/* mips16 floating point support code
- Copyright (C) 1996, 1997, 1998 Free Software Foundation, Inc.
+ Copyright (C) 1996, 1997, 1998, 2008, 2009 Free Software Foundation, Inc.
Contributed by Cygnus Support
This file is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
-Free Software Foundation; either version 2, or (at your option) any
+Free Software Foundation; either version 3, or (at your option) any
later version.
-In addition to the permissions in the GNU General Public License, the
-Free Software Foundation gives you unlimited permission to link the
-compiled version of this file with other programs, and to distribute
-those programs without any restriction coming from the use of this
-file. (The General Public License restrictions do apply in other
-respects; for example, they cover modification of the file, and
-distribution when not linked into another program.)
-
This file is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
-You should have received a copy of the GNU General Public License
-along with this program; see the file COPYING. If not, write to
-the Free Software Foundation, 51 Franklin Street, Fifth Floor,
-Boston, MA 02110-1301, USA. */
+Under Section 7 of GPL version 3, you are granted additional
+permissions described in the GCC Runtime Library Exception, version
+3.1, as published by the Free Software Foundation.
-/* As a special exception, if you link this library with other files,
- some of which are compiled with GCC, to produce an executable,
- this library does not by itself cause the resulting executable
- to be covered by the GNU General Public License.
- This exception does not however invalidate any other reasons why
- the executable file might be covered by the GNU General Public License. */
+You should have received a copy of the GNU General Public License and
+a copy of the GCC Runtime Library Exception along with this program;
+see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+<http://www.gnu.org/licenses/>. */
/* This file contains mips16 floating point support functions. These
functions are called by mips16 code to handle floating point when
values using the soft-float calling convention, but do the actual
operation using the hard floating point instructions. */
+#if defined _MIPS_SIM && (_MIPS_SIM == _ABIO32 || _MIPS_SIM == _ABIO64)
+
/* This file contains 32-bit assembly code. */
.set nomips16
#define ENDFN(NAME) .end NAME
-/* Single precision math. */
+/* ARG1
+ The FPR that holds the first floating-point argument.
+
+ ARG2
+ The FPR that holds the second floating-point argument.
+
+ RET
+ The FPR that holds a floating-point return value. */
+
+#define RET $f0
+#define ARG1 $f12
+#ifdef __mips64
+#define ARG2 $f13
+#else
+#define ARG2 $f14
+#endif
+
+/* Set 64-bit register GPR so that its high 32 bits contain HIGH_FPR
+ and so that its low 32 bits contain LOW_FPR. */
+#define MERGE_GPRf(GPR, HIGH_FPR, LOW_FPR) \
+ .set noat; \
+ mfc1 GPR, HIGH_FPR; \
+ mfc1 $1, LOW_FPR; \
+ dsll GPR, GPR, 32; \
+ or GPR, GPR, $1; \
+ .set at
+
+/* Move the high 32 bits of GPR to HIGH_FPR and the low 32 bits of
+ GPR to LOW_FPR. */
+#define MERGE_GPRt(GPR, HIGH_FPR, LOW_FPR) \
+ .set noat; \
+ dsrl $1, GPR, 32; \
+ mtc1 GPR, LOW_FPR; \
+ mtc1 $1, HIGH_FPR; \
+ .set at
+
+/* Jump to T, and use "OPCODE, OP2" to implement a delayed move. */
+#define DELAYt(T, OPCODE, OP2) \
+ .set noreorder; \
+ jr T; \
+ OPCODE, OP2; \
+ .set reorder
+
+/* Use "OPCODE. OP2" and jump to T. */
+#define DELAYf(T, OPCODE, OP2) OPCODE, OP2; jr T
+
+/* MOVE_SF_BYTE0(D)
+ Move the first single-precision floating-point argument between
+ GPRs and FPRs.
+
+ MOVE_SI_BYTE0(D)
+ Likewise the first single-precision integer argument.
+
+ MOVE_SF_BYTE4(D)
+ Move the second single-precision floating-point argument between
+ GPRs and FPRs, given that the first argument occupies 4 bytes.
+
+ MOVE_SF_BYTE8(D)
+ Move the second single-precision floating-point argument between
+ GPRs and FPRs, given that the first argument occupies 8 bytes.
-/* This macro defines a function which loads two single precision
- values, performs an operation, and returns the single precision
- result. */
+ MOVE_DF_BYTE0(D)
+ Move the first double-precision floating-point argument between
+ GPRs and FPRs.
+
+ MOVE_DF_BYTE8(D)
+ Likewise the second double-precision floating-point argument.
+
+ MOVE_SF_RET(D, T)
+ Likewise a single-precision floating-point return value,
+ then jump to T.
+
+ MOVE_SC_RET(D, T)
+ Likewise a complex single-precision floating-point return value.
+
+ MOVE_DF_RET(D, T)
+ Likewise a double-precision floating-point return value.
+
+ MOVE_DC_RET(D, T)
+ Likewise a complex double-precision floating-point return value.
