| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <012701cbab3c$46d283e0$d4778ba0$@mprc.pku.edu.cn>
Date: Mon, 3 Jan 2011 19:49:28 +0800
From: "Guan Xuetao" <guanxuetao@...c.pku.edu.cn>
To: <linux-arch@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: [PATCHv1 2/8] unicore32 additional architecture files: float point unit
From: Guan Xuetao <guanxuetao@...c.pku.edu.cn>
Patch 2 implements support for float point unit, which using UniCore-F64 FPU hardware
in UniCore32 ISA.
Signed-off-by: Guan Xuetao <guanxuetao@...c.pku.edu.cn>
---
arch/unicore32/include/asm/fpstate.h | 41 ++
arch/unicore32/include/asm/uc-f64.h | 60 +++
arch/unicore32/uc-f64/Makefile | 13 +
arch/unicore32/uc-f64/entry.S | 33 ++
arch/unicore32/uc-f64/f64_double_cmp.h | 245 ++++++++++
arch/unicore32/uc-f64/f64_single_cmp.h | 245 ++++++++++
arch/unicore32/uc-f64/f64double.c | 758 +++++++++++++++++++++++++++++++
arch/unicore32/uc-f64/f64hw.S | 155 +++++++
arch/unicore32/uc-f64/f64instr.h | 101 +++++
arch/unicore32/uc-f64/f64module.c | 180 ++++++++
arch/unicore32/uc-f64/f64single.c | 771 ++++++++++++++++++++++++++++++++
arch/unicore32/uc-f64/f64sint.c | 94 ++++
arch/unicore32/uc-f64/uc-f64.h | 332 ++++++++++++++
13 files changed, 3028 insertions(+), 0 deletions(-)
diff --git a/arch/unicore32/include/asm/fpstate.h b/arch/unicore32/include/asm/fpstate.h
new file mode 100644
index 0000000..818042a
--- /dev/null
+++ b/arch/unicore32/include/asm/fpstate.h
@@ -0,0 +1,41 @@
+/*
+ * linux/arch/unicore32/include/asm/fpstate.h
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef __UNICORE_FPSTATE_H__
+#define __UNICORE_FPSTATE_H__
+
+
+#ifndef __ASSEMBLY__
+
+
+#define FP_HARD_SIZE 33
+
+struct fp_hard_struct {
+ unsigned int save[FP_HARD_SIZE]; /* as yet undefined */
+};
+
+#define FP_SOFT_SIZE 33
+
+struct fp_soft_struct {
+ unsigned int save[FP_SOFT_SIZE]; /* undefined information */
+};
+
+union fp_state {
+ struct fp_hard_struct hard;
+ struct fp_soft_struct soft;
+} __attribute__((aligned(8)));
+
+#define FP_SIZE (sizeof(union fp_state) / sizeof(int))
+
+#endif
+
+#endif
diff --git a/arch/unicore32/include/asm/uc-f64.h b/arch/unicore32/include/asm/uc-f64.h
new file mode 100644
index 0000000..1a7faf1
--- /dev/null
+++ b/arch/unicore32/include/asm/uc-f64.h
@@ -0,0 +1,60 @@
+/*
+ * linux/arch/unicore32/include/asm/uc-f64.h
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ *
+ * Maintained by GUAN Xue-tao <gxt@...c.pku.edu.cn>
+ * Copyright (C) 2001-2010 Guan Xuetao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Based on:
+ *
+ * Contributors & Additions/Fixes:
+ *
+ * TODO:
+ */
+#define FPSCR s31
+
+
+/* FPSCR bits */
+#define FPSCR_DEFAULT_NAN (1<<25)
+
+#define FPSCR_CMPINSTR_BIT (1<<31)
+
+#define FPSCR_CON (1<<29)
+#define FPSCR_TRAP (1<<27)
+
+/* RND mode */
+#define FPSCR_ROUND_NEAREST (0<<0)
+#define FPSCR_ROUND_PLUSINF (2<<0)
+#define FPSCR_ROUND_MINUSINF (3<<0)
+#define FPSCR_ROUND_TOZERO (1<<0)
+#define FPSCR_RMODE_BIT (0)
+#define FPSCR_RMODE_MASK (7 << FPSCR_RMODE_BIT)
+
+/* trap enable */
+#define FPSCR_IOE (1<<16)
+#define FPSCR_OFE (1<<14)
+#define FPSCR_UFE (1<<13)
+#define FPSCR_IXE (1<<12)
+#define FPSCR_HIE (1<<11)
+#define FPSCR_NDE (1<<10) /* non denomal */
+
+/* flags */
+#define FPSCR_IDC (1<<24)
+#define FPSCR_HIC (1<<23)
+#define FPSCR_IXC (1<<22)
+#define FPSCR_OFC (1<<21)
+#define FPSCR_UFC (1<<20)
+#define FPSCR_IOC (1<<19)
+
+/* stick bits */
+#define FPSCR_IOS (1<<9)
+#define FPSCR_OFS (1<<7)
+#define FPSCR_UFS (1<<6)
+#define FPSCR_IXS (1<<5)
+#define FPSCR_HIS (1<<4)
+#define FPSCR_NDS (1<<3) /*non denomal */
diff --git a/arch/unicore32/uc-f64/Makefile b/arch/unicore32/uc-f64/Makefile
new file mode 100644
index 0000000..ac33d55
--- /dev/null
+++ b/arch/unicore32/uc-f64/Makefile
@@ -0,0 +1,13 @@
+#
+# linux/arch/unicore32/uc-f64/Makefile
+#
+
+# EXTRA_CFLAGS := -DDEBUG
+# EXTRA_AFLAGS := -DDEBUG
+
+#AFLAGS :=$(AFLAGS:-msoft-float=-Wa,-mfpu=softfpu)
+#LDFLAGS +=--no-warn-mismatch
+
+obj-y += uc-f64.o
+
+uc-f64-$(CONFIG_UNICORE_FPU_F64) += f64module.o entry.o f64hw.o f64single.o f64double.o f64sint.o
diff --git a/arch/unicore32/uc-f64/entry.S b/arch/unicore32/uc-f64/entry.S
new file mode 100644
index 0000000..4991ca6
--- /dev/null
+++ b/arch/unicore32/uc-f64/entry.S
@@ -0,0 +1,33 @@
+/*
+ * linux/arch/unicore32/uc-f64/entry.S
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Basic entry code, called from the kernel's data abort trap.
+ * r0 = faulted instruction
+ * r19 = successful return
+ * r20 = thread_info structure
+ * lr = failure return
+ */
+#include <linux/linkage.h>
+#include <asm/assembler.h>
+#include <asm/thread_info.h>
+#include <generated/asm-offsets.h>
+
+ENTRY(do_uc_f64)
+ enable_irq r4
+ ldw r4, .LC_f64
+ add r20, r20, #TI_FPSTATE @ r20 = workspace
+ ldw pc, [r4] @ call FP entry point
+ENDPROC(do_uc_f64)
+
+ .align 2
+.LC_f64:
+ .word f64_vector
+
diff --git a/arch/unicore32/uc-f64/f64_double_cmp.h b/arch/unicore32/uc-f64/f64_double_cmp.h
new file mode 100644
index 0000000..ceda0d6
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64_double_cmp.h
@@ -0,0 +1,245 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64_double_cmp.h
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Compare operations for double format
+ */
+
+static u32 f64_double_fcf(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ return exceptions;
+}
+static u32 f64_double_fcun(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FUOD)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fceq(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FEQ)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcueq(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcolt(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FLT)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcult(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FLT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcole(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FLT) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcule(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 0, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) != FGT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+
+static u32 f64_double_fcsf(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ return exceptions;
+}
+static u32 f64_double_fcngle(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FUOD)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcseq(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FEQ)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcngl(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fclt(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FLT)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcnge(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FLT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcle(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FLT) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_double_fcngt(int unused, int dd, int dm, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(dd, 1, dm, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) != FGT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static struct op f_cmp_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FCF)] = { f64_double_fcf, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCUN)] = { f64_double_fcun, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCEQ)] = { f64_double_fceq, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCUEQ)] = { f64_double_fcueq, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCOLT)] = { f64_double_fcolt, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCULT)] = { f64_double_fcult, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCOLE)] = { f64_double_fcole, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCUL)] = { f64_double_fcule, OP_DD },
+
+ [FFUNC_TO_IDX(FOP_FCSF)] = { f64_double_fcsf, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCNGLE)] = { f64_double_fcngle, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCSEQ)] = { f64_double_fcseq, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCNGL)] = { f64_double_fcngl, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCLT)] = { f64_double_fclt, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCNGE)] = { f64_double_fcnge, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCLE)] = { f64_double_fcle, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCNGT)] = { f64_double_fcngt, OP_DD },
+};
+
+u32 f64_double_mffcdo(u32 inst, u32 ff, struct pt_regs *regs)
+{
+ u32 op1 = (inst & FOP1_MASK) >> 26;
+ u32 func = (inst & FFUNC_MASK) >> 6;
+ u32 exceptions = 0;
+ u32 except;
+ s32 m;
+ char type;
+ unsigned int dest;
+ unsigned int dn = f64_get_dn(inst);
+ unsigned int dm = f64_get_dm(inst);
+ struct op *fop;
+
+ fop = &f_cmp_ops[func];
+
+ if (!fop->fn)
+ goto invalid;
+
+ dest = f64_get_rd(inst);
+ m = f64_get_double(dm);
+
+ type = 'w';
+ pr_debug("UniCore-F64: (%c%u) = (s%u) op1[%u] func[%u] (s%u=%08x)\n",
+ type, dest, dn,
+ op1, func, dm, m);
+
+ except = fop->fn(dest, dn, m, ff);
+
+ pr_debug("UniCore-F64: exceptions=%08x\n", except);
+ if (dest < 31)
+ regs->uregs[dest] = except & 0x20000000;
+ else {
+ if (except & 0x20000000) {
+ regs->uregs[32] &= 0x0fffffff;
+ regs->uregs[32] |= 0x20000000;
+ } else
+ regs->uregs[32] &= 0x0fffffff;
+ }
+
+ exceptions |= except;
+ return exceptions;
+
+ invalid:
+ return (u32)-1;
+}
diff --git a/arch/unicore32/uc-f64/f64_single_cmp.h b/arch/unicore32/uc-f64/f64_single_cmp.h
new file mode 100644
index 0000000..8797c6d
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64_single_cmp.h
@@ -0,0 +1,245 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64_single_cmp.h
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Compare operations for single format
+ */
+static u32 f64_single_fcf(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ return exceptions;
+}
+static u32 f64_single_fcun(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FUOD)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fceq(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FEQ)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcueq(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcolt(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FLT)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcult(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FLT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcole(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FLT) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcule(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 0, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) != FGT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+
+static u32 f64_single_fcsf(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ return exceptions;
+}
+static u32 f64_single_fcngle(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FUOD)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcseq(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FEQ)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcngl(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fclt(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if (FSTATUS(result) == FLT)
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcnge(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FUOD) || (FSTATUS(result) == FLT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcle(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) == FLT) || (FSTATUS(result) == FEQ))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+static u32 f64_single_fcngt(int unused, int sd, s32 m, u32 fpscr)
+{
+ u32 exceptions = FPSCR_CMPINSTR_BIT;
+ u32 result;
+ result = f64_compare(sd, 1, m, fpscr);
+ if (result & FPSCR_IOC)
+ exceptions |= FPSCR_IOC;
+ if ((FSTATUS(result) != FGT))
+ exceptions |= FPSCR_CON;
+ return exceptions;
+}
+
+static struct op f_cmp_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FCF)] = { f64_single_fcf, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCUN)] = { f64_single_fcun, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCEQ)] = { f64_single_fceq, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCUEQ)] = { f64_single_fcueq, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCOLT)] = { f64_single_fcolt, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCULT)] = { f64_single_fcult, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCOLE)] = { f64_single_fcole, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCUL)] = { f64_single_fcule, OP_SD },
+
+ [FFUNC_TO_IDX(FOP_FCSF)] = { f64_single_fcsf, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCNGLE)] = { f64_single_fcngle, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCSEQ)] = { f64_single_fcseq, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCNGL)] = { f64_single_fcngl, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCLT)] = { f64_single_fclt, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCNGE)] = { f64_single_fcnge, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCLE)] = { f64_single_fcle, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCNGT)] = { f64_single_fcngt, OP_SD },
+};
+
+u32 f64_single_mffcdo(u32 inst, u32 ff, struct pt_regs *regs)
+{
+ u32 op1 = (inst & FOP1_MASK) >> 26;
+ u32 func = (inst & FFUNC_MASK) >> 6;
+ u32 exceptions = 0;
+ u32 except;
+ s32 m;
+ char type;
+ unsigned int dest;
+ unsigned int sn = f64_get_sn(inst);
+ unsigned int sm = f64_get_sm(inst);
+ struct op *fop;
+
+ fop = &f_cmp_ops[func];
+
+ if (!fop->fn)
+ goto invalid;
+
+ dest = f64_get_rd(inst);
+ m = f64_get_float(sm);
+
+ type = 'w';
+ pr_debug("UniCore-F64: (%c%u) = (s%u) op1[%u] func[%u] (s%u=%08x)\n",
+ type, dest, sn,
+ op1, func, sm, m);
+
+ except = fop->fn(dest, sn, m, ff);
+
+ pr_debug("UniCore-F64: exceptions=%08x\n", except);
+ if (dest < 31)
+ regs->uregs[dest] = except & 0x20000000;
+ else {
+ if (except & 0x20000000) {
+ regs->uregs[32] &= 0x0fffffff;
+ regs->uregs[32] |= 0x20000000;
+ } else
+ regs->uregs[32] &= 0x0fffffff;
+ }
+
+ exceptions |= except;
+ return exceptions;
+
+ invalid:
+ return (u32)-1;
+}
diff --git a/arch/unicore32/uc-f64/f64double.c b/arch/unicore32/uc-f64/f64double.c
new file mode 100644
index 0000000..454b5e6
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64double.c
@@ -0,0 +1,758 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64double.c
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This code is derived in part from John R. Housers softfloat library, which
+ * carries the following notice:
+ *
+ * ===========================================================================
+ * This C source file is part of the SoftFloat IEC/IEEE Floating-point
+ * Arithmetic Package, Release 2.
