From: Vasileios Kalintiris Date: Thu, 16 Oct 2014 15:41:51 +0000 (+0000) Subject: [mips] Account for endianess when expanding BuildPairF64/ExtractElementF64 nodes. X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=commitdiff_plain;h=3b72ec5083457366ae4b1c230ed1172f1e428dbe [mips] Account for endianess when expanding BuildPairF64/ExtractElementF64 nodes. Summary: In order to support big endian targets for the BuildPairF64 nodes we just need to swap the low/high pair registers. Additionally, for the ExtractElementF64 nodes we have to calculate the correct stack offset with respect to the node's register/operand that we want to extract. Reviewers: dsanders Reviewed By: dsanders Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D5753 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219931 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/Target/Mips/MipsSEFrameLowering.cpp b/lib/Target/Mips/MipsSEFrameLowering.cpp index 67f96d16fd0..97d9edfe6a5 100644 --- a/lib/Target/Mips/MipsSEFrameLowering.cpp +++ b/lib/Target/Mips/MipsSEFrameLowering.cpp @@ -325,6 +325,8 @@ bool ExpandPseudo::expandBuildPairF64(MachineBasicBlock &MBB, // We re-use the same spill slot each time so that the stack frame doesn't // grow too much in functions with a large number of moves. int FI = MF.getInfo()->getMoveF64ViaSpillFI(RC2); + if (!Subtarget.isLittle()) + std::swap(LoReg, HiReg); TII.storeRegToStack(MBB, I, LoReg, I->getOperand(1).isKill(), FI, RC, &TRI, 0); TII.storeRegToStack(MBB, I, HiReg, I->getOperand(2).isKill(), FI, RC, &TRI, @@ -369,6 +371,7 @@ bool ExpandPseudo::expandExtractElementF64(MachineBasicBlock &MBB, unsigned DstReg = I->getOperand(0).getReg(); unsigned SrcReg = I->getOperand(1).getReg(); unsigned N = I->getOperand(2).getImm(); + int64_t Offset = 4 * (Subtarget.isLittle() ? N : (1 - N)); // It should be impossible to have FGR64 on MIPS-II or MIPS32r1 (which are // the cases where mfhc1 is not available). 64-bit architectures and @@ -385,7 +388,7 @@ bool ExpandPseudo::expandExtractElementF64(MachineBasicBlock &MBB, int FI = MF.getInfo()->getMoveF64ViaSpillFI(RC); TII.storeRegToStack(MBB, I, SrcReg, I->getOperand(1).isKill(), FI, RC, &TRI, 0); - TII.loadRegFromStack(MBB, I, DstReg, FI, RC2, &TRI, N * 4); + TII.loadRegFromStack(MBB, I, DstReg, FI, RC2, &TRI, Offset); return true; } diff --git a/test/CodeGen/Mips/fp64a.ll b/test/CodeGen/Mips/fp64a.ll index 5c2c87373a3..fadce5cb748 100644 --- a/test/CodeGen/Mips/fp64a.ll +++ b/test/CodeGen/Mips/fp64a.ll @@ -12,9 +12,9 @@ ; this check here. ; RUN: llc -march=mips -mcpu=mips32r2 -mattr=fp64 < %s | FileCheck %s -check-prefix=ALL -check-prefix=32R2-NO-FP64A-BE -; RUN: llc -march=mips -mcpu=mips32r2 -mattr=fp64,nooddspreg < %s | FileCheck %s -check-prefix=ALL -check-prefix=32R2-FP64A-BE +; RUN: llc -march=mips -mcpu=mips32r2 -mattr=fp64,nooddspreg < %s | FileCheck %s -check-prefix=ALL -check-prefix=32R2-FP64A ; RUN: llc -march=mipsel -mcpu=mips32r2 -mattr=fp64 < %s | FileCheck %s -check-prefix=ALL -check-prefix=32R2-NO-FP64A-LE -; RUN: llc -march=mipsel -mcpu=mips32r2 -mattr=fp64,nooddspreg < %s | FileCheck %s -check-prefix=ALL -check-prefix=32R2-FP64A-LE +; RUN: llc -march=mipsel -mcpu=mips32r2 -mattr=fp64,nooddspreg < %s | FileCheck %s -check-prefix=ALL -check-prefix=32R2-FP64A ; RUN: llc -march=mips64 -mcpu=mips64 -mattr=fp64 < %s | FileCheck %s -check-prefix=ALL -check-prefix=64-NO-FP64A ; RUN: not llc -march=mips64 -mcpu=mips64 -mattr=fp64,nooddspreg < %s 2>&1 | FileCheck %s -check-prefix=64-FP64A @@ -38,15 +38,10 @@ define double @call1(double %d, ...) { ; 32R2-NO-FP64A-BE: mtc1 $5, $f0 ; 32R2-NO-FP64A-BE: mthc1 $4, $f0 -; 32R2-FP64A-LE: addiu $sp, $sp, -8 -; 32R2-FP64A-LE: sw $4, 0($sp) -; 32R2-FP64A-LE: sw $5, 4($sp) -; 32R2-FP64A-LE: ldc1 $f0, 0($sp) - -; 32R2-FP64A-BE: addiu $sp, $sp, -8 -; 32R2-FP64A-BE: sw $5, 0($sp) -; 32R2-FP64A-BE: sw $4, 4($sp) -; 32R2-FP64A-BE: ldc1 $f0, 0($sp) +; 32R2-FP64A: addiu $sp, $sp, -8 +; 32R2-FP64A: sw $4, 0($sp) +; 32R2-FP64A: sw $5, 4($sp) +; 32R2-FP64A: ldc1 $f0, 0($sp) ; 64-NO-FP64A: daddiu $sp, $sp, -64 ; 64-NO-FP64A: mov.d $f0, $f12 @@ -63,15 +58,10 @@ define double @call2(i32 %i, double %d) { ; 32R2-NO-FP64A-BE: mtc1 $7, $f0 ; 32R2-NO-FP64A-BE: mthc1 $6, $f0 -; 32R2-FP64A-LE: addiu $sp, $sp, -8 -; 32R2-FP64A-LE: sw $6, 0($sp) -; 32R2-FP64A-LE: sw $7, 4($sp) -; 32R2-FP64A-LE: ldc1 $f0, 0($sp) - -; 32R2-FP64A-BE: addiu $sp, $sp, -8 -; 32R2-FP64A-BE: sw $7, 0($sp) -; 32R2-FP64A-BE: sw $6, 4($sp) -; 32R2-FP64A-BE: ldc1 $f0, 0($sp) +; 32R2-FP64A: addiu $sp, $sp, -8 +; 32R2-FP64A: sw $6, 0($sp) +; 32R2-FP64A: sw $7, 4($sp) +; 32R2-FP64A: ldc1 $f0, 0($sp) ; 64-NO-FP64A-NOT: daddiu $sp, $sp ; 64-NO-FP64A: mov.d $f0, $f13 @@ -88,15 +78,10 @@ define double @call3(float %f1, float %f2, double %d) { ; 32R2-NO-FP64A-BE: mtc1 $7, $f0 ; 32R2-NO-FP64A-BE: mthc1 $6, $f0 -; 32R2-FP64A-LE: addiu $sp, $sp, -8 -; 32R2-FP64A-LE: sw $6, 0($sp) -; 32R2-FP64A-LE: sw $7, 4($sp) -; 32R2-FP64A-LE: ldc1 $f0, 0($sp) - -; 32R2-FP64A-BE: addiu $sp, $sp, -8 -; 32R2-FP64A-BE: sw $7, 0($sp) -; 32R2-FP64A-BE: sw $6, 4($sp) -; 32R2-FP64A-BE: ldc1 $f0, 0($sp) +; 32R2-FP64A: addiu $sp, $sp, -8 +; 32R2-FP64A: sw $6, 