+
+ MOVE_SI_RET(D, T)
+ Likewise a single-precision integer return value.
+
+ The D argument is "t" to move to FPRs and "f" to move from FPRs.
+ The return macros may assume that the target of the jump does not
+ use a floating-point register. */
+
+#define MOVE_SF_RET(D, T) DELAY##D (T, m##D##c1 $2,$f0)
+#define MOVE_SI_RET(D, T) DELAY##D (T, m##D##c1 $2,$f0)
+
+#if defined(__mips64) && defined(__MIPSEB__)
+#define MOVE_SC_RET(D, T) MERGE_GPR##D ($2, $f0, $f1); jr T
+#elif defined(__mips64)
+/* The high 32 bits of $2 correspond to the second word in memory;
+ i.e. the imaginary part. */
+#define MOVE_SC_RET(D, T) MERGE_GPR##D ($2, $f1, $f0); jr T
+#elif __mips_fpr == 64
+#define MOVE_SC_RET(D, T) m##D##c1 $2,$f0; DELAY##D (T, m##D##c1 $3,$f1)
+#else
+#define MOVE_SC_RET(D, T) m##D##c1 $2,$f0; DELAY##D (T, m##D##c1 $3,$f2)
+#endif
-#define SFOP(NAME, OPCODE) \
+#if defined(__mips64)
+#define MOVE_SF_BYTE0(D) m##D##c1 $4,$f12
+#define MOVE_SF_BYTE4(D) m##D##c1 $5,$f13
+#define MOVE_SF_BYTE8(D) m##D##c1 $5,$f13
+#else
+#define MOVE_SF_BYTE0(D) m##D##c1 $4,$f12
+#define MOVE_SF_BYTE4(D) m##D##c1 $5,$f14
+#define MOVE_SF_BYTE8(D) m##D##c1 $6,$f14
+#endif
+#define MOVE_SI_BYTE0(D) MOVE_SF_BYTE0(D)
+
+#if defined(__mips64)
+#define MOVE_DF_BYTE0(D) dm##D##c1 $4,$f12
+#define MOVE_DF_BYTE8(D) dm##D##c1 $5,$f13
+#define MOVE_DF_RET(D, T) DELAY##D (T, dm##D##c1 $2,$f0)
+#define MOVE_DC_RET(D, T) dm##D##c1 $3,$f1; MOVE_DF_RET (D, T)
+#elif __mips_fpr == 64 && defined(__MIPSEB__)
+#define MOVE_DF_BYTE0(D) m##D##c1 $5,$f12; m##D##hc1 $4,$f12
+#define MOVE_DF_BYTE8(D) m##D##c1 $7,$f14; m##D##hc1 $6,$f14
+#define MOVE_DF_RET(D, T) m##D##c1 $3,$f0; DELAY##D (T, m##D##hc1 $2,$f0)
+#define MOVE_DC_RET(D, T) m##D##c1 $5,$f1; m##D##hc1 $4,$f1; MOVE_DF_RET (D, T)
+#elif __mips_fpr == 64
+#define MOVE_DF_BYTE0(D) m##D##c1 $4,$f12; m##D##hc1 $5,$f12
+#define MOVE_DF_BYTE8(D) m##D##c1 $6,$f14; m##D##hc1 $7,$f14
+#define MOVE_DF_RET(D, T) m##D##c1 $2,$f0; DELAY##D (T, m##D##hc1 $3,$f0)
+#define MOVE_DC_RET(D, T) m##D##c1 $4,$f1; m##D##hc1 $5,$f1; MOVE_DF_RET (D, T)
+#elif defined(__MIPSEB__)
+/* FPRs are little-endian. */
+#define MOVE_DF_BYTE0(D) m##D##c1 $4,$f13; m##D##c1 $5,$f12
+#define MOVE_DF_BYTE8(D) m##D##c1 $6,$f15; m##D##c1 $7,$f14
+#define MOVE_DF_RET(D, T) m##D##c1 $2,$f1; DELAY##D (T, m##D##c1 $3,$f0)
+#define MOVE_DC_RET(D, T) m##D##c1 $4,$f3; m##D##c1 $5,$f2; MOVE_DF_RET (D, T)
+#else
+#define MOVE_DF_BYTE0(D) m##D##c1 $4,$f12; m##D##c1 $5,$f13
+#define MOVE_DF_BYTE8(D) m##D##c1 $6,$f14; m##D##c1 $7,$f15
+#define MOVE_DF_RET(D, T) m##D##c1 $2,$f0; DELAY##D (T, m##D##c1 $3,$f1)
+#define MOVE_DC_RET(D, T) m##D##c1 $4,$f2; m##D##c1 $5,$f3; MOVE_DF_RET (D, T)
+#endif
+
+/* Single-precision math. */
+
+/* Define a function NAME that loads two single-precision values,
+ performs FPU operation OPCODE on them, and returns the single-
+ precision result. */
+
+#define OPSF3(NAME, OPCODE) \
STARTFN (NAME); \
- .set noreorder; \
- mtc1 $4,$f0; \
- mtc1 $5,$f2; \
- nop; \
- OPCODE $f0,$f0,$f2; \
- mfc1 $2,$f0; \
- j $31; \
- nop; \
- .set reorder; \
+ MOVE_SF_BYTE0 (t); \
+ MOVE_SF_BYTE4 (t); \
+ OPCODE RET,ARG1,ARG2; \
+ MOVE_SF_RET (f, $31); \
ENDFN (NAME)
#ifdef L_m16addsf3
-SFOP(__mips16_addsf3, add.s)
+OPSF3 (__mips16_addsf3, add.s)
#endif
#ifdef L_m16subsf3
-SFOP(__mips16_subsf3, sub.s)
+OPSF3 (__mips16_subsf3, sub.s)
#endif
#ifdef L_m16mulsf3
-SFOP(__mips16_mulsf3, mul.s)
+OPSF3 (__mips16_mulsf3, mul.s)
#endif
#ifdef L_m16divsf3
-SFOP(__mips16_divsf3, div.s)
+OPSF3 (__mips16_divsf3, div.s)
#endif
-#define SFOP2(NAME, OPCODE) \
+/* Define a function NAME that loads a single-precision value,
+ performs FPU operation OPCODE on it, and returns the single-
+ precision result. */
+
+#define OPSF2(NAME, OPCODE) \
STARTFN (NAME); \
- .set noreorder; \
- mtc1 $4,$f0; \
- nop; \
- OPCODE $f0,$f0; \
- mfc1 $2,$f0; \
- j $31; \
- nop; \
- .set reorder; \
+ MOVE_SF_BYTE0 (t); \
+ OPCODE RET,ARG1; \
+ MOVE_SF_RET (f, $31); \
ENDFN (NAME)
#ifdef L_m16negsf2
-SFOP2(__mips16_negsf2, neg.s)
+OPSF2 (__mips16_negsf2, neg.s)
#endif
#ifdef L_m16abssf2
-SFOP2(__mips16_abssf2, abs.s)
+OPSF2 (__mips16_abssf2, abs.s)
#endif
-/* Single precision comparisons. */
+/* Single-precision comparisons. */
-/* This macro defines a function which loads two single precision
- values, performs a floating point comparison, and returns the
- specified values according to whether the comparison is true or
- false. */
+/* Define a function NAME that loads two single-precision values,
+ performs floating point comparison OPCODE, and returns TRUE or
+ FALSE depending on the result. */
-#define SFCMP(NAME, OPCODE, TRUE, FALSE) \
+#define CMPSF(NAME, OPCODE, TRUE, FALSE) \
STARTFN (NAME); \
- mtc1 $4,$f0; \
- mtc1 $5,$f2; \
- OPCODE $f0,$f2; \
+ MOVE_SF_BYTE0 (t); \
+ MOVE_SF_BYTE4 (t); \
+ OPCODE ARG1,ARG2; \
li $2,TRUE; \
bc1t 1f; \
li $2,FALSE; \
j $31; \
ENDFN (NAME)
-/* This macro is like SFCMP, but it reverses the comparison. */
+/* Like CMPSF, but reverse the comparison operands. */
-#define SFREVCMP(NAME, OPCODE, TRUE, FALSE) \
+#define REVCMPSF(NAME, OPCODE, TRUE, FALSE) \
STARTFN (NAME); \
- mtc1 $4,$f0; \
- mtc1 $5,$f2; \
- OPCODE $f2,$f0; \
+ MOVE_SF_BYTE0 (t); \
+ MOVE_SF_BYTE4 (t); \
+ OPCODE ARG2,ARG1; \
li $2,TRUE; \
bc1t 1f; \
li $2,FALSE; \
ENDFN (NAME)
#ifdef L_m16eqsf2
-SFCMP(__mips16_eqsf2, c.eq.s, 0, 1)
+CMPSF (__mips16_eqsf2, c.eq.s, 0, 1)
#endif
#ifdef L_m16nesf2
-SFCMP(__mips16_nesf2, c.eq.s, 0, 1)
+CMPSF (__mips16_nesf2, c.eq.s, 0, 1)
#endif
#ifdef L_m16gtsf2
-SFREVCMP(__mips16_gtsf2, c.lt.s, 1, 0)
+REVCMPSF (__mips16_gtsf2, c.lt.s, 1, 0)
#endif
#ifdef L_m16gesf2
-SFREVCMP(__mips16_gesf2, c.le.s, 0, -1)
+REVCMPSF (__mips16_gesf2, c.le.s, 0, -1)
#endif
#ifdef L_m16lesf2
-SFCMP(__mips16_lesf2, c.le.s, 0, 1)
+CMPSF (__mips16_lesf2, c.le.s, 0, 1)
#endif
#ifdef L_m16ltsf2
-SFCMP(__mips16_ltsf2, c.lt.s, -1, 0)
+CMPSF (__mips16_ltsf2, c.lt.s, -1, 0)
+#endif
+#ifdef L_m16unordsf2
+CMPSF(__mips16_unordsf2, c.un.s, 1, 0)
#endif
-/* Single precision conversions. */
+
+/* Single-precision conversions. */
#ifdef L_m16fltsisf
STARTFN (__mips16_floatsisf)
- .set noreorder
- mtc1 $4,$f0
- nop
- cvt.s.w $f0,$f0
- mfc1 $2,$f0
- j $31
- nop
- .set reorder
+ MOVE_SF_BYTE0 (t)
+ cvt.s.w RET,ARG1
+ MOVE_SF_RET (f, $31)
ENDFN (__mips16_floatsisf)
#endif
-#ifdef L_m16fix_truncsfsi
-STARTFN (__mips16_fix_truncsfsi)
+#ifdef L_m16fltunsisf
+STARTFN (__mips16_floatunsisf)
.set noreorder
- mtc1 $4,$f0
- nop
- trunc.w.s $f0,$f0,$4
- mfc1 $2,$f0
- j $31
- nop
+ bltz $4,1f
+ MOVE_SF_BYTE0 (t)
.set reorder
+ cvt.s.w RET,ARG1
+ MOVE_SF_RET (f, $31)
+1:
+ and $2,$4,1
+ srl $3,$4,1
+ or $2,$2,$3
+ mtc1 $2,RET
+ cvt.s.w RET,RET
+ add.s RET,RET,RET
+ MOVE_SF_RET (f, $31)
+ ENDFN (__mips16_floatunsisf)
+#endif
+
+#ifdef L_m16fix_truncsfsi
+STARTFN (__mips16_fix_truncsfsi)
+ MOVE_SF_BYTE0 (t)
+ trunc.w.s RET,ARG1,$4
+ MOVE_SI_RET (f, $31)
ENDFN (__mips16_fix_truncsfsi)
#endif
#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
-/* The double precision operations. We need to use different code
- based on the preprocessor symbol __mips64, because the way in which
- double precision values will change. Without __mips64, the value
- is passed in two 32-bit registers. With __mips64, the value is
- passed in a single 64-bit register. */
+/* Double-precision math. */
-/* Load the first double precision operand. */
+/* Define a function NAME that loads two double-precision values,
+ performs FPU operation OPCODE on them, and returns the double-
+ precision result. */
-#if defined(__mips64)
-#define LDDBL1 dmtc1 $4,$f12
-#elif defined(__mipsfp64)
-#define LDDBL1 sw $4,0($29); sw $5,4($29); l.d $f12,0($29)
-#elif defined(__MIPSEB__)
-#define LDDBL1 mtc1 $4,$f13; mtc1 $5,$f12
-#else
-#define LDDBL1 mtc1 $4,$f12; mtc1 $5,$f13
-#endif
-
-/* Load the second double precision operand. */
-
-#if defined(__mips64)
-/* XXX this should be $6 for Algo arg passing model */
-#define LDDBL2 dmtc1 $5,$f14
-#elif defined(__mipsfp64)
-#define LDDBL2 sw $6,8($29); sw $7,12($29); l.d $f14,8($29)
-#elif defined(__MIPSEB__)
-#define LDDBL2 mtc1 $6,$f15; mtc1 $7,$f14
-#else
-#define LDDBL2 mtc1 $6,$f14; mtc1 $7,$f15
-#endif
-
-/* Move the double precision return value to the right place. */
-
-#if defined(__mips64)
-#define RETDBL dmfc1 $2,$f0
-#elif defined(__mipsfp64)
-#define RETDBL s.d $f0,0($29); lw $2,0($29); lw $3,4($29)
-#elif defined(__MIPSEB__)
-#define RETDBL mfc1 $2,$f1; mfc1 $3,$f0
-#else
-#define RETDBL mfc1 $2,$f0; mfc1 $3,$f1
-#endif
-
-/* Double precision math. */
-
-/* This macro defines a function which loads two double precision
- values, performs an operation, and returns the double precision
- result. */
-
-#define DFOP(NAME, OPCODE) \
+#define OPDF3(NAME, OPCODE) \
STARTFN (NAME); \
- .set noreorder; \
- LDDBL1; \
- LDDBL2; \
- nop; \
- OPCODE $f0,$f12,$f14; \
- RETDBL; \
- j $31; \
- nop; \
- .set reorder; \
+ MOVE_DF_BYTE0 (t); \
+ MOVE_DF_BYTE8 (t); \
+ OPCODE RET,ARG1,ARG2; \
+ MOVE_DF_RET (f, $31); \
ENDFN (NAME)
#ifdef L_m16adddf3
-DFOP(__mips16_adddf3, add.d)
+OPDF3 (__mips16_adddf3, add.d)
#endif
#ifdef L_m16subdf3
-DFOP(__mips16_subdf3, sub.d)
+OPDF3 (__mips16_subdf3, sub.d)
#endif
#ifdef L_m16muldf3
-DFOP(__mips16_muldf3, mul.d)
+OPDF3 (__mips16_muldf3, mul.d)
#endif
#ifdef L_m16divdf3
-DFOP(__mips16_divdf3, div.d)
+OPDF3 (__mips16_divdf3, div.d)
#endif
-#define DFOP2(NAME, OPCODE) \
+/* Define a function NAME that loads a double-precision value,
+ performs FPU operation OPCODE on it, and returns the double-
+ precision result. */
+
+#define OPDF2(NAME, OPCODE) \
STARTFN (NAME); \
- .set noreorder; \
- LDDBL1; \
- nop; \
- OPCODE $f0,$f12; \
- RETDBL; \
- j $31; \
- nop; \
- .set reorder; \
+ MOVE_DF_BYTE0 (t); \
+ OPCODE RET,ARG1; \
+ MOVE_DF_RET (f, $31); \
ENDFN (NAME)
#ifdef L_m16negdf2