+ *
+ * Written by John R. Hauser. This work was made possible in part by the
+ * International Computer Science Institute, located at Suite 600, 1947 Center
+ * Street, Berkeley, California 94704. Funding was partially provided by the
+ * National Science Foundation under grant MIP-9311980. The original version
+ * of this code was written as part of a project to build a fixed-point vector
+ * processor in collaboration with the University of California at Berkeley,
+ * overseen by Profs. Nelson Morgan and John Wawrzynek. More information
+ * is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
+ * arithmetic/softfloat.html'.
+ *
+ * THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
+ * has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
+ * TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
+ * PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
+ * AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
+ *
+ * Derivative works are acceptable, even for commercial purposes, so long as
+ * (1) they include prominent notice that the work is derivative, and (2) they
+ * include prominent notice akin to these three paragraphs for those parts of
+ * this code that are retained.
+ * ===========================================================================
+ */
+#include <linux/kernel.h>
+#include <linux/bitops.h>
+
+#include <asm/div64.h>
+#include <asm/uc-f64.h>
+
+#include "f64instr.h"
+#include "uc-f64.h"
+
+static struct f64_double f64_double_default_qnan = {
+ .exponent = 2047,
+ .sign = 0,
+ .significand = F64_DOUBLE_SIGNIFICAND_QNAN,
+};
+
+static void f64_double_dump(const char *str, struct f64_double *d)
+{
+ pr_debug("UniCore-F64 %s: sign=%d exponent=%d significand=%016llx\n",
+ str, d->sign != 0, d->exponent, d->significand);
+}
+
+static void f64_double_normalise_denormal(struct f64_double *vd)
+{
+ int bits = 31 - fls(vd->significand >> 32);
+ if (bits == 31)
+ bits = 63 - fls(vd->significand);
+
+ f64_double_dump("normalise_denormal: in", vd);
+
+ if (bits) {
+ vd->exponent -= bits - 1;
+ vd->significand <<= bits;
+ }
+
+ f64_double_dump("normalise_denormal: out", vd);
+}
+
+u32 f64_double_normaliseround(int dd, struct f64_double *vd, u32 fpscr,
+ u32 exceptions, const char *func)
+{
+ u64 significand, incr;
+ int exponent, shift, underflow;
+ u32 rmode;
+
+ f64_double_dump("pack: in", vd);
+
+ /*
+ * Infinities and NaNs are a special case.
+ */
+ if (vd->exponent == 2047 && (vd->significand == 0 || exceptions))
+ goto pack;
+
+ /*
+ * Special-case zero.
+ */
+ if (vd->significand == 0) {
+ vd->exponent = 0;
+ goto pack;
+ }
+
+ exponent = vd->exponent;
+ significand = vd->significand;
+
+ shift = 32 - fls(significand >> 32);
+ if (shift == 32)
+ shift = 64 - fls(significand);
+ if (shift) {
+ exponent -= shift;
+ significand <<= shift;
+ }
+
+#ifdef DEBUG
+ vd->exponent = exponent;
+ vd->significand = significand;
+ f64_double_dump("pack: normalised", vd);
+#endif
+
+ /*
+ * Tiny number?
+ */
+ underflow = exponent < 0;
+ if (underflow) {
+ significand = f64_shiftright64jamming(significand, -exponent);
+ exponent = 0;
+#ifdef DEBUG
+ vd->exponent = exponent;
+ vd->significand = significand;
+ f64_double_dump("pack: tiny number", vd);
+#endif
+ if (!(significand & ((1ULL << (F64_DOUBLE_LOW_BITS + 1)) - 1)))
+ underflow = 0;
+ }
+
+ /*
+ * Select rounding increment.
+ */
+ incr = 0;
+ rmode = fpscr & FPSCR_RMODE_MASK;
+
+ if (rmode == FPSCR_ROUND_NEAREST) {
+ incr = 1ULL << F64_DOUBLE_LOW_BITS;
+ if ((significand & (1ULL << (F64_DOUBLE_LOW_BITS + 1))) == 0)
+ incr -= 1;
+ } else if (rmode == FPSCR_ROUND_TOZERO) {
+ incr = 0;
+ } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vd->sign != 0))
+ incr = (1ULL << (F64_DOUBLE_LOW_BITS + 1)) - 1;
+
+ pr_debug("UniCore-F64 rounding increment = 0x%08llx\n", incr);
+
+ /*
+ * Is our rounding going to overflow?
+ */
+ if ((significand + incr) < significand) {
+ exponent += 1;
+ significand = (significand >> 1) | (significand & 1);
+ incr >>= 1;
+#ifdef DEBUG
+ vd->exponent = exponent;
+ vd->significand = significand;
+ f64_double_dump("pack: overflow", vd);
+#endif
+ }
+
+ /*
+ * If any of the low bits (which will be shifted out of the
+ * number) are non-zero, the result is inexact.
+ */
+ if (significand & ((1 << (F64_DOUBLE_LOW_BITS + 1)) - 1))
+ exceptions |= FPSCR_IXC;
+
+ /*
+ * Do our rounding.
+ */
+ significand += incr;
+
+ /*
+ * Infinity?
+ */
+ if (exponent >= 2046) {
+ exceptions |= FPSCR_OFC | FPSCR_IXC;
+ if (incr == 0) {
+ vd->exponent = 2045;
+ vd->significand = 0x7fffffffffffffffULL;
+ } else {
+ vd->exponent = 2047; /* infinity */
+ vd->significand = 0;
+ }
+ } else {
+ if (significand >> (F64_DOUBLE_LOW_BITS + 1) == 0)
+ exponent = 0;
+ if (exponent || significand > 0x8000000000000000ULL)
+ underflow = 0;
+ if (underflow)
+ exceptions |= FPSCR_UFC;
+ vd->exponent = exponent;
+ vd->significand = significand >> 1;
+ }
+
+ pack:
+ f64_double_dump("pack: final", vd);
+ {
+ s64 d = f64_double_pack(vd);
+ pr_debug("UniCore-F64 %s: d(d%d)=%016llx exceptions=%08x\n",
+ func, dd, d, exceptions);
+ f64_put_double(d, dd);
+ }
+ return exceptions;
+}
+
+/*
+ * Propagate the NaN, setting exceptions if it is signalling.
+ * 'n' is always a NaN. 'm' may be a number, NaN or infinity.
+ */
+static u32
+f64_propagate_nan(struct f64_double *vdd, struct f64_double *vdn,
+ struct f64_double *vdm, u32 fpscr)
+{
+ struct f64_double *nan;
+ int tn, tm = 0;
+
+ tn = f64_double_type(vdn);
+
+ if (vdm)
+ tm = f64_double_type(vdm);
+
+ if (fpscr & FPSCR_DEFAULT_NAN)
+ /*
+ * Default NaN mode - always returns a quiet NaN
+ */
+ nan = &f64_double_default_qnan;
+ else {
+ /*
+ * Contemporary mode - select the first signalling
+ * NAN, or if neither are signalling, the first
+ * quiet NAN.
+ */
+ if (tn == F64_SNAN || (tm != F64_SNAN && tn == F64_QNAN))
+ nan = vdn;
+ else
+ nan = vdm;
+ /*
+ * Make the NaN quiet.
+ */
+ nan->significand |= F64_DOUBLE_SIGNIFICAND_QNAN;
+ }
+
+ *vdd = *nan;
+
+ /*
+ * If one was a signalling NAN, raise invalid operation.
+ */
+ return tn == F64_SNAN || tm == F64_SNAN ? FPSCR_IOC : F64_NAN_FLAG;
+}
+
+/*
+ * Extended operations
+ */
+static u32 f64_double_fabs(int dd, int unused, int dm, u32 fpscr)
+{
+ f64_put_double(f64_double_packed_abs(f64_get_double(dm)), dd);
+ return 0;
+}
+
+static u32 f64_double_fneg(int dd, int unused, int dm, u32 fpscr)
+{
+ f64_put_double(f64_double_packed_negate(f64_get_double(dm)), dd);
+ return 0;
+}
+/*
+ * Equal := ZC
+ * Less than := N
+ * Greater than := C
+ * Unordered := CV
+ */
+static u32 f64_compare(int dd, int signal_on_qnan, int dm, u32 fpscr)
+{
+ s64 d, m;
+ u32 ret = 0;
+
+ m = f64_get_double(dm);
+ if (f64_double_packed_exponent(m) == 2047
+ && f64_double_packed_mantissa(m)) {
+ ret |= FPSCR_C | FPSCR_V;
+ if (signal_on_qnan || !(f64_double_packed_mantissa(m)
+ & (1ULL << (F64_DOUBLE_MANTISSA_BITS - 1))))
+ /*
+ * Signalling NaN, or signalling on quiet NaN
+ */
+ ret |= FPSCR_IOC;
+ }
+
+ d = f64_get_double(dd);
+ if (f64_double_packed_exponent(d) == 2047
+ && f64_double_packed_mantissa(d)) {
+ ret |= FPSCR_C | FPSCR_V;
+ if (signal_on_qnan || !(f64_double_packed_mantissa(d)
+ & (1ULL << (F64_DOUBLE_MANTISSA_BITS - 1))))
+ /*
+ * Signalling NaN, or signalling on quiet NaN
+ */
+ ret |= FPSCR_IOC;
+ }
+
+ if (ret == 0) {
+ if (d == m || f64_double_packed_abs(d | m) == 0) {
+ /*
+ * equal
+ */
+ ret |= FPSCR_Z | FPSCR_C;
+ } else if (f64_double_packed_sign(d ^ m)) {
+ /*
+ * different signs
+ */
+ if (f64_double_packed_sign(d))
+ /*
+ * d is negative, so d < m
+ */
+ ret |= FPSCR_N;
+ else
+ /*
+ * d is positive, so d > m
+ */
+ ret |= FPSCR_C;
+ } else if ((f64_double_packed_sign(d) != 0) ^ (d < m)) {
+ /*
+ * d < m
+ */
+ ret |= FPSCR_N;
+ } else if ((f64_double_packed_sign(d) != 0) ^ (d > m)) {
+ /*
+ * d > m
+ */
+ ret |= FPSCR_C;
+ }
+ }
+
+ return ret;
+}
+
+/*
+ * Conversion operations
+ */
+static u32 f64_double_fcvts(int sd, int unused, int dm, u32 fpscr)
+{
+ struct f64_double vdm;
+ struct f64_single vsd;
+ int tm;
+ u32 exceptions = 0;
+
+ f64_double_unpack(&vdm, f64_get_double(dm));
+
+ tm = f64_double_type(&vdm);
+
+ /*
+ * If we have a signalling NaN, signal invalid operation.
+ */
+ if (tm == F64_SNAN)
+ exceptions = FPSCR_IOC;
+
+ if (tm & F64_DENORMAL)
+ f64_double_normalise_denormal(&vdm);
+
+ vsd.sign = vdm.sign;
+ vsd.significand = f64_hi64to32jamming(vdm.significand);
+
+ /*
+ * If we have an infinity or a NaN, the exponent must be 255
+ */
+ if (tm & (F64_INFINITY|F64_NAN)) {
+ vsd.exponent = 255;
+ if (tm == F64_QNAN)
+ vsd.significand |= F64_SINGLE_SIGNIFICAND_QNAN;
+ goto pack_nan;
+ } else if (tm & F64_ZERO)
+ vsd.exponent = 0;
+ else
+ vsd.exponent = vdm.exponent - (1023 - 127);
+
+ return f64_single_normaliseround(sd, &vsd, fpscr, exceptions, "fcvts");
+
+ pack_nan:
+ f64_put_float(f64_single_pack(&vsd), sd);
+ return exceptions;
+}
+static u32 f64_double_fcvtd(int dd, int unused, int dm, u32 fpscr)
+{
+ return 0;
+}
+static u32 f64_double_ftosi(int sd, int unused, int dm, u32 fpscr)
+{
+ struct f64_double vdm;
+ u32 d, exceptions = 0;
+ int rmode = fpscr & FPSCR_RMODE_MASK;
+ int tm;
+
+ f64_double_unpack(&vdm, f64_get_double(dm));
+ f64_double_dump("VDM", &vdm);
+
+ /*
+ * Do we have denormalised number?
+ */
+ tm = f64_double_type(&vdm);
+ if (tm & F64_DENORMAL)
+ exceptions |= FPSCR_IDC;
+
+ if (tm & F64_NAN) {
+ d = 0;
+ exceptions |= FPSCR_IOC;
+ } else if (vdm.exponent >= 1023 + 32) {
+ d = 0x7fffffff;
+ if (vdm.sign)
+ d = ~d;
+ exceptions |= FPSCR_IOC;
+ } else if (vdm.exponent >= 1023 - 1) {
+ int shift = 1023 + 63 - vdm.exponent; /* 58 */
+ u64 rem, incr = 0;
+
+ d = (vdm.significand << 1) >> shift;
+ rem = vdm.significand << (65 - shift);
+
+ if (rmode == FPSCR_ROUND_NEAREST) {
+ incr = 0x8000000000000000ULL;
+ if ((d & 1) == 0)
+ incr -= 1;
+ } else if (rmode == FPSCR_ROUND_TOZERO) {
+ incr = 0;
+ } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vdm.sign != 0)) {
+ incr = ~0ULL;
+ }
+
+ if ((rem + incr) < rem && d < 0xffffffff)
+ d += 1;
+ if (d > 0x7fffffff + (vdm.sign != 0)) {
+ d = 0x7fffffff + (vdm.sign != 0);
+ exceptions |= FPSCR_IOC;
+ } else if (rem)
+ exceptions |= FPSCR_IXC;
+
+ if (vdm.sign)
+ d = -d;
+ } else {
+ d = 0;
+ if (vdm.exponent | vdm.significand) {
+ exceptions |= FPSCR_IXC;
+ if (rmode == FPSCR_ROUND_PLUSINF && vdm.sign == 0)
+ d = 1;
+ else if (rmode == FPSCR_ROUND_MINUSINF && vdm.sign)
+ d = -1;
+ }
+ }
+
+ pr_debug("UniCore-F64 ftosi: d(s%d)=%08x exceptions=%08x\n",
+ sd, d, exceptions);
+
+ f64_put_float((s32)d, sd);
+
+ return exceptions;
+}
+static struct op f_conv_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FCVTS)] = { f64_double_fcvts, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCVTD)] = { f64_double_fcvtd, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCVTW)] = { f64_double_ftosi, OP_SD },
+};
+static u32
+f64_double_fadd_nonnumber(struct f64_double *vdd, struct f64_double *vdn,
+ struct f64_double *vdm, u32 fpscr)
+{
+ struct f64_double *vdp;
+ u32 exceptions = 0;
+ int tn, tm;
+
+ tn = f64_double_type(vdn);
+ tm = f64_double_type(vdm);
+
+ if (tn & tm & F64_INFINITY) {
+ /*
+ * Two infinities. Are they different signs?
+ */
+ if (vdn->sign ^ vdm->sign) {
+ /*
+ * different signs -> invalid
+ */
+ exceptions = FPSCR_IOC;
+ vdp = &f64_double_default_qnan;
+ } else {
+ /*
+ * same signs -> valid
+ */
+ vdp = vdn;
+ }
+ } else if (tn & F64_INFINITY && tm & F64_NUMBER) {
+ /*
+ * One infinity and one number -> infinity
+ */
+ vdp = vdn;
+ } else {
+ /*
+ * 'n' is a NaN of some type
+ */
+ return f64_propagate_nan(vdd, vdn, vdm, fpscr);
+ }
+ *vdd = *vdp;
+ return exceptions;
+}
+
+static u32
+f64_double_add(struct f64_double *vdd, struct f64_double *vdn,
+ struct f64_double *vdm, u32 fpscr)
+{
+ u32 exp_diff;
+ u64 m_sig;
+
+ if (vdn->significand & (1ULL << 63) ||
+ vdm->significand & (1ULL << 63)) {
+ pr_info("UniCore-F64 bad FP values in %s\n", __func__);
+ f64_double_dump("VDN", vdn);
+ f64_double_dump("VDM", vdm);
+ }
+
+ /*
+ * Ensure that 'n' is the largest magnitude number. Note that
+ * if 'n' and 'm' have equal exponents, we do not swap them.
+ * This ensures that NaN propagation works correctly.
+ */
+ if (vdn->exponent < vdm->exponent) {
+ struct f64_double *t = vdn;
+ vdn = vdm;
+ vdm = t;
+ }
+
+ /*
+ * Is 'n' an infinity or a NaN? Note that 'm' may be a number,
+ * infinity or a NaN here.
+ */
+ if (vdn->exponent == 2047)
+ return f64_double_fadd_nonnumber(vdd, vdn, vdm, fpscr);
+
+ /*
+ * We have two proper numbers, where 'vdn' is the larger magnitude.
+ *
+ * Copy 'n' to 'd' before doing the arithmetic.
+ */
+ *vdd = *vdn;
+
+ /*
+ * Align 'm' with the result.
+ */
+ exp_diff = vdn->exponent - vdm->exponent;
+ m_sig = f64_shiftright64jamming(vdm->significand, exp_diff);
+
+ /*
+ * If the signs are different, we are really subtracting.
+ */
+ if (vdn->sign ^ vdm->sign) {
+ m_sig = vdn->significand - m_sig;
+ if ((s64)m_sig < 0) {
+ vdd->sign = f64_sign_negate(vdd->sign);
+ m_sig = -m_sig;
+ } else if (m_sig == 0) {
+ vdd->sign = (fpscr & FPSCR_RMODE_MASK) ==
+ FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
+ }
+ } else {
+ m_sig += vdn->significand;
+ }
+ vdd->significand = m_sig;
+
+ return 0;
+}
+
+static u32
+f64_double_multiply(struct f64_double *vdd, struct f64_double *vdn,
+ struct f64_double *vdm, u32 fpscr)
+{
+ f64_double_dump("VDN", vdn);
+ f64_double_dump("VDM", vdm);
+
+ /*
+ * Ensure that 'n' is the largest magnitude number. Note that
+ * if 'n' and 'm' have equal exponents, we do not swap them.
+ * This ensures that NaN propagation works correctly.
+ */
+ if (vdn->exponent < vdm->exponent) {
+ struct f64_double *t = vdn;
+ vdn = vdm;
+ vdm = t;
+ pr_debug("UniCore-F64 swapping M <-> N\n");
+ }
+
+ vdd->sign = vdn->sign ^ vdm->sign;
+
+ /*
+ * If 'n' is an infinity or NaN, handle it. 'm' may be anything.
+ */
+ if (vdn->exponent == 2047) {
+ if (vdn->significand || (vdm->exponent == 2047
+ && vdm->significand))
+ return f64_propagate_nan(vdd, vdn, vdm, fpscr);
+ if ((vdm->exponent | vdm->significand) == 0) {
+ *vdd = f64_double_default_qnan;
+ return FPSCR_IOC;
+ }
+ vdd->exponent = vdn->exponent;
+ vdd->significand = 0;
+ return 0;
+ }
+
+ /*
+ * If 'm' is zero, the result is always zero. In this case,
+ * 'n' may be zero or a number, but it doesn't matter which.
+ */
+ if ((vdm->exponent | vdm->significand) == 0) {
+ vdd->exponent = 0;
+ vdd->significand = 0;
+ return 0;
+ }
+
+ /*
+ * We add 2 to the destination exponent for the same reason
+ * as the addition case - though this time we have +1 from
+ * each input operand.
+ */
+ vdd->exponent = vdn->exponent + vdm->exponent - 1023 + 2;
+ vdd->significand = f64_hi64multiply64(vdn->significand,
+ vdm->significand);
+
+ f64_double_dump("VDD", vdd);
+ return 0;
+}
+/*
+ * Standard operations
+ */
+
+/*
+ * sd = sn * sm
+ */
+static u32 f64_double_fmul(int dd, int dn, int dm, u32 fpscr)
+{
+ struct f64_double vdd, vdn, vdm;
+ u32 exceptions;
+
+ f64_double_unpack(&vdn, f64_get_double(dn));
+ if (vdn.exponent == 0 && vdn.significand)
+ f64_double_normalise_denormal(&vdn);
+
+ f64_double_unpack(&vdm, f64_get_double(dm));
+ if (vdm.exponent == 0 && vdm.significand)
+ f64_double_normalise_denormal(&vdm);
+
+ exceptions = f64_double_multiply(&vdd, &vdn, &vdm, fpscr);
+ return f64_double_normaliseround(dd, &vdd, fpscr, exceptions, "fmul");
+}
+/*
+ * sd = sn + sm
+ */
+static u32 f64_double_fadd(int dd, int dn, int dm, u32 fpscr)
+{
+ struct f64_double vdd, vdn, vdm;
+ u32 exceptions;
+
+ f64_double_unpack(&vdn, f64_get_double(dn));
+ if (vdn.exponent == 0 && vdn.significand)
+ f64_double_normalise_denormal(&vdn);
+
+ f64_double_unpack(&vdm, f64_get_double(dm));
+ if (vdm.exponent == 0 && vdm.significand)
+ f64_double_normalise_denormal(&vdm);
+
+ exceptions = f64_double_add(&vdd, &vdn, &vdm, fpscr);
+
+ return f64_double_normaliseround(dd, &vdd, fpscr, exceptions, "fadd");
+}
+
+/*
+ * sd = sn - sm
+ */
+static u32 f64_double_fsub(int dd, int dn, int dm, u32 fpscr)
+{
+ struct f64_double vdd, vdn, vdm;
+ u32 exceptions;
+
+ f64_double_unpack(&vdn, f64_get_double(dn));
+ if (vdn.exponent == 0 && vdn.significand)
+ f64_double_normalise_denormal(&vdn);
+
+ f64_double_unpack(&vdm, f64_get_double(dm));
+ if (vdm.exponent == 0 && vdm.significand)
+ f64_double_normalise_denormal(&vdm);
+
+ /*
+ * Subtraction is like addition, but with a negated operand.
+ */
+ vdm.sign = f64_sign_negate(vdm.sign);
+
+ exceptions = f64_double_add(&vdd, &vdn, &vdm, fpscr);
+
+ return f64_double_normaliseround(dd, &vdd, fpscr, exceptions, "fsub");
+}
+static struct op f_arith_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FABS)] = { f64_double_fabs, OP_DD },
+ [FFUNC_TO_IDX(FOP_FMUL)] = { f64_double_fmul, OP_DD },
+ [FFUNC_TO_IDX(FOP_FADD)] = { f64_double_fadd, OP_DD },
+ [FFUNC_TO_IDX(FOP_FSUB)] = { f64_double_fsub, OP_DD },
+ [FFUNC_TO_IDX(FOP_FNEG)] = { f64_double_fneg, OP_DD },
+};
+
+
+#include "f64_double_cmp.h"
+
+u32 f64_double_cpdo(u32 inst, u32 fpscr)
+{
+ u32 op1 = (inst & FOP1_MASK) >> 26;
+ u32 func = (inst & FFUNC_MASK) >> 6;
+ u32 exceptions = 0;
+ u32 except;
+ char type;
+ unsigned int dest;
+ unsigned int dn = f64_get_dn(inst);
+ unsigned int dm = f64_get_dm(inst);
+ struct op *fop;
+
+ switch (op1) {
+ case 0:
+ fop = &f_arith_ops[func];
+ break;
+ case 2:
+ fop = &f_conv_ops[func];
+ break;
+ case 3:
+ fop = &f_cmp_ops[func];
+ break;
+ default:
+ goto invalid;
+ }
+
+ if (!fop->fn)
+ goto invalid;
+
+ if (fop->flags & OP_SD)
+ dest = f64_get_sd(inst);
+ else
+ dest = f64_get_dd(inst);
+
+ type = fop->flags & OP_SD ? 's' : 'd';
+ pr_debug("UniCore-F64 (%c%u) = (d%u) op1[%u] func[%u] (d%u)\n",
+ type, dest, dn, op1, func, dm);
+
+ except = fop->fn(dest, dn, dm, fpscr);
+ pr_debug("UniCore-F64 exceptions=%08x\n", except);
+
+ exceptions |= except;
+ return exceptions;
+
+ invalid:
+ return (u32)-1;
+}
diff --git a/arch/unicore32/uc-f64/f64hw.S b/arch/unicore32/uc-f64/f64hw.S
new file mode 100644
index 0000000..5388999
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64hw.S
@@ -0,0 +1,155 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64hw.S
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This code is called from the kernel's data abort trap.
+ * r19 holds the return address for successful handling.
+ * lr holds the return address for unrecognised instructions.
+ * r20 points at the start of the private FP workspace in the thread structure
+ * sp points to a struct pt_regs (as defined in include/asm/proc/ptrace.h)
+ */
+#include <linux/linkage.h>
+#include <asm/thread_info.h>
+
+#undef DEBUG
+ .macro DBGSTR, str
+#ifdef DEBUG
+ stm.w (r0-r3), [sp-]
+ stm.w (ip, lr), [sp-]
+ add r0, pc, #8
+ b.l printk
+ b 1f
+ .asciz "<7>UniCore-F64: \str\n"
+ .balign 4
+1: ldm.w (ip, lr), [sp]+
+ ldm.w (r0-r3), [sp]+
+#endif
+ .endm
+
+ .macro DBGSTR1, str, arg
+#ifdef DEBUG
+ stm.w (r0-r3), [sp-]
+ stm.w (ip, lr), [sp-]
+ mov r1, \arg
+ add r0, pc, #8
+ b.l printk
+ b 1f
+ .asciz "<7>UniCore-F64: \str\n"
+ .balign 4
+1: ldm.w (ip, lr), [sp]+
+ ldm.w (r0-r3), [sp]+
+#endif
+ .endm
+
+ .macro DBGSTR3, str, arg1, arg2, arg3
+#ifdef DEBUG
+ stm.w (r0-r3), [sp-]
+ stm.w (ip, lr), [sp-]
+ mov r3, \arg3
+ mov r2, \arg2
+ mov r1, \arg1
+ add r0, pc, #8
+ b.l printk
+ b 1f
+ .asciz "<7>UniCore-F64: \str\n"
+ .balign 4
+1: ldm.w (ip, lr), [sp]+
+ ldm.w (r0-r3), [sp]+
+#endif
+ .endm
+
+
+@ F64 hardware support entry point.
+@
+@ r0 = faulted instruction
+@ r2 = faulted PC
+@ r19 = successful return
+@ r20 = fp_state union
+@ lr = failure return
+
+ENTRY(f64_support_entry)
+ DBGSTR3 "instr %08x pc %08x state %p", r0, r2, r20
+
+ cff r1, s31 @ get fpu FPSCR
+ DBGSTR1 "fpscr %08x", r1
+ andn r2, r1, #0x08000000
+ ctf r2, s31 @ clear 27 bit
+process_exception:
+ DBGSTR "bounce"
+ mov r2, sp @ nothing stacked - regdump is at TOS
+ mov lr, r19 @ setup for a return to the user code.
+
+ @ Now call the C code to package up the bounce to the support code
+ @ r0 holds the trigger instruction
+ @ r1 holds the FPSCR value
+ @ r2 pointer to register dump
+ b F64_bounce @ we have handled this - the support
+ @ code will raise an exception if
+ @ required. If not, the user code will
+ @ retry the faulted instruction
+ENDPROC(f64_support_entry)
+
+ .macro tbl_branch, base, shift
+ add pc, pc, \base << \shift
+ .endm
+
+ENTRY(f64_get_float)
+ tbl_branch r0, #3
+ .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+1: mff r0, f\dr
+ mov pc, lr
+ .org 1b + 8
+ .endr
+ENDPROC(f64_get_float)
+
+ENTRY(f64_put_float)
+ tbl_branch r1, #3
+ .irp dr,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15
+1: mtf r0, f\dr
+ mov pc, lr
+ .org 1b + 8
+ .endr
+ENDPROC(f64_put_float)
+
+ENTRY(f64_get_double)
+ tbl_branch r0, #3
+ .irp dr 1,3,4,7,9,11,13,15,17,19,21,23,25,27,29,31
+1: mtf r1, f\dr
+ b 100f
+ .org 1b + 8
+ .endr
+100: tbl_branch r0, #3
+ .irp dr,0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30
+1: mtf r0, f\dr
+ mov pc, lr
+ .org 1b + 8
+ .endr
+
+ @ virtual register 16 for compare with zero
+ mov r0, #0
+ mov r1, #0
+ mov pc, lr
+ENDPROC(f64_get_double)
+
+ENTRY(f64_put_double)
+ tbl_branch r2, #3
+ .irp dr,0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30
+1: mtf r0, f\dr
+ b 100f
+ .org 1b + 8
+ .endr
+100: tbl_branch r2, #3
+ .irp dr 1,3,4,7,9,11,13,15,17,19,21,23,25,27,29,31
+1: mtf r1, f\dr
+ mov pc, lr
+ .org 1b + 8
+ .endr
+ENDPROC(f64_put_double)
diff --git a/arch/unicore32/uc-f64/f64instr.h b/arch/unicore32/uc-f64/f64instr.h
new file mode 100644
index 0000000..8355705
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64instr.h
@@ -0,0 +1,101 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64instr.h
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * UniCore-F64 instruction masks.
+ */
+#define INST_CPRTDO(inst) (((inst) & 0xf0000000) == 0xe0000000)
+#define INST_CPRT_DP(inst) ((inst) & (1 << 5))
+#define INST_CPNUM(inst) ((inst) & 0x00003c00)
+#define CPNUM(cp) ((cp) << 10)
+
+#define FOP1_MASK (0x0c000000)
+#define FFUNC_MASK (0x000003c0)
+
+#define FOP_FADD (0x00000000)
+#define FOP_FSUB (0x00000040)
+#define FOP_FMUL (0x00000080)
+#define FOP_FABS (0x00000140)
+#define FOP_FNEG (0x000001c0)
+
+#define FOP_FCVTS (0x08000000)
+#define FOP_FCVTD (0x08000040)
+#define FOP_FCVTW (0x08000100)
+
+#define FOP_FCF (0x0c000000)
+#define FOP_FCUN (0x0c000040)
+#define FOP_FCEQ (0x0c000080)
+#define FOP_FCUEQ (0x0c0000c0)
+#define FOP_FCOLT (0x0c000100)
+#define FOP_FCULT (0x0c000140)
+#define FOP_FCOLE (0x0c000180)
+#define FOP_FCUL (0x0c0001c0)
+#define FOP_FCSF (0x0c000200)
+#define FOP_FCNGLE (0x0c000240)
+#define FOP_FCSEQ (0x0c000280)
+#define FOP_FCNGL (0x0c0002c0)
+#define FOP_FCLT (0x0c000300)
+#define FOP_FCNGE (0x0c000340)
+#define FOP_FCLE (0x0c000380)
+#define FOP_FCNGT (0x0c0003c0)
+
+#define FFUNC_TO_IDX(inst) ((inst & (0x000003c0)) >> 6)
+
+#define f64_get_sd(inst) ((inst & 0x0007c000) >> 14)
+#define f64_get_dd(inst) ((f64_get_sd(inst)) >> 1)
+#define f64_get_sm(inst) ((inst & 0x0000001f))
+#define f64_get_dm(inst) ((f64_get_sm(inst)) >> 1)
+#define f64_get_sn(inst) ((inst & 0x00f80000) >> 19)
+#define f64_get_dn(inst) ((f64_get_sn(inst)) >> 1)
+#define f64_get_rd(inst) ((inst & 0x0007c000) >> 14)
+
+#define f64_get_fmt(inst) ((inst & 0x03000000) >> 24)
+#define f64_get_mffc_fmt(inst) ((inst & 0x04000000) >> 26)
+
+#define f64_single(inst) (((inst) & 0x03000000) == 0x00000000)
+
+#define FPSCR_N (1 << 31)
+#define FPSCR_Z (1 << 30)
+#define FPSCR_C (1 << 29)
+#define FPSCR_V (1 << 28)
+
+#define FSTATUS(except) ((except & (0xc0000000)) >> 28 | \
+ (except & (0x10000000)) >> 27 | \
+ (except & (0x04000000)) >> 26)
+#define FGT (0x2)
+#define FUOD (0x3)
+#define FEQ (0x6)
+#define FLT (0x8)
+
+/*
+ * Since we aren't building with -mfpu=f64, we need to code
+ * these instructions using their MRC/MCR equivalents.
+ */
+#define f64reg(_f64_) #_f64_
+
+#define cff(_f64_) ({ \
+ u32 __v; \
+ asm("cff %0, " f64reg(_f64_) "@ fmrx %0, " #_f64_ \
+ : "=r" (__v) : : "cc"); \
+ __v; \
+ })
+
+#define ctf(_f64_, _var_) \
+ asm("ctf %0, " f64reg(_f64_) "@ fmxr " #_f64_ ", %0" \
+ : : "r" (_var_) : "cc")
+
+u32 f64_single_cpdo(u32 inst, u32 fpscr);
+u32 f64_single_mffcdo(u32 inst, u32 ff, struct pt_regs *regs);
+
+u32 f64_sint_cpdo(u32 inst, u32 fpscr);
+
+u32 f64_double_cpdo(u32 inst, u32 fpscr);
+u32 f64_double_mffcdo(u32 inst, u32 ff, struct pt_regs *regs);
diff --git a/arch/unicore32/uc-f64/f64module.c b/arch/unicore32/uc-f64/f64module.c
new file mode 100644
index 0000000..bba5624
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64module.c
@@ -0,0 +1,180 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64module.c
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/init.h>
+
+#include <asm/uc-f64.h>
+#include "f64instr.h"
+#include "uc-f64.h"
+
+void (*f64_vector)(void) = f64_support_entry;
+
+/*
+ * Raise a SIGFPE for the current process.
+ * sicode describes the signal being raised.
+ */
+void f64_raise_sigfpe(unsigned int sicode, struct pt_regs *regs)
+{
+ siginfo_t info;
+
+ memset(&info, 0, sizeof(info));
+
+ info.si_signo = SIGFPE;
+ info.si_code = sicode;
+ info.si_addr = (void __user *)(instruction_pointer(regs) - 4);
+
+ /*
+ * This is the same as NWFPE, because it's not clear what
+ * this is used for
+ */
+ current->thread.error_code = 0;
+ current->thread.trap_no = 6;
+
+ send_sig_info(SIGFPE, &info, current);
+}
+
+static void f64_panic(char *reason, u32 inst)
+{
+ int i;
+
+ printk(KERN_ERR "UniCore-F64 Error: %s\n", reason);
+ printk(KERN_ERR "UniCore-F64 FPSCR 0x%08x INST 0x%08x\n",
+ cff(FPSCR), inst);
+ for (i = 0; i < 32; i += 2)
+ printk(KERN_ERR "UniCore-F64 s%2u: 0x%08x s%2u: 0x%08x\n",
+ i, f64_get_float(i), i+1, f64_get_float(i+1));
+}
+
+/*
+ * Process bitmask of exception conditions.
+ */
+static void f64_raise_exceptions(u32 exceptions, u32 inst, u32 fpscr,
+ struct pt_regs *regs)
+{
+ if (exceptions & FPSCR_CMPINSTR_BIT) {
+ if (exceptions & FPSCR_CON)
+ fpscr |= FPSCR_CON;
+ else
+ fpscr &= ~(FPSCR_CON);
+ exceptions &= ~(FPSCR_CMPINSTR_BIT | FPSCR_CON);
+ } else
+ pr_debug("UniCore-F64 raising exceptions %08x\n", exceptions);
+
+ if (exceptions == F64_EXCEPTION_ERROR) {
+ f64_panic("unhandled bounce", inst);
+ f64_raise_sigfpe(0, regs);
+ return;
+ }
+
+ /*
+ * Update the FPSCR with the additional exception flags.
+ * Comparison instructions always return at least one of
+ * these flags set.
+ */
+ fpscr &= ~(FPSCR_TRAP | FPSCR_IOS | FPSCR_OFS | FPSCR_UFS |
+ FPSCR_IXS | FPSCR_HIS | FPSCR_IOC | FPSCR_OFC |
+ FPSCR_UFC | FPSCR_IXC | FPSCR_HIC);
+
+ fpscr |= exceptions;
+ ctf(FPSCR, fpscr);
+}
+/*
+ * Emulate a F64 instruction.
+ */
+static u32 f64_emulate_instruction(u32 inst, u32 fpscr, struct pt_regs *regs)
+{
+ u32 exceptions = F64_EXCEPTION_ERROR;
+ u32 fmt = f64_get_fmt(inst);
+
+ pr_debug("UniCore-F64 emulate: INST=0x%08x SCR=0x%08x\n", inst, fpscr);
+
+ if (INST_CPRTDO(inst)) {
+ if (INST_CPRT_DP(inst)) {
+ /*
+ * CPDO
+ */
+ switch (fmt) {
+ case 0:
+ exceptions = f64_single_cpdo(inst, fpscr);
+ break;
+ case 1:
+ exceptions = f64_double_cpdo(inst, fpscr);
+ break;
+ case 2:
+ exceptions = f64_sint_cpdo(inst, fpscr);
+ break;
+ default:
+ return exceptions;
+ }
+ } else {
+ /* deal with MFFC */
+ fmt = f64_get_mffc_fmt(inst);
+ switch (fmt) {
+ case 0:
+ exceptions = f64_single_mffcdo(inst, fpscr,
+ regs);
+ break;
+ case 1:
+ exceptions = f64_double_mffcdo(inst, fpscr,
+ regs);
+ break;
+ default:
+ return exceptions;
+ }
+ }
+ } else {
+ /*
+ * A CPDT instruction can not cause an exception.
+ * Therefore, we do not have to emulate it.
+ */
+ }
+
+ return exceptions & ~F64_NAN_FLAG;
+}
+/*
+ * Package up a bounce condition.
+ */
+void F64_bounce(u32 trigger, u32 fpexc, struct pt_regs *regs)
+{
+ u32 exceptions;
+ u32 orig_fpscr = fpexc;
+ pr_debug("UniCore-F64: bounce: trigger %08x fpscr %08x\n",
+ trigger, fpexc);
+ /* emulate the inst */
+ exceptions = f64_emulate_instruction(trigger, orig_fpscr, regs);
+ if (exceptions)
+ f64_raise_exceptions(exceptions, trigger, orig_fpscr, regs);
+}
+
+#include <linux/smp.h>
+
+/*
+ * F64 support code initialisation.
+ */
+static int __init f64_init(void)
+{
+ f64_vector = f64_support_entry;
+ barrier();
+
+ ctf(FPSCR, 0x0); /* FPSCR_UFE | FPSCR_NDE perhaps better */
+
+ printk(KERN_INFO "UniCore-F64 support\n");
+
+ return 0;
+}
+
+late_initcall(f64_init);
diff --git a/arch/unicore32/uc-f64/f64single.c b/arch/unicore32/uc-f64/f64single.c
new file mode 100644
index 0000000..f6ef44c
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64single.c
@@ -0,0 +1,771 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64single.c
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This code is derived in part from John R. Housers softfloat library, which
+ * carries the following notice:
+ *
+ * ===========================================================================
+ * This C source file is part of the SoftFloat IEC/IEEE Floating-point
+ * Arithmetic Package, Release 2.
+ *
+ * Written by John R. Hauser. This work was made possible in part by the
+ * International Computer Science Institute, located at Suite 600, 1947 Center
+ * Street, Berkeley, California 94704. Funding was partially provided by the
+ * National Science Foundation under grant MIP-9311980. The original version
+ * of this code was written as part of a project to build a fixed-point vector
+ * processor in collaboration with the University of California at Berkeley,
+ * overseen by Profs. Nelson Morgan and John Wawrzynek. More information
+ * is available through the web page `http://HTTP.CS.Berkeley.EDU/~jhauser/
+ * arithmetic/softfloat.html'.
+ *
+ * THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort
+ * has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT
+ * TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO
+ * PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY
+ * AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE.
+ *
+ * Derivative works are acceptable, even for commercial purposes, so long as
+ * (1) they include prominent notice that the work is derivative, and (2) they
+ * include prominent notice akin to these three paragraphs for those parts of
+ * this code that are retained.
+ * ===========================================================================
+ */
+#include <linux/kernel.h>
+#include <linux/bitops.h>
+
+#include <asm/div64.h>
+#include <asm/uc-f64.h>
+
+#include "f64instr.h"
+#include "uc-f64.h"
+
+static struct f64_single f64_single_default_qnan = {
+ .exponent = 255,
+ .sign = 0,
+ .significand = F64_SINGLE_SIGNIFICAND_QNAN,
+};
+
+static void f64_single_dump(const char *str, struct f64_single *s)
+{
+ pr_debug("UniCore-F64: %s: sign=%d exponent=%d significand=%08x\n",
+ str, s->sign != 0, s->exponent, s->significand);
+}
+
+static void f64_single_normalise_denormal(struct f64_single *vs)
+{
+ int bits = 31 - fls(vs->significand);
+
+ f64_single_dump("normalise_denormal: in", vs);
+
+ if (bits) {
+ vs->exponent -= bits - 1;
+ vs->significand <<= bits;
+ }
+
+ f64_single_dump("normalise_denormal: out", vs);
+}
+
+#ifndef DEBUG
+#define f64_single_normaliseround(sd, vsd, fpscr, except, func) \
+ __f64_single_normaliseround(sd, vsd, fpscr, except)
+u32 __f64_single_normaliseround(int sd, struct f64_single *vs, u32 fpscr,
+ u32 exceptions)
+#else
+u32 f64_single_normaliseround(int sd, struct f64_single *vs, u32 fpscr,
+ u32 exceptions, const char *func)
+#endif
+{
+ u32 significand, incr, rmode;
+ int exponent, shift, underflow;
+
+ f64_single_dump("pack: in", vs);
+
+ /*
+ * Infinities and NaNs are a special case.
+ */
+ if (vs->exponent == 255 && (vs->significand == 0 || exceptions))
+ goto pack;
+
+ /*
+ * Special-case zero.
+ */
+ if (vs->significand == 0) {
+ vs->exponent = 0;
+ goto pack;
+ }
+
+ exponent = vs->exponent;
+ significand = vs->significand;
+
+ /*
+ * Normalise first. Note that we shift the significand up to
+ * bit 31, so we have F64_SINGLE_LOW_BITS + 1 below the least
+ * significant bit.
+ */
+ shift = 32 - fls(significand);
+ if (shift < 32 && shift) {
+ exponent -= shift;
+ significand <<= shift;
+ }
+
+#ifdef DEBUG
+ vs->exponent = exponent;
+ vs->significand = significand;
+ f64_single_dump("pack: normalised", vs);
+#endif
+
+ /*
+ * Tiny number?
+ */
+ underflow = exponent < 0;
+ if (underflow) {
+ significand = f64_shiftright32jamming(significand, -exponent);
+ exponent = 0;
+#ifdef DEBUG
+ vs->exponent = exponent;
+ vs->significand = significand;
+ f64_single_dump("pack: tiny number", vs);
+#endif
+ if (!(significand & ((1 << (F64_SINGLE_LOW_BITS + 1)) - 1)))
+ underflow = 0;
+ }
+
+ /*
+ * Select rounding increment.
+ */
+ incr = 0;
+ rmode = fpscr & FPSCR_RMODE_MASK;
+
+ if (rmode == FPSCR_ROUND_NEAREST) {
+ incr = 1 << F64_SINGLE_LOW_BITS;
+ if ((significand & (1 << (F64_SINGLE_LOW_BITS + 1))) == 0)
+ incr -= 1;
+ } else if (rmode == FPSCR_ROUND_TOZERO) {
+ incr = 0;
+ } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vs->sign != 0))
+ incr = (1 << (F64_SINGLE_LOW_BITS + 1)) - 1;
+
+ pr_debug("UniCore-F64: rounding increment = 0x%08x\n", incr);
+
+ /*
+ * Is our rounding going to overflow?
+ */
+ if ((significand + incr) < significand) {
+ exponent += 1;
+ significand = (significand >> 1) | (significand & 1);
+ incr >>= 1;
+#ifdef DEBUG
+ vs->exponent = exponent;
+ vs->significand = significand;
+ f64_single_dump("pack: overflow", vs);
+#endif
+ }
+
+ /*
+ * If any of the low bits (which will be shifted out of the
+ * number) are non-zero, the result is inexact.
+ */
+ if (significand & ((1 << (F64_SINGLE_LOW_BITS + 1)) - 1))
+ exceptions |= FPSCR_IXC;
+
+ /*
+ * Do our rounding.
+ */
+ significand += incr;
+
+ /*
+ * Infinity?
+ */
+ if (exponent >= 254) {
+ exceptions |= FPSCR_OFC | FPSCR_IXC;
+ if (incr == 0) {
+ vs->exponent = 253;
+ vs->significand = 0x7fffffff;
+ } else {
+ vs->exponent = 255; /* infinity */
+ vs->significand = 0;
+ }
+ } else {
+ if (significand >> (F64_SINGLE_LOW_BITS + 1) == 0)
+ exponent = 0;
+ if (exponent || significand > 0x80000000)
+ underflow = 0;
+ if (underflow)
+ exceptions |= FPSCR_UFC;
+ vs->exponent = exponent;
+ vs->significand = significand >> 1;
+ }
+
+ pack:
+ f64_single_dump("pack: final", vs);
+ {
+ s32 d = f64_single_pack(vs);
+#ifdef DEBUG
+ pr_debug("UniCore-F64: %s: d(s%d)=%08x exceptions=%08x\n", func,
+ sd, d, exceptions);
+#endif
+ f64_put_float(d, sd);
+ }
+
+ return exceptions;
+}
+
+/*
+ * Propagate the NaN, setting exceptions if it is signalling.
+ * 'n' is always a NaN. 'm' may be a number, NaN or infinity.
+ */
+static u32 f64_propagate_nan(struct f64_single *vsd, struct f64_single *vsn,
+ struct f64_single *vsm, u32 fpscr)
+{
+ struct f64_single *nan;
+ int tn, tm = 0;
+
+ tn = f64_single_type(vsn);
+
+ if (vsm)
+ tm = f64_single_type(vsm);
+
+ if (fpscr & FPSCR_DEFAULT_NAN)
+ /*
+ * Default NaN mode - always returns a quiet NaN
+ */
+ nan = &f64_single_default_qnan;
+ else {
+ /*
+ * Contemporary mode - select the first signalling
+ * NAN, or if neither are signalling, the first
+ * quiet NAN.
+ */
+ if (tn == F64_SNAN || (tm != F64_SNAN && tn == F64_QNAN))
+ nan = vsn;
+ else
+ nan = vsm;
+ /*
+ * Make the NaN quiet.
+ */
+ nan->significand |= F64_SINGLE_SIGNIFICAND_QNAN;
+ }
+
+ *vsd = *nan;
+
+ /*
+ * If one was a signalling NAN, raise invalid operation.
+ */
+ return tn == F64_SNAN || tm == F64_SNAN ? FPSCR_IOC : F64_NAN_FLAG;
+}
+
+
+/*
+ * Extended operations
+ */
+static u32 f64_single_fabs(int sd, int unused, s32 m, u32 fpscr)
+{
+ f64_put_float(f64_single_packed_abs(m), sd);
+ return 0;
+}
+
+static u32 f64_single_fneg(int sd, int unused, s32 m, u32 fpscr)
+{
+ f64_put_float(f64_single_packed_negate(m), sd);
+ return 0;
+}
+
+/*
+ * Equal := ZC
+ * Less than := N
+ * Greater than := C
+ * Unordered := CV
+ */
+static u32 f64_compare(int sd, int signal_on_qnan, s32 m, u32 fpscr)
+{
+ s32 d;
+ u32 ret = 0;
+
+ d = f64_get_float(sd);
+ if (f64_single_packed_exponent(m) == 255
+ && f64_single_packed_mantissa(m)) {
+ ret |= FPSCR_C | FPSCR_V;
+ if (signal_on_qnan || !(f64_single_packed_mantissa(m)
+ & (1 << (F64_SINGLE_MANTISSA_BITS - 1))))
+ /*
+ * Signalling NaN, or signalling on quiet NaN
+ */
+ ret |= FPSCR_IOC;
+ }
+
+ if (f64_single_packed_exponent(d) == 255
+ && f64_single_packed_mantissa(d)) {
+ ret |= FPSCR_C | FPSCR_V;
+ if (signal_on_qnan || !(f64_single_packed_mantissa(d)
+ & (1 << (F64_SINGLE_MANTISSA_BITS - 1))))
+ /*
+ * Signalling NaN, or signalling on quiet NaN
+ */
+ ret |= FPSCR_IOC;
+ }
+
+ if (ret == 0) {
+ if (d == m || f64_single_packed_abs(d | m) == 0) {
+ /*
+ * equal
+ */
+ ret |= FPSCR_Z | FPSCR_C;
+ } else if (f64_single_packed_sign(d ^ m)) {
+ /*
+ * different signs
+ */
+ if (f64_single_packed_sign(d))
+ /*
+ * d is negative, so d < m
+ */
+ ret |= FPSCR_N;
+ else
+ /*
+ * d is positive, so d > m
+ */
+ ret |= FPSCR_C;
+ } else if ((f64_single_packed_sign(d) != 0) ^ (d < m)) {
+ /*
+ * d < m
+ */
+ ret |= FPSCR_N;
+ } else if ((f64_single_packed_sign(d) != 0) ^ (d > m)) {
+ /*
+ * d > m
+ */
+ ret |= FPSCR_C;
+ }
+ }
+ return ret;
+}
+
+/*
+ * Conversion operations
+ */
+
+static u32 f64_single_fcvts(int dd, int unused, s32 m, u32 fpscr)
+{
+ return 0;
+}
+
+static u32 f64_single_fcvtd(int dd, int unused, s32 m, u32 fpscr)
+{
+ struct f64_single vsm;
+ struct f64_double vdd;
+ int tm;
+ u32 exceptions = 0;
+
+ f64_single_unpack(&vsm, m);
+
+ tm = f64_single_type(&vsm);
+
+ /*
+ * If we have a signalling NaN, signal invalid operation.
+ */
+ if (tm == F64_SNAN)
+ exceptions = FPSCR_IOC;
+
+ if (tm & F64_DENORMAL)
+ f64_single_normalise_denormal(&vsm);
+
+ vdd.sign = vsm.sign;
+ vdd.significand = (u64)vsm.significand << 32;
+
+ /*
+ * If we have an infinity or NaN, the exponent must be 2047.
+ */
+ if (tm & (F64_INFINITY|F64_NAN)) {
+ vdd.exponent = 2047;
+ if (tm == F64_QNAN)
+ vdd.significand |= F64_DOUBLE_SIGNIFICAND_QNAN;
+ goto pack_nan;
+ } else if (tm & F64_ZERO)
+ vdd.exponent = 0;
+ else
+ vdd.exponent = vsm.exponent + (1023 - 127);
+
+ return f64_double_normaliseround(dd, &vdd, fpscr, exceptions, "fcvtd");
+
+ pack_nan:
+ f64_put_double(f64_double_pack(&vdd), dd);
+ return exceptions;
+}
+
+static u32 f64_single_ftosi(int sd, int unused, s32 m, u32 fpscr)
+{
+ struct f64_single vsm;
+ u32 d, exceptions = 0;
+ int rmode = fpscr & FPSCR_RMODE_MASK;
+ int tm;
+
+ f64_single_unpack(&vsm, m);
+ f64_single_dump("VSM", &vsm);
+
+ /*
+ * Do we have a denormalised number?
+ */
+ tm = f64_single_type(&vsm);
+ if (f64_single_type(&vsm) & F64_DENORMAL)
+ exceptions |= FPSCR_IDC;
+
+ if (tm & F64_NAN) {
+ d = 0;
+ exceptions |= FPSCR_IOC;
+ } else if (vsm.exponent >= 127 + 32) {
+ /*
+ * m >= 2^31-2^7: invalid
+ */
+ d = 0x7fffffff;
+ if (vsm.sign)
+ d = ~d;
+ exceptions |= FPSCR_IOC;
+ } else if (vsm.exponent >= 127 - 1) {
+ int shift = 127 + 31 - vsm.exponent;
+ u32 rem, incr = 0;
+
+ /* 2^0 <= m <= 2^31-2^7 */
+ d = (vsm.significand << 1) >> shift;
+ rem = vsm.significand << (33 - shift);
+
+ if (rmode == FPSCR_ROUND_NEAREST) {
+ incr = 0x80000000;
+ if ((d & 1) == 0)
+ incr -= 1;
+ } else if (rmode == FPSCR_ROUND_TOZERO) {
+ incr = 0;
+ } else if ((rmode == FPSCR_ROUND_PLUSINF) ^ (vsm.sign != 0)) {
+ incr = ~0;
+ }
+
+ if ((rem + incr) < rem && d < 0xffffffff)
+ d += 1;
+ if (d > 0x7fffffff + (vsm.sign != 0)) {
+ d = 0x7fffffff + (vsm.sign != 0);
+ exceptions |= FPSCR_IOC;
+ } else if (rem)
+ exceptions |= FPSCR_IXC;
+
+ if (vsm.sign)
+ d = -d;
+ } else {
+ d = 0;
+ if (vsm.exponent | vsm.significand) {
+ exceptions |= FPSCR_IXC;
+ if (rmode == FPSCR_ROUND_PLUSINF && vsm.sign == 0)
+ d = 1;
+ else if (rmode == FPSCR_ROUND_MINUSINF && vsm.sign)
+ d = -1;
+ }
+ }
+
+ pr_debug("UniCore-F64: ftosi: d(s%d)=%08x exceptions=%08x\n",
+ sd, d, exceptions);
+
+ f64_put_float((s32)d, sd);
+
+ return exceptions;
+}
+
+static struct op f_conv_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FCVTS)] = { f64_single_fcvts, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCVTD)] = { f64_single_fcvtd, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCVTW)] = { f64_single_ftosi, OP_SD },
+};
+
+static u32 f64_single_fadd_nonnumber(struct f64_single *vsd,
+ struct f64_single *vsn, struct f64_single *vsm, u32 fpscr)
+{
+ struct f64_single *vsp;
+ u32 exceptions = 0;
+ int tn, tm;
+
+ tn = f64_single_type(vsn);
+ tm = f64_single_type(vsm);
+
+ if (tn & tm & F64_INFINITY) {
+ /*
+ * Two infinities. Are they different signs?
+ */
+ if (vsn->sign ^ vsm->sign) {
+ /*
+ * different signs -> invalid
+ */
+ exceptions = FPSCR_IOC;
+ vsp = &f64_single_default_qnan;
+ } else {
+ /*
+ * same signs -> valid
+ */
+ vsp = vsn;
+ }
+ } else if (tn & F64_INFINITY && tm & F64_NUMBER) {
+ /*
+ * One infinity and one number -> infinity
+ */
+ vsp = vsn;
+ } else {
+ /*
+ * 'n' is a NaN of some type
+ */
+ return f64_propagate_nan(vsd, vsn, vsm, fpscr);
+ }
+ *vsd = *vsp;
+ return exceptions;
+}
+
+static u32 f64_single_add(struct f64_single *vsd, struct f64_single *vsn,
+ struct f64_single *vsm, u32 fpscr)
+{
+ u32 exp_diff, m_sig;
+
+ if (vsn->significand & 0x80000000 ||
+ vsm->significand & 0x80000000) {
+ pr_info("UniCore-F64: bad FP values in %s\n", __func__);
+ f64_single_dump("VSN", vsn);
+ f64_single_dump("VSM", vsm);
+ }
+
+ /*
+ * Ensure that 'n' is the largest magnitude number. Note that
+ * if 'n' and 'm' have equal exponents, we do not swap them.
+ * This ensures that NaN propagation works correctly.
+ */
+ if (vsn->exponent < vsm->exponent) {
+ struct f64_single *t = vsn;
+ vsn = vsm;
+ vsm = t;
+ }
+
+ /*
+ * Is 'n' an infinity or a NaN? Note that 'm' may be a number,
+ * infinity or a NaN here.
+ */
+ if (vsn->exponent == 255)
+ return f64_single_fadd_nonnumber(vsd, vsn, vsm, fpscr);
+
+ /*
+ * We have two proper numbers, where 'vsn' is the larger magnitude.
+ *
+ * Copy 'n' to 'd' before doing the arithmetic.
+ */
+ *vsd = *vsn;
+
+ /*
+ * Align both numbers.
+ */
+ exp_diff = vsn->exponent - vsm->exponent;
+ m_sig = f64_shiftright32jamming(vsm->significand, exp_diff);
+
+ /*
+ * If the signs are different, we are really subtracting.
+ */
+ if (vsn->sign ^ vsm->sign) {
+ m_sig = vsn->significand - m_sig;
+ if ((s32)m_sig < 0) {
+ vsd->sign = f64_sign_negate(vsd->sign);
+ m_sig = -m_sig;
+ } else if (m_sig == 0) {
+ vsd->sign = (fpscr & FPSCR_RMODE_MASK) ==
+ FPSCR_ROUND_MINUSINF ? 0x8000 : 0;
+ }
+ } else {
+ m_sig = vsn->significand + m_sig;
+ }
+ vsd->significand = m_sig;
+
+ return 0;
+}
+
+static u32 f64_single_multiply(struct f64_single *vsd, struct f64_single *vsn,
+ struct f64_single *vsm, u32 fpscr)
+{
+ f64_single_dump("VSN", vsn);
+ f64_single_dump("VSM", vsm);
+
+ /*
+ * Ensure that 'n' is the largest magnitude number. Note that
+ * if 'n' and 'm' have equal exponents, we do not swap them.
+ * This ensures that NaN propagation works correctly.
+ */
+ if (vsn->exponent < vsm->exponent) {
+ struct f64_single *t = vsn;
+ vsn = vsm;
+ vsm = t;
+ pr_debug("UniCore-F64: swapping M <-> N\n");
+ }
+
+ vsd->sign = vsn->sign ^ vsm->sign;
+
+ /*
+ * If 'n' is an infinity or NaN, handle it. 'm' may be anything.
+ */
+ if (vsn->exponent == 255) {
+ if (vsn->significand || (vsm->exponent == 255
+ && vsm->significand))
+ return f64_propagate_nan(vsd, vsn, vsm, fpscr);
+ if ((vsm->exponent | vsm->significand) == 0) {
+ *vsd = f64_single_default_qnan;
+ return FPSCR_IOC;
+ }
+ vsd->exponent = vsn->exponent;
+ vsd->significand = 0;
+ return 0;
+ }
+
+ /*
+ * If 'm' is zero, the result is always zero. In this case,
+ * 'n' may be zero or a number, but it doesn't matter which.
+ */
+ if ((vsm->exponent | vsm->significand) == 0) {
+ vsd->exponent = 0;
+ vsd->significand = 0;
+ return 0;
+ }
+
+ /*
+ * We add 2 to the destination exponent for the same reason as
+ * the addition case - though this time we have +1 from each
+ * input operand.
+ */
+ vsd->exponent = vsn->exponent + vsm->exponent - 127 + 2;
+ vsd->significand = f64_hi64to32jamming((u64)vsn->significand
+ * vsm->significand);
+
+ f64_single_dump("VSD", vsd);
+ return 0;
+}
+
+/*
+ * Standard operations
+ */
+
+/*
+ * sd = sn * sm
+ */
+static u32 f64_single_fmul(int sd, int sn, s32 m, u32 fpscr)
+{
+ struct f64_single vsd, vsn, vsm;
+ u32 exceptions;
+ s32 n = f64_get_float(sn);
+
+ pr_debug("UniCore-F64: s%u = %08x\n", sn, n);
+
+ f64_single_unpack(&vsn, n);
+ if (vsn.exponent == 0 && vsn.significand)
+ f64_single_normalise_denormal(&vsn);
+
+ f64_single_unpack(&vsm, m);
+ if (vsm.exponent == 0 && vsm.significand)
+ f64_single_normalise_denormal(&vsm);
+
+ exceptions = f64_single_multiply(&vsd, &vsn, &vsm, fpscr);
+ return f64_single_normaliseround(sd, &vsd, fpscr, exceptions, "fmul");
+}
+
+/*
+ * sd = sn + sm
+ */
+static u32 f64_single_fadd(int sd, int sn, s32 m, u32 fpscr)
+{
+ struct f64_single vsd, vsn, vsm;
+ u32 exceptions;
+ s32 n = f64_get_float(sn);
+
+ pr_debug("UniCore-F64: s%u = %08x\n", sn, n);
+
+ /*
+ * Unpack and normalise denormals.
+ */
+ f64_single_unpack(&vsn, n);
+ if (vsn.exponent == 0 && vsn.significand)
+ f64_single_normalise_denormal(&vsn);
+
+ f64_single_unpack(&vsm, m);
+ if (vsm.exponent == 0 && vsm.significand)
+ f64_single_normalise_denormal(&vsm);
+
+ exceptions = f64_single_add(&vsd, &vsn, &vsm, fpscr);
+
+ return f64_single_normaliseround(sd, &vsd, fpscr, exceptions, "fadd");
+}
+
+/*
+ * sd = sn - sm
+ */
+static u32 f64_single_fsub(int sd, int sn, s32 m, u32 fpscr)
+{
+ /*
+ * Subtraction is addition with one sign inverted.
+ */
+ return f64_single_fadd(sd, sn, f64_single_packed_negate(m), fpscr);
+}
+
+
+static struct op f_arith_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FABS)] = { f64_single_fabs, OP_SD },
+ [FFUNC_TO_IDX(FOP_FMUL)] = { f64_single_fmul, OP_SD },
+ [FFUNC_TO_IDX(FOP_FADD)] = { f64_single_fadd, OP_SD },
+ [FFUNC_TO_IDX(FOP_FSUB)] = { f64_single_fsub, OP_SD },
+ [FFUNC_TO_IDX(FOP_FNEG)] = { f64_single_fneg, OP_SD },
+};
+
+#include "f64_single_cmp.h"
+u32 f64_single_cpdo(u32 inst, u32 fpscr)
+{
+ u32 op1 = (inst & FOP1_MASK) >> 26;
+ u32 func = (inst & FFUNC_MASK) >> 6;
+ u32 exceptions = 0;
+ u32 except;
+ s32 m;
+ char type;
+ unsigned int dest;
+ unsigned int sn = f64_get_sn(inst);
+ unsigned int sm = f64_get_sm(inst);
+ struct op *fop;
+
+ switch (op1) {
+ case 0:
+ fop = &f_arith_ops[func];
+ break;
+ case 2:
+ fop = &f_conv_ops[func];
+ break;
+ case 3:
+ fop = &f_cmp_ops[func];
+ break;
+ default:
+ goto invalid;
+ }
+
+ if (!fop->fn)
+ goto invalid;
+
+ if (fop->flags & OP_DD)
+ dest = f64_get_dd(inst);
+ else
+ dest = f64_get_sd(inst);
+ m = f64_get_float(sm);
+
+ type = fop->flags & OP_DD ? 'd' : 's';
+ pr_debug("UniCore-F64: (%c%u) = (s%u) op1[%u] func[%u] (s%u=%08x)\n",
+ type, dest, sn,
+ op1, func, sm, m);
+
+ except = fop->fn(dest, sn, m, fpscr);
+ pr_debug("UniCore-F64: exceptions=%08x\n", except);
+ exceptions |= except;
+
+ return exceptions;
+
+invalid:
+ return (u32)-1;
+}
diff --git a/arch/unicore32/uc-f64/f64sint.c b/arch/unicore32/uc-f64/f64sint.c
new file mode 100644
index 0000000..0e909ac
--- /dev/null
+++ b/arch/unicore32/uc-f64/f64sint.c
@@ -0,0 +1,94 @@
+/*
+ * linux/arch/unicore32/uc-f64/f64sint.c
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/kernel.h>
+#include <linux/bitops.h>
+
+#include <asm/div64.h>
+#include <asm/uc-f64.h>
+
+#include "f64instr.h"
+#include "uc-f64.h"
+
+static u32 f64_sint_fcvts(int sd, int unused, s32 m, u32 fpscr)
+{
+ struct f64_single vs;
+
+ vs.sign = (m & 0x80000000) >> 16;
+ vs.exponent = 127 + 31 - 1;
+ vs.significand = vs.sign ? -m : m;
+
+ return f64_single_normaliseround(sd, &vs, fpscr, 0, "f64_sint_fcvts");
+}
+
+static u32 f64_sint_fcvtd(int dd, int unused, s32 m, u32 fpscr)
+{
+ struct f64_double vdm;
+
+ vdm.sign = (m & 0x80000000) >> 16;
+ vdm.exponent = 1023 + 63 - 1;
+ vdm.significand = vdm.sign ? -m : m;
+
+ return f64_double_normaliseround(dd, &vdm, fpscr, 0, "f64_sint_fcvtd");
+}
+
+static u32 f64_sint_ftosi(int dd, int unused, s32 m, u32 fpscr)
+{
+ return 0;
+}
+static struct op f_conv_ops[16] = {
+ [FFUNC_TO_IDX(FOP_FCVTS)] = { f64_sint_fcvts, OP_SD },
+ [FFUNC_TO_IDX(FOP_FCVTD)] = { f64_sint_fcvtd, OP_DD },
+ [FFUNC_TO_IDX(FOP_FCVTW)] = { f64_sint_ftosi, OP_SD },
+};
+
+u32 f64_sint_cpdo(u32 inst, u32 ff)
+{
+ u32 op1 = (inst & FOP1_MASK) >> 26;
+ u32 func = (inst & FFUNC_MASK) >> 6;
+ u32 exceptions = 0;
+ u32 except;
+ s32 m;
+ char type;
+ unsigned int dest;
+ unsigned int sn = f64_get_sn(inst);
+ unsigned int sm = f64_get_sm(inst);
+ struct op *fop;
+
+ if (op1 == 2)
+ fop = &f_conv_ops[func];
+ else
+ goto invalid;
+
+ if (!fop->fn)
+ goto invalid;
+
+ if (fop->flags & OP_SD)
+ dest = f64_get_sd(inst);
+ else
+ dest = f64_get_dd(inst);
+
+ m = f64_get_float(sm);
+
+ type = fop->flags & OP_DD ? 'd' : 's';
+ pr_debug("UniCore-F64: (%c%u) = (s%u) op1[%u] func[%u] (s%u=%08x)\n",
+ type, dest, sn,
+ op1, func, sm, m);
+
+ except = fop->fn(dest, sn, m, ff);
+ exceptions |= except;
+
+ return exceptions;
+
+ invalid:
+ return (u32)-1;
+}
diff --git a/arch/unicore32/uc-f64/uc-f64.h b/arch/unicore32/uc-f64/uc-f64.h
new file mode 100644
index 0000000..1e458d7
--- /dev/null
+++ b/arch/unicore32/uc-f64/uc-f64.h
@@ -0,0 +1,332 @@
+/*
+ * linux/arch/unicore/uc-f64/uc-f64.h
+ *
+ * Code specific to PKUnity SoC and UniCore ISA
+ * Fragments that appear the same as the files in arm vfp
+ *
+ * Copyright (C) 2001-2010 GUAN Xue-tao
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+static inline u32 f64_shiftright32jamming(u32 val, unsigned int shift)
+{
+ if (shift) {
+ if (shift < 32)
+ val = val >> shift | ((val << (32 - shift)) != 0);
+ else
+ val = val != 0;
+ }
+ return val;
+}
+
+static inline u64 f64_shiftright64jamming(u64 val, unsigned int shift)
+{
+ if (shift) {
+ if (shift < 64)
+ val = val >> shift | ((val << (64 - shift)) != 0);
+ else
+ val = val != 0;
+ }
+ return val;
+}
+
+static inline u32 f64_hi64to32jamming(u64 val)
+{
+ u32 v;
+
+ asm(
+ "cmpsub.a %Q1, #1 @ f64_hi64to32jamming\n\t"
+ "cmovub %0, %R1\n\t"
+ "bub 100f\n\t"
+ "or %0, %R1, #1\n\t"
+"100: "
+ : "=r" (v) : "r" (val) : "cc");
+
+ return v;
+}
+
+static inline void mul64to128(u64 *resh, u64 *resl, u64 n, u64 m)
+{
+ u32 nh, nl, mh, ml;
+ u64 rh, rma, rmb, rl;
+
+ nl = n;
+ ml = m;
+ rl = (u64)nl * ml;
+
+ nh = n >> 32;
+ rma = (u64)nh * ml;
+
+ mh = m >> 32;
+ rmb = (u64)nl * mh;
+ rma += rmb;
+
+ rh = (u64)nh * mh;
+ rh += ((u64)(rma < rmb) << 32) + (rma >> 32);
+
+ rma <<= 32;
+ rl += rma;
+ rh += (rl < rma);
+
+ *resl = rl;
+ *resh = rh;
+}
+
+static inline void shift64left(u64 *resh, u64 *resl, u64 n)
+{
+ *resh = n >> 63;
+ *resl = n << 1;
+}
+
+static inline u64 f64_hi64multiply64(u64 n, u64 m)
+{
+ u64 rh, rl;
+ mul64to128(&rh, &rl, n, m);
+ return rh | (rl != 0);
+}
+
+/*
+ * Operations on unpacked elements
+ */
+#define f64_sign_negate(sign) (sign ^ 0x8000)
+
+/*
+ * Single-precision
+ */
+struct f64_single {
+ s16 exponent;
+ u16 sign;
+ u32 significand;
+};
+
+extern s32 f64_get_float(unsigned int reg);
+extern void f64_put_float(s32 val, unsigned int reg);
+
+/*
+ * F64_SINGLE_MANTISSA_BITS - number of bits in the mantissa
+ * F64_SINGLE_EXPONENT_BITS - number of bits in the exponent
+ * F64_SINGLE_LOW_BITS - number of low bits in the unpacked significand
+ * which are not propagated to the float upon packing.