0($sp) +; 32R2-FP64A: sw $7, 4($sp) +; 32R2-FP64A: ldc1 $f0, 0($sp) ; 64-NO-FP64A-NOT: daddiu $sp, $sp ; 64-NO-FP64A: mov.d $f0, $f14 @@ -113,15 +98,10 @@ define double @call4(float %f, double %d, ...) { ; 32R2-NO-FP64A-BE: mtc1 $7, $f0 ; 32R2-NO-FP64A-BE: mthc1 $6, $f0 -; 32R2-FP64A-LE: addiu $sp, $sp, -8 -; 32R2-FP64A-LE: sw $6, 0($sp) -; 32R2-FP64A-LE: sw $7, 4($sp) -; 32R2-FP64A-LE: ldc1 $f0, 0($sp) - -; 32R2-FP64A-BE: addiu $sp, $sp, -8 -; 32R2-FP64A-BE: sw $7, 0($sp) -; 32R2-FP64A-BE: sw $6, 4($sp) -; 32R2-FP64A-BE: ldc1 $f0, 0($sp) +; 32R2-FP64A: addiu $sp, $sp, -8 +; 32R2-FP64A: sw $6, 0($sp) +; 32R2-FP64A: sw $7, 4($sp) +; 32R2-FP64A: ldc1 $f0, 0($sp) ; 64-NO-FP64A: daddiu $sp, $sp, -48 ; 64-NO-FP64A: mov.d $f0, $f13 @@ -145,23 +125,14 @@ define double @call5(double %a, double %b, ...) { ; 32R2-NO-FP64A-BE-DAG: mthc1 $6, $[[T1:f[0-9]+]] ; 32R2-NO-FP64A-BE: sub.d $f0, $[[T0]], $[[T1]] -; 32R2-FP64A-LE: addiu $sp, $sp, -8 -; 32R2-FP64A-LE: sw $6, 0($sp) -; 32R2-FP64A-LE: sw $7, 4($sp) -; 32R2-FP64A-LE: ldc1 $[[T1:f[0-9]+]], 0($sp) -; 32R2-FP64A-LE: sw $4, 0($sp) -; 32R2-FP64A-LE: sw $5, 4($sp) -; 32R2-FP64A-LE: ldc1 $[[T0:f[0-9]+]], 0($sp) -; 32R2-FP64A-LE: sub.d $f0, $[[T0]], $[[T1]] - -; 32R2-FP64A-BE: addiu $sp, $sp, -8 -; 32R2-FP64A-BE: sw $7, 0($sp) -; 32R2-FP64A-BE: sw $6, 4($sp) -; 32R2-FP64A-BE: ldc1 $[[T1:f[0-9]+]], 0($sp) -; 32R2-FP64A-BE: sw $5, 0($sp) -; 32R2-FP64A-BE: sw $4, 4($sp) -; 32R2-FP64A-BE: ldc1 $[[T0:f[0-9]+]], 0($sp) -; 32R2-FP64A-BE: sub.d $f0, $[[T0]], $[[T1]] +; 32R2-FP64A: addiu $sp, $sp, -8 +; 32R2-FP64A: sw $6, 0($sp) +; 32R2-FP64A: sw $7, 4($sp) +; 32R2-FP64A: ldc1 $[[T1:f[0-9]+]], 0($sp) +; 32R2-FP64A: sw $4, 0($sp) +; 32R2-FP64A: sw $5, 4($sp) +; 32R2-FP64A: ldc1 $[[T0:f[0-9]+]], 0($sp) +; 32R2-FP64A: sub.d $f0, $[[T0]], $[[T1]] ; 64-NO-FP64A: sub.d $f0, $f12, $f13 } @@ -179,19 +150,12 @@ define double @move_from(double %d) { ; 32R2-NO-FP64A-BE-DAG: mfc1 $7, $f0 ; 32R2-NO-FP64A-BE-DAG: mfhc1 $6, $f0 -; 32R2-FP64A-LE: addiu $sp, $sp, -32 -; 32R2-FP64A-LE: sdc1 $f0, 16($sp) -; 32R2-FP64A-LE: lw $6, 16($sp) -; FIXME: This store is redundant -; 32R2-FP64A-LE: sdc1 $f0, 16($sp) -; 32R2-FP64A-LE: lw $7, 20($sp) - -; 32R2-FP64A-BE: addiu $sp, $sp, -32 -; 32R2-FP64A-BE: sdc1 $f0, 16($sp) -; 32R2-FP64A-BE: lw $6, 20($sp) +; 32R2-FP64A: addiu $sp, $sp, -32 +; 32R2-FP64A: sdc1 $f0, 16($sp) +; 32R2-FP64A: lw $6, 16($sp) ; FIXME: This store is redundant -; 32R2-FP64A-BE: sdc1 $f0, 16($sp) -; 32R2-FP64A-BE: lw $7, 16($sp) +; 32R2-FP64A: sdc1 $f0, 16($sp) +; 32R2-FP64A: lw $7, 20($sp) ; 64-NO-FP64A: mov.d $f13, $f0 }