-DFOP2(__mips16_negdf2, neg.d)
+OPDF2 (__mips16_negdf2, neg.d)
#endif
#ifdef L_m16absdf2
-DFOP2(__mips16_absdf2, abs.d)
+OPDF2 (__mips16_absdf2, abs.d)
#endif
-
/* Conversions between single and double precision. */
#ifdef L_m16extsfdf2
STARTFN (__mips16_extendsfdf2)
- .set noreorder
- mtc1 $4,$f12
- nop
- cvt.d.s $f0,$f12
- RETDBL
- j $31
- nop
- .set reorder
+ MOVE_SF_BYTE0 (t)
+ cvt.d.s RET,ARG1
+ MOVE_DF_RET (f, $31)
ENDFN (__mips16_extendsfdf2)
#endif
#ifdef L_m16trdfsf2
STARTFN (__mips16_truncdfsf2)
- .set noreorder
- LDDBL1
- nop
- cvt.s.d $f0,$f12
- mfc1 $2,$f0
- j $31
- nop
- .set reorder
+ MOVE_DF_BYTE0 (t)
+ cvt.s.d RET,ARG1
+ MOVE_SF_RET (f, $31)
ENDFN (__mips16_truncdfsf2)
#endif
-/* Double precision comparisons. */
+/* Double-precision comparisons. */
-/* This macro defines a function which loads two double precision
- values, performs a floating point comparison, and returns the
- specified values according to whether the comparison is true or
- false. */
+/* Define a function NAME that loads two double-precision values,
+ performs floating point comparison OPCODE, and returns TRUE or
+ FALSE depending on the result. */
-#define DFCMP(NAME, OPCODE, TRUE, FALSE) \
+#define CMPDF(NAME, OPCODE, TRUE, FALSE) \
STARTFN (NAME); \
- LDDBL1; \
- LDDBL2; \
- OPCODE $f12,$f14; \
+ MOVE_DF_BYTE0 (t); \
+ MOVE_DF_BYTE8 (t); \
+ OPCODE ARG1,ARG2; \
li $2,TRUE; \
bc1t 1f; \
li $2,FALSE; \
j $31; \
ENDFN (NAME)
-/* This macro is like DFCMP, but it reverses the comparison. */
+/* Like CMPDF, but reverse the comparison operands. */
-#define DFREVCMP(NAME, OPCODE, TRUE, FALSE) \
+#define REVCMPDF(NAME, OPCODE, TRUE, FALSE) \
STARTFN (NAME); \
- LDDBL1; \
- LDDBL2; \
- OPCODE $f14,$f12; \
+ MOVE_DF_BYTE0 (t); \
+ MOVE_DF_BYTE8 (t); \
+ OPCODE ARG2,ARG1; \
li $2,TRUE; \
bc1t 1f; \
li $2,FALSE; \
ENDFN (NAME)
#ifdef L_m16eqdf2
-DFCMP(__mips16_eqdf2, c.eq.d, 0, 1)
+CMPDF (__mips16_eqdf2, c.eq.d, 0, 1)
#endif
#ifdef L_m16nedf2
-DFCMP(__mips16_nedf2, c.eq.d, 0, 1)
+CMPDF (__mips16_nedf2, c.eq.d, 0, 1)
#endif
#ifdef L_m16gtdf2
-DFREVCMP(__mips16_gtdf2, c.lt.d, 1, 0)
+REVCMPDF (__mips16_gtdf2, c.lt.d, 1, 0)
#endif
#ifdef L_m16gedf2
-DFREVCMP(__mips16_gedf2, c.le.d, 0, -1)
+REVCMPDF (__mips16_gedf2, c.le.d, 0, -1)
#endif
#ifdef L_m16ledf2
-DFCMP(__mips16_ledf2, c.le.d, 0, 1)
+CMPDF (__mips16_ledf2, c.le.d, 0, 1)
#endif
#ifdef L_m16ltdf2
-DFCMP(__mips16_ltdf2, c.lt.d, -1, 0)
+CMPDF (__mips16_ltdf2, c.lt.d, -1, 0)
+#endif
+#ifdef L_m16unorddf2
+CMPDF(__mips16_unorddf2, c.un.d, 1, 0)
#endif
-/* Double precision conversions. */
+/* Double-precision conversions. */
#ifdef L_m16fltsidf
STARTFN (__mips16_floatsidf)
- .set noreorder
- mtc1 $4,$f12
- nop
- cvt.d.w $f0,$f12
- RETDBL
- j $31
- nop
- .set reorder
+ MOVE_SI_BYTE0 (t)
+ cvt.d.w RET,ARG1
+ MOVE_DF_RET (f, $31)
ENDFN (__mips16_floatsidf)
#endif
-
+
+#ifdef L_m16fltunsidf
+STARTFN (__mips16_floatunsidf)
+ MOVE_SI_BYTE0 (t)
+ cvt.d.w RET,ARG1
+ bgez $4,1f
+ li.d ARG1, 4.294967296e+9
+ add.d RET, RET, ARG1
+1: MOVE_DF_RET (f, $31)
+ ENDFN (__mips16_floatunsidf)
+#endif
+
#ifdef L_m16fix_truncdfsi
STARTFN (__mips16_fix_truncdfsi)
- .set noreorder
- LDDBL1
- nop
- trunc.w.d $f0,$f12,$4
- mfc1 $2,$f0
- j $31
- nop
- .set reorder
+ MOVE_DF_BYTE0 (t)
+ trunc.w.d RET,ARG1,$4
+ MOVE_SI_RET (f, $31)
ENDFN (__mips16_fix_truncdfsi)
#endif
#endif /* !__mips_single_float */
-/* These functions are used to return floating point values from
- mips16 functions. In this case we can put mtc1 in a jump delay slot,
- because we know that the next instruction will not refer to a floating
- point register. */
+/* Define a function NAME that moves a return value of mode MODE from
+ FPRs to GPRs. */
+
+#define RET_FUNCTION(NAME, MODE) \
+STARTFN (NAME); \
+ MOVE_##MODE##_RET (t, $31); \
+ ENDFN (NAME)
#ifdef L_m16retsf
-STARTFN (__mips16_ret_sf)
- .set noreorder
- j $31
- mtc1 $2,$f0
- .set reorder
- ENDFN (__mips16_ret_sf)
+RET_FUNCTION (__mips16_ret_sf, SF)
+#endif
+
+#ifdef L_m16retsc
+RET_FUNCTION (__mips16_ret_sc, SC)
#endif
#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
#ifdef L_m16retdf
-STARTFN (__mips16_ret_df)
- .set noreorder
-#if defined(__mips64)
- j $31
- dmtc1 $2,$f0
-#elif defined(__mipsfp64)
- sw $2,0($29)
- sw $3,4($29)
- l.d $f0,0($29)
-#elif defined(__MIPSEB__)
- mtc1 $2,$f1
- j $31
- mtc1 $3,$f0
-#else
- mtc1 $2,$f0
- j $31
- mtc1 $3,$f1
+RET_FUNCTION (__mips16_ret_df, DF)
#endif
- .set reorder
- ENDFN (__mips16_ret_df)
+
+#ifdef L_m16retdc
+RET_FUNCTION (__mips16_ret_dc, DC)
#endif
#endif /* !__mips_single_float */
+/* STUB_ARGS_X copies the arguments from GPRs to FPRs for argument
+ code X. X is calculated as ARG1 + ARG2 * 4, where ARG1 and ARG2
+ classify the first and second arguments as follows:
+
+ 1: a single-precision argument
+ 2: a double-precision argument
+ 0: no argument, or not one of the above. */
+
+#define STUB_ARGS_0 /* () */
+#define STUB_ARGS_1 MOVE_SF_BYTE0 (t) /* (sf) */
+#define STUB_ARGS_5 MOVE_SF_BYTE0 (t); MOVE_SF_BYTE4 (t) /* (sf, sf) */
+#define STUB_ARGS_9 MOVE_SF_BYTE0 (t); MOVE_DF_BYTE8 (t) /* (sf, df) */
+#define STUB_ARGS_2 MOVE_DF_BYTE0 (t) /* (df) */
+#define STUB_ARGS_6 MOVE_DF_BYTE0 (t); MOVE_SF_BYTE8 (t) /* (df, sf) */
+#define STUB_ARGS_10 MOVE_DF_BYTE0 (t); MOVE_DF_BYTE8 (t) /* (df, df) */
+
/* These functions are used by 16-bit code when calling via a function
- pointer. They must copy the floating point arguments from the gp
- regs into the fp regs. The function to call will be in $2. The
- exact set of floating point arguments to copy is encoded in the
- function name; the final number is an fp_code, as described in
- mips.h in the comment about CUMULATIVE_ARGS. */
+ pointer. They must copy the floating point arguments from the GPRs
+ to FPRs and then call function $2. */
+
+#define CALL_STUB_NO_RET(NAME, CODE) \
+STARTFN (NAME); \
+ STUB_ARGS_##CODE; \
+ .set noreorder; \
+ jr $2; \
+ move $25,$2; \
+ .set reorder; \
+ ENDFN (NAME)
#ifdef L_m16stub1
-/* (float) */
-STARTFN (__mips16_call_stub_1)
- .set noreorder
- mtc1 $4,$f12
- j $2
- nop
- .set reorder
- ENDFN (__mips16_call_stub_1)
+CALL_STUB_NO_RET (__mips16_call_stub_1, 1)
#endif
#ifdef L_m16stub5
-/* (float, float) */
-STARTFN (__mips16_call_stub_5)
- .set noreorder
- mtc1 $4,$f12
- mtc1 $5,$f14
- j $2
- nop
- .set reorder
- ENDFN (__mips16_call_stub_5)
+CALL_STUB_NO_RET (__mips16_call_stub_5, 5)
#endif
#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
#ifdef L_m16stub2
-/* (double) */
-STARTFN (__mips16_call_stub_2)
- .set noreorder
- LDDBL1
- j $2
- nop
- .set reorder
- ENDFN (__mips16_call_stub_2)
+CALL_STUB_NO_RET (__mips16_call_stub_2, 2)
#endif
#ifdef L_m16stub6
-/* (double, float) */
-STARTFN (__mips16_call_stub_6)
- .set noreorder
- LDDBL1
- mtc1 $6,$f14
- j $2
- nop
- .set reorder
- ENDFN (__mips16_call_stub_6)
+CALL_STUB_NO_RET (__mips16_call_stub_6, 6)
#endif
#ifdef L_m16stub9
-/* (float, double) */
-STARTFN (__mips16_call_stub_9)
- .set noreorder
- mtc1 $4,$f12
- LDDBL2
- j $2
- nop
- .set reorder
- ENDFN (__mips16_call_stub_9)
+CALL_STUB_NO_RET (__mips16_call_stub_9, 9)
#endif
#ifdef L_m16stub10
-/* (double, double) */
-STARTFN (__mips16_call_stub_10)
- .set noreorder
- LDDBL1
- LDDBL2
- j $2
- nop
- .set reorder
- ENDFN (__mips16_call_stub_10)
+CALL_STUB_NO_RET (__mips16_call_stub_10, 10)
#endif
#endif /* !__mips_single_float */
/* Now we have the same set of functions, except that this time the
- function being called returns an SFmode value. The calling
+ function being called returns an SFmode, SCmode, DFmode or DCmode
+ value; we need to instantiate a set for each case. The calling
function will arrange to preserve $18, so these functions are free
to use it to hold the return address.
being called is 16 bits, in which case the copy is unnecessary;
however, it's faster to always do the copy. */
+#define CALL_STUB_RET(NAME, CODE, MODE) \
+STARTFN (NAME); \
+ move $18,$31; \
+ STUB_ARGS_##CODE; \
+ .set noreorder; \
+ jalr $2; \
+ move $25,$2; \
+ .set reorder; \
+ MOVE_##MODE##_RET (f, $18); \
+ ENDFN (NAME)
+
+/* First, instantiate the single-float set. */
+
#ifdef L_m16stubsf0
-/* () */
-STARTFN (__mips16_call_stub_sf_0)
- .set noreorder
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_0)
+CALL_STUB_RET (__mips16_call_stub_sf_0, 0, SF)
#endif
#ifdef L_m16stubsf1
-/* (float) */
-STARTFN (__mips16_call_stub_sf_1)
- .set noreorder
- mtc1 $4,$f12
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_1)
+CALL_STUB_RET (__mips16_call_stub_sf_1, 1, SF)
#endif
#ifdef L_m16stubsf5
-/* (float, float) */
-STARTFN (__mips16_call_stub_sf_5)
- .set noreorder
- mtc1 $4,$f12
- mtc1 $5,$f14
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_5)
+CALL_STUB_RET (__mips16_call_stub_sf_5, 5, SF)
#endif
#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
#ifdef L_m16stubsf2
-/* (double) */
-STARTFN (__mips16_call_stub_sf_2)
- .set noreorder
- LDDBL1
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_2)
+CALL_STUB_RET (__mips16_call_stub_sf_2, 2, SF)
#endif
#ifdef L_m16stubsf6
-/* (double, float) */
-STARTFN (__mips16_call_stub_sf_6)
- .set noreorder
- LDDBL1
- mtc1 $6,$f14
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_6)
+CALL_STUB_RET (__mips16_call_stub_sf_6, 6, SF)
#endif
#ifdef L_m16stubsf9
-/* (float, double) */
-STARTFN (__mips16_call_stub_sf_9)
- .set noreorder
- mtc1 $4,$f12
- LDDBL2
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_9)
+CALL_STUB_RET (__mips16_call_stub_sf_9, 9, SF)
#endif
#ifdef L_m16stubsf10
-/* (double, double) */
-STARTFN (__mips16_call_stub_sf_10)
- .set noreorder
- LDDBL1
- LDDBL2
- move $18,$31
- jal $2
- nop
- mfc1 $2,$f0
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_sf_10)
+CALL_STUB_RET (__mips16_call_stub_sf_10, 10, SF)
#endif
+#endif /* !__mips_single_float */
+
/* Now we have the same set of functions again, except that this time
the function being called returns an DFmode value. */
+#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
#ifdef L_m16stubdf0
-/* () */
-STARTFN (__mips16_call_stub_df_0)
- .set noreorder
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_0)
+CALL_STUB_RET (__mips16_call_stub_df_0, 0, DF)
#endif
#ifdef L_m16stubdf1
-/* (float) */
-STARTFN (__mips16_call_stub_df_1)
- .set noreorder
- mtc1 $4,$f12
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_1)
+CALL_STUB_RET (__mips16_call_stub_df_1, 1, DF)
#endif
-#ifdef L_m16stubdf2
-/* (double) */
-STARTFN (__mips16_call_stub_df_2)
- .set noreorder
- LDDBL1
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_2)
+#ifdef L_m16stubdf5
+CALL_STUB_RET (__mips16_call_stub_df_5, 5, DF)
#endif
-#ifdef L_m16stubdf5
-/* (float, float) */
-STARTFN (__mips16_call_stub_df_5)
- .set noreorder
- mtc1 $4,$f12
- mtc1 $5,$f14
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_5)
+#ifdef L_m16stubdf2
+CALL_STUB_RET (__mips16_call_stub_df_2, 2, DF)
#endif
#ifdef L_m16stubdf6
-/* (double, float) */
-STARTFN (__mips16_call_stub_df_6)
- .set noreorder
- LDDBL1
- mtc1 $6,$f14
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_6)
+CALL_STUB_RET (__mips16_call_stub_df_6, 6, DF)
#endif
#ifdef L_m16stubdf9
-/* (float, double) */
-STARTFN (__mips16_call_stub_df_9)
- .set noreorder
- mtc1 $4,$f12
- LDDBL2
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_9)
+CALL_STUB_RET (__mips16_call_stub_df_9, 9, DF)
#endif
#ifdef L_m16stubdf10
-/* (double, double) */
-STARTFN (__mips16_call_stub_df_10)
- .set noreorder
- LDDBL1
- LDDBL2
- move $18,$31
- jal $2
- nop
- RETDBL
- j $18
- nop
- .set reorder
- ENDFN (__mips16_call_stub_df_10)
+CALL_STUB_RET (__mips16_call_stub_df_10, 10, DF)
+#endif
+#endif /* !__mips_single_float */
+
+
+/* Ho hum. Here we have the same set of functions again, this time
+ for when the function being called returns an SCmode value. */
+
+#ifdef L_m16stubsc0
+CALL_STUB_RET (__mips16_call_stub_sc_0, 0, SC)
+#endif
+
+#ifdef L_m16stubsc1
+CALL_STUB_RET (__mips16_call_stub_sc_1, 1, SC)
+#endif
+
+#ifdef L_m16stubsc5
+CALL_STUB_RET (__mips16_call_stub_sc_5, 5, SC)
+#endif
+
+#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
+#ifdef L_m16stubsc2
+CALL_STUB_RET (__mips16_call_stub_sc_2, 2, SC)
+#endif
+
+#ifdef L_m16stubsc6
+CALL_STUB_RET (__mips16_call_stub_sc_6, 6, SC)
+#endif
+
+#ifdef L_m16stubsc9
+CALL_STUB_RET (__mips16_call_stub_sc_9, 9, SC)
+#endif
+
+#ifdef L_m16stubsc10
+CALL_STUB_RET (__mips16_call_stub_sc_10, 10, SC)
+#endif
+#endif /* !__mips_single_float */
+
+
+/* Finally, another set of functions for DCmode. */
+
+#if !defined(__mips_single_float) && !defined(__SINGLE_FLOAT)
+#ifdef L_m16stubdc0
+CALL_STUB_RET (__mips16_call_stub_dc_0, 0, DC)
+#endif
+
+#ifdef L_m16stubdc1
+CALL_STUB_RET (__mips16_call_stub_dc_1, 1, DC)
+#endif
+
+#ifdef L_m16stubdc5
+CALL_STUB_RET (__mips16_call_stub_dc_5, 5, DC)
+#endif
+
+#ifdef L_m16stubdc2
+CALL_STUB_RET (__mips16_call_stub_dc_2, 2, DC)
+#endif
+
+#ifdef L_m16stubdc6
+CALL_STUB_RET (__mips16_call_stub_dc_6, 6, DC)
+#endif
+
+#ifdef L_m16stubdc9
+CALL_STUB_RET (__mips16_call_stub_dc_9, 9, DC)
+#endif
+
+#ifdef L_m16stubdc10
+CALL_STUB_RET (__mips16_call_stub_dc_10, 10, DC)
#endif
#endif /* !__mips_single_float */
+#endif