+ */
+#define F64_SINGLE_MANTISSA_BITS (23)
+#define F64_SINGLE_EXPONENT_BITS (8)
+#define F64_SINGLE_LOW_BITS (32 - F64_SINGLE_MANTISSA_BITS - 2)
+#define F64_SINGLE_LOW_BITS_MASK ((1 << F64_SINGLE_LOW_BITS) - 1)
+
+/*
+ * The bit in an unpacked float which indicates that it is a quiet NaN
+ */
+#define F64_SINGLE_SIGNIFICAND_QNAN (1 << (F64_SINGLE_MANTISSA_BITS - 1 \
+ + F64_SINGLE_LOW_BITS))
+
+/*
+ * Operations on packed single-precision numbers
+ */
+#define f64_single_packed_sign(v) ((v) & 0x80000000)
+#define f64_single_packed_negate(v) ((v) ^ 0x80000000)
+#define f64_single_packed_abs(v) ((v) & ~0x80000000)
+#define f64_single_packed_exponent(v) (((v) >> F64_SINGLE_MANTISSA_BITS) \
+ & ((1 << F64_SINGLE_EXPONENT_BITS) \
+ - 1))
+#define f64_single_packed_mantissa(v) ((v) & \
+ ((1 << F64_SINGLE_MANTISSA_BITS) \
+ - 1))
+
+/*
+ * Unpack a single-precision float. Note that this returns the magnitude
+ * of the single-precision float mantissa with the 1. if necessary,
+ * aligned to bit 30.
+ */
+static inline void f64_single_unpack(struct f64_single *s, s32 val)
+{
+ u32 significand;
+
+ s->sign = f64_single_packed_sign(val) >> 16,
+ s->exponent = f64_single_packed_exponent(val);
+
+ significand = (u32) val;
+ significand = (significand << (32 - F64_SINGLE_MANTISSA_BITS)) >> 2;
+ if (s->exponent && s->exponent != 255)
+ significand |= 0x40000000;
+ s->significand = significand;
+}
+
+/*
+ * Re-pack a single-precision float. This assumes that the float is
+ * already normalised such that the MSB is bit 30, _not_ bit 31.
+ */
+static inline s32 f64_single_pack(struct f64_single *s)
+{
+ u32 val;
+ val = (s->sign << 16) +
+ (s->exponent << F64_SINGLE_MANTISSA_BITS) +
+ (s->significand >> F64_SINGLE_LOW_BITS);
+ return (s32)val;
+}
+
+#define F64_NUMBER (1<<0)
+#define F64_ZERO (1<<1)
+#define F64_DENORMAL (1<<2)
+#define F64_INFINITY (1<<3)
+#define F64_NAN (1<<4)
+#define F64_NAN_SIGNAL (1<<5)
+
+#define F64_QNAN (F64_NAN)
+#define F64_SNAN (F64_NAN|F64_NAN_SIGNAL)
+
+static inline int f64_single_type(struct f64_single *s)
+{
+ int type = F64_NUMBER;
+ if (s->exponent == 255) {
+ if (s->significand == 0)
+ type = F64_INFINITY;
+ else if (s->significand & F64_SINGLE_SIGNIFICAND_QNAN)
+ type = F64_QNAN;
+ else
+ type = F64_SNAN;
+ } else if (s->exponent == 0) {
+ if (s->significand == 0)
+ type |= F64_ZERO;
+ else
+ type |= F64_DENORMAL;
+ }
+ return type;
+}
+
+#ifndef DEBUG
+#define f64_single_normaliseround(sd, vsd, fpscr, except, func) \
+ __f64_single_normaliseround(sd, vsd, fpscr, except)
+u32 __f64_single_normaliseround(int sd, struct f64_single *vs, u32 fpscr,
+ u32 exceptions);
+#else
+u32 f64_single_normaliseround(int sd, struct f64_single *vs, u32 fpscr,
+ u32 exceptions, const char *func);
+#endif
+
+/*
+ * Double-precision
+ */
+struct f64_double {
+ s16 exponent;
+ u16 sign;
+ u64 significand;
+};
+
+extern u64 f64_get_double(unsigned int reg);
+extern void f64_put_double(u64 val, unsigned int reg);
+
+#define F64_DOUBLE_MANTISSA_BITS (52)
+#define F64_DOUBLE_EXPONENT_BITS (11)
+#define F64_DOUBLE_LOW_BITS (64 - F64_DOUBLE_MANTISSA_BITS - 2)
+#define F64_DOUBLE_LOW_BITS_MASK ((1 << F64_DOUBLE_LOW_BITS) - 1)
+
+/*
+ * The bit in an unpacked double which indicates that it is a quiet NaN
+ */
+#define F64_DOUBLE_SIGNIFICAND_QNAN (1ULL << (F64_DOUBLE_MANTISSA_BITS \
+ - 1 + F64_DOUBLE_LOW_BITS))
+
+/*
+ * Operations on packed single-precision numbers
+ */
+#define f64_double_packed_sign(v) ((v) & (1ULL << 63))
+#define f64_double_packed_negate(v) ((v) ^ (1ULL << 63))
+#define f64_double_packed_abs(v) ((v) & ~(1ULL << 63))
+#define f64_double_packed_exponent(v) (((v) >> F64_DOUBLE_MANTISSA_BITS) \
+ & ((1 << F64_DOUBLE_EXPONENT_BITS) \
+ - 1))
+#define f64_double_packed_mantissa(v) ((v) & \
+ ((1ULL << F64_DOUBLE_MANTISSA_BITS) \
+ - 1))
+
+/*
+ * Unpack a double-precision float. Note that this returns the magnitude
+ * of the double-precision float mantissa with the 1. if necessary,
+ * aligned to bit 62.
+ */
+static inline void f64_double_unpack(struct f64_double *s, s64 val)
+{
+ u64 significand;
+
+ s->sign = f64_double_packed_sign(val) >> 48;
+ s->exponent = f64_double_packed_exponent(val);
+
+ significand = (u64) val;
+ significand = (significand << (64 - F64_DOUBLE_MANTISSA_BITS)) >> 2;
+ if (s->exponent && s->exponent != 2047)
+ significand |= (1ULL << 62);
+ s->significand = significand;
+}
+
+/*
+ * Re-pack a double-precision float. This assumes that the float is
+ * already normalised such that the MSB is bit 30, _not_ bit 31.
+ */
+static inline s64 f64_double_pack(struct f64_double *s)
+{
+ u64 val;
+ val = ((u64)s->sign << 48) +
+ ((u64)s->exponent << F64_DOUBLE_MANTISSA_BITS) +
+ (s->significand >> F64_DOUBLE_LOW_BITS);
+ return (s64)val;
+}
+
+static inline int f64_double_type(struct f64_double *s)
+{
+ int type = F64_NUMBER;
+ if (s->exponent == 2047) {
+ if (s->significand == 0)
+ type = F64_INFINITY;
+ else if (s->significand & F64_DOUBLE_SIGNIFICAND_QNAN)
+ type = F64_QNAN;
+ else
+ type = F64_SNAN;
+ } else if (s->exponent == 0) {
+ if (s->significand == 0)
+ type |= F64_ZERO;
+ else
+ type |= F64_DENORMAL;
+ }
+ return type;
+}
+
+u32 f64_double_normaliseround(int dd, struct f64_double *vd, u32 fpscr,
+ u32 exceptions, const char *func);
+
+u32 f64_estimate_sqrt_significand(u32 exponent, u32 significand);
+
+/*
+ * A special flag to tell the normalisation code not to normalise.
+ */
+#define F64_NAN_FLAG 0x100
+
+/*
+ * A bit pattern used to indicate the initial (unset) value of the
+ * exception mask, in case nothing handles an instruction. This
+ * doesn't include the NAN flag, which get masked out before
+ * we check for an error.
+ */
+#define F64_EXCEPTION_ERROR ((u32)-1 & ~F64_NAN_FLAG)
+
+/*
+ * A flag to tell f64 instruction type.
+ * OP_SD - the instruction exceptionally writes to a single precision result.
+ * OP_DD - the instruction exceptionally writes to a double precision result.
+ */
+#define OP_SD (1 << 0)
+#define OP_DD (1 << 1)
+
+struct op {
+ u32 (* const fn)(int dd, int dn, int dm, u32 fpscr);
+ u32 flags;
+};
+
+extern void f64_save_state(void *location, u32 fpexc);
+
+/*
+ * Our undef handlers (in entry.S)
+ */
+extern void f64_support_entry(void);
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists