mirror/guile
mirror/guile
1
Fork 0
mirror of https://git.savannah.gnu.org/git/guile.git synced 2025-06-06 12:10:28 +02:00
Code Activity

Correct jit implementation to pass several test cases.

Browse source 
* include/lightning/jit_ia64.h, lib/jit_ia64-cpu.c,
	lib/jit_ia64-fpu.c, lib/jit_ia64.c:
	  Relocate JIT_Rn registers to the local registers, as, like
	float registers, div/rem and sqrt are implemented as function
	calls, and may overwrite non saved scratch registers.
	  Change patch_at to receive a jit_code_t instead of a
	jit_node_t, so that it is easier to "inline" patches when
	some instruction requires complex code to implement, e.g.
	uneq and ltgt.
	  Correct arguments to FMA and FMA like instructions that,
	due to a cut&paste error were passing the wrong argument
	to the related F- implementation function.
	  Rewrite ltgt to return the proper result if one (or both)
	of the arguments is unordered.
This commit is contained in:
pcpa 2013-04-26 21:07:40 -03:00
parent 89f1e2f608
commit 746f3bb6c7
5 changed files with 80 additions and 44 deletions
Download patch file Download diff file Expand all files Collapse all files

17
ChangeLog
View file

@ -1,3 +1,20 @@
2013-04-26 Paulo Andrade <pcpa@gnu.org>
* include/lightning/jit_ia64.h, lib/jit_ia64-cpu.c,
lib/jit_ia64-fpu.c, lib/jit_ia64.c:
Relocate JIT_Rn registers to the local registers, as, like
float registers, div/rem and sqrt are implemented as function
calls, and may overwrite non saved scratch registers.
Change patch_at to receive a jit_code_t instead of a
jit_node_t, so that it is easier to "inline" patches when
some instruction requires complex code to implement, e.g.
uneq and ltgt.
Correct arguments to FMA and FMA like instructions that,
due to a cut&paste error were passing the wrong argument
to the related F- implementation function.
Rewrite ltgt to return the proper result if one (or both)
of the arguments is unordered.
2013-04-26 Paulo Andrade <pcpa@gnu.org> 2013-04-26 Paulo Andrade <pcpa@gnu.org>
* include/lightning/jit_ia64.h, include/lightning/jit_private.h, * include/lightning/jit_ia64.h, include/lightning/jit_private.h,

22
include/lightning/jit_ia64.h
View file

@ -26,16 +26,18 @@
*/ */
#define JIT_FP _R4 /* Not abi specific */ #define JIT_FP _R4 /* Not abi specific */
typedef enum { typedef enum {
#define JIT_R0 _R14 #define JIT_R0 _R40
#define JIT_R1 _R15 #define JIT_R1 _R41
#define JIT_R2 _R16 #define JIT_R2 _R42
#define JIT_V0 _R40 #define JIT_R3 _R43
#define JIT_V1 _R41 #define JIT_V0 _R44
#define JIT_V2 _R42 #define JIT_V1 _R45
#define jit_r_num() 16 #define JIT_V2 _R46
#define jit_r(n) (_R14 - (n)) #define JIT_V3 _R47
#define jit_v_num() 8 #define jit_r_num() 4
#define jit_v(n) (_R40 + (n)) #define jit_r(n) (_R40 - (n))
#define jit_v_num() 4
#define jit_v(n) (_R44 + (n))
_R0, /* constant - Always 0 */ _R0, /* constant - Always 0 */
_R1, /* special - Global Data pointer (gp) */ _R1, /* special - Global Data pointer (gp) */
/* r2-r3 - scratch - Use with 22-bit immediate add - scratch */ /* r2-r3 - scratch - Use with 22-bit immediate add - scratch */

15
lib/jit_ia64-cpu.c
View file

@ -1423,7 +1423,7 @@ static void _prolog(jit_state_t*,jit_node_t*);
#define epilog(node) _epilog(_jit,node) #define epilog(node) _epilog(_jit,node)
static void _epilog(jit_state_t*,jit_node_t*); static void _epilog(jit_state_t*,jit_node_t*);
#define patch_at(node,instr,label) _patch_at(_jit,node,instr,label) #define patch_at(node,instr,label) _patch_at(_jit,node,instr,label)
static void _patch_at(jit_state_t*,jit_node_t*,jit_word_t,jit_word_t); static void _patch_at(jit_state_t*,jit_code_t,jit_word_t,jit_word_t);
#endif #endif
#if CