %a0) {
   ret i32 %a0
 }
 
-; CHECK: ret_nozext
+; CHECK-LABEL: ret_nozext:
 ; CHECK-NOT: srl
 define signext i32 @ret_nozext(i32 signext %a0) {
   ret i32 %a0
 }
 
-; CHECK-LABEL: test_register_directive
+; CHECK-LABEL: test_register_directive:
 ; CHECK:       .register %g2, #scratch
 ; CHECK:       .register %g3, #scratch
 ; CHECK:       add %i0, 2, %g2
@@ -391,7 +391,7 @@ entry:
   ret i32 %2
 }
 
-; CHECK-LABEL: test_large_stack
+; CHECK-LABEL: test_large_stack:
 
 ; CHECK:       sethi 16, %g1
 ; CHECK:       xor %g1, -176, %g1
@@ -412,7 +412,7 @@ entry:
 
 declare i32 @use_buf(i32, i8*)
 
-; CHECK-LABEL: test_fp128_args
+; CHECK-LABEL: test_fp128_args:
 ; CHECK-DAG:   std %f0, [%fp+{{.+}}]
 ; CHECK-DAG:   std %f2, [%fp+{{.+}}]
 ; CHECK-DAG:   std %f6, [%fp+{{.+}}]
@@ -428,7 +428,7 @@ entry:
 
 declare i64 @receive_fp128(i64 %a, ...)
 
-; CHECK-LABEL: test_fp128_variable_args
+; CHECK-LABEL: test_fp128_variable_args:
 ; CHECK-DAG:   std %f4, [%sp+[[Offset0:[0-9]+]]]
 ; CHECK-DAG:   std %f6, [%sp+[[Offset1:[0-9]+]]]
 ; CHECK-DAG:   ldx [%sp+[[Offset0]]], %o2
@@ -440,7 +440,7 @@ entry:
   ret i64 %0
 }
 
-; CHECK-LABEL: test_call_libfunc
+; CHECK-LABEL: test_call_libfunc:
 ; CHECK:       st %f1, [%fp+[[Offset0:[0-9]+]]]
 ; CHECK:       fmovs %f3, %f1
 ; CHECK:       call cosf

diff --git a/test/CodeGen/SPARC/basictest.ll b/test/CodeGen/SPARC/basictest.ll
index 3f5dfda9ad19dbae9f5704f09417379cac6099f3..889f5144413f2cf83d4837504c6fe1bf6891137d 100644 (file)
--- a/test/CodeGen/SPARC/basictest.ll
+++ b/test/CodeGen/SPARC/basictest.ll
@@ -71,12 +71,12 @@ define i64 @signed_multiply_32x32_64(i32 %a, i32 %b) {
 }
 
 ; CHECK-LABEL: unsigned_multiply_32x32_64:
-; CHECK: umul %o0, %o1, %o2
-; CHECK: rd %y, %o2
 ;FIXME: the smul in the output is totally redundant and should not there.
-; CHECK: smul %o0, %o1, %o1
+; CHECK: smul %o0, %o1, %o2
+; CHECK: umul %o0, %o1, %o0
+; CHECK: rd %y, %o0
 ; CHECK: retl
-; CHECK: mov      %o2, %o0
+; CHECK: mov      %o2, %o1
 define i64 @unsigned_multiply_32x32_64(i32 %a, i32 %b) {
   %xa = zext i32 %a to i64
   %xb = zext i32 %b to i64

diff --git a/test/CodeGen/SPARC/float.ll b/test/CodeGen/SPARC/float.ll
index d7a79cb05a82c51e5ef5ce885eafdb64f73502a0..c4cc04420ad7ac399e9fe926bf2d08ba7ea5ae9c 100644 (file)
--- a/test/CodeGen/SPARC/float.ll
+++ b/test/CodeGen/SPARC/float.ll
@@ -53,20 +53,18 @@ declare double @get_double()
 declare double @llvm.fabs.f64(double) nounwind readonly
 
 ; V8-LABEL:    test_v9_floatreg:
-; V8:          fsubd {{.+}}, {{.+}}, {{.+}}
-; V8:          faddd {{.+}}, {{.+}}, [[R:%f(((1|2)?(0|2|4|6|8))|30)]]
+; V8:          fsubd {{.+}}, {{.+}}, [[R:%f(((1|2)?(0|2|4|6|8))|30)]]
 ; V8:          std [[R]], [%{{.+}}]
 ; V8:          ldd [%{{.+}}], %f0
+; V8:          faddd {{.+}}, {{.+}}, {{.+}}
 
 ; V9-LABEL:    test_v9_floatreg:
 ; V9:          fsubd {{.+}}, {{.+}}, {{.+}}
-; V9:          faddd {{.+}}, {{.+}}, [[R:%f((3(2|4|6|8))|((4|5)(0|2|4|6|8))|(60|62))]]
-; V9:          fmovd [[R]], %f0
+; V9:          faddd {{.+}}, {{.+}}, %f0
 
 ; SPARC64-LABEL:    test_v9_floatreg:
 ; SPARC64:          fsubd {{.+}}, {{.+}}, {{.+}}
-; SPARC64:          faddd {{.+}}, {{.+}}, [[R:%f((3(2|4|6|8))|((4|5)(0|2|4|6|8))|(60|62))]]
-; SPARC64:          fmovd [[R]], %f0
+; SPARC64:          faddd {{.+}}, {{.+}}, %f0
 
 define double @test_v9_floatreg() {
 entry:

diff --git a/test/CodeGen/SPARC/tls.ll b/test/CodeGen/SPARC/tls.ll
index a70637b283f51b2d85fe1247d178cb78b30adb07..8ebd36833ba129ec3a635bf384b294c9a2b61766 100644 (file)
--- a/test/CodeGen/SPARC/tls.ll
+++ b/test/CodeGen/SPARC/tls.ll
@@ -103,10 +103,10 @@ entry:
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_PC22 _GLOBAL_OFFSET_TABLE_ 0x4
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_PC10 _GLOBAL_OFFSET_TABLE_ 0x8
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDO_HIX22 local_symbol 0x0
-; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDO_LOX10 local_symbol 0x0
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDM_HI22 local_symbol 0x0
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDM_LO10 local_symbol 0x0
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDM_ADD local_symbol 0x0
+; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDO_LOX10 local_symbol 0x0
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDM_CALL local_symbol 0x0
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_TLS_LDO_ADD local_symbol 0x0
 ; pic-obj:    0x{{[0-9,A-F]+}} R_SPARC_PC22 _GLOBAL_OFFSET_TABLE_ 0x4

diff --git a/test/CodeGen/SPARC/varargs.ll b/test/CodeGen/SPARC/varargs.ll
index c2d1e98b698be1ef41ce498510abd54f5f4ff284..576acc284fb95c85180e763c250023026d9372af 100644 (file)
--- a/test/CodeGen/SPARC/varargs.ll
+++ b/test/CodeGen/SPARC/varargs.ll
@@ -67,8 +67,8 @@ declare void @llvm.va_start(i8*)
 ; CHECK: call_1d
 ; The fixed-arg double goes in %d2, the second goes in %o2.
 ; CHECK: sethi 1048576
-; CHECK: , %o2
 ; CHECK: , %f2
+; CHECK: , %o2
 define i32 @call_1d() #0 {
 entry:
   %call = call double (i8*, double, ...) @varargsfunc(i8* undef, double 1.000000e+00, double 2.000000e+00)