[X86][X87] Mark pseudo memory fold instructions as load/sideeffects (PR21160, PR34080...

Match regular x87 memory fold instructions with load/sideeffects tags, to prevent the schedulers from re-ordering them across the fnstcw/fldcw sequences for truncating stores while they are still pseudo during the stack conversion pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@321424 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86InstrFPStack.td b/lib/Target/X86/X86InstrFPStack.td
index 7e89a4111d86f1b6abbcb86df23ede64ef44c3b1..619b399ef8d8c3ad3a13cbed038809c7e0cf8e11 100644 (file)
--- a/lib/Target/X86/X86InstrFPStack.td
+++ b/lib/Target/X86/X86InstrFPStack.td
@@ -141,6 +141,7 @@ def _Fp80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, RFP80:$src2), TwoArgFP,
 // These instructions cannot address 80-bit memory.
 multiclass FPBinary<SDNode OpNode, Format fp, string asmstring,
                     bit Forward = 1> {
+let mayLoad = 1, hasSideEffects = 1 in {
 // ST(0) = ST(0) + [mem]
 def _Fp32m  : FpIf32<(outs RFP32:$dst),
                      (ins RFP32:$src1, f32mem:$src2), OneArgFPRW,
@@ -177,10 +178,8 @@ def _Fp80m64: FpI_<(outs RFP80:$dst),
                         (OpNode RFP80:$src1, (f80 (extloadf64 addr:$src2)))),
                        (set RFP80:$dst,
                         (OpNode (f80 (extloadf64 addr:$src2)), RFP80:$src1)))]>;
-let mayLoad = 1 in
 def _F32m  : FPI<0xD8, fp, (outs), (ins f32mem:$src),
                  !strconcat("f", asmstring, "{s}\t$src")>;
-let mayLoad = 1 in
 def _F64m  : FPI<0xDC, fp, (outs), (ins f64mem:$src),
                  !strconcat("f", asmstring, "{l}\t$src")>;
 // ST(0) = ST(0) + [memint]
@@ -226,12 +225,11 @@ def _FpI32m80 : FpI_<(outs RFP80:$dst), (ins RFP80:$src1, i32mem:$src2),
                            (OpNode RFP80:$src1, (X86fild addr:$src2, i32))),
                           (set RFP80:$dst,
                            (OpNode (X86fild addr:$src2, i32), RFP80:$src1)))]>;
-let mayLoad = 1 in
 def _FI16m  : FPI<0xDE, fp, (outs), (ins i16mem:$src),
                   !strconcat("fi", asmstring, "{s}\t$src")>;
-let mayLoad = 1 in
 def _FI32m  : FPI<0xDA, fp, (outs), (ins i32mem:$src),
                   !strconcat("fi", asmstring, "{l}\t$src")>;
+} // mayLoad = 1, hasSideEffects = 1
 }
 
 let Defs = [FPSW] in {

diff --git a/test/CodeGen/X86/pr34080-2.ll b/test/CodeGen/X86/pr34080-2.ll
index 290e076be09ee1155274788ba71f9b52ea6384e6..5c00f0e3706b22f5dd624b5ff373cd64f2aaab93 100644 (file)
--- a/test/CodeGen/X86/pr34080-2.ll
+++ b/test/CodeGen/X86/pr34080-2.ll
@@ -62,6 +62,7 @@ define void @computeJD(%struct.DateTime*) nounwind {
 ; CHECK-NEXT:    imull $60000, 24(%ebx), %ecx # imm = 0xEA60
 ; CHECK-NEXT:    addl %eax, %ecx
 ; CHECK-NEXT:    fldl 28(%ebx)
+; CHECK-NEXT:    fmuls {{\.LCPI.*}}
 ; CHECK-NEXT:    fnstcw (%esp)
 ; CHECK-NEXT:    movzwl (%esp), %eax
 ; CHECK-NEXT:    movw $3199, (%esp) # imm = 0xC7F
@@ -69,7 +70,6 @@ define void @computeJD(%struct.DateTime*) nounwind {
 ; CHECK-NEXT:    movw %ax, (%esp)
 ; CHECK-NEXT:    movl %ecx, %eax
 ; CHECK-NEXT:    sarl $31, %eax
-; CHECK-NEXT:    fmuls {{\.LCPI.*}}
 ; CHECK-NEXT:    fistpll {{[0-9]+}}(%esp)
 ; CHECK-NEXT:    fldcw (%esp)
 ; CHECK-NEXT:    addl {{[0-9]+}}(%esp), %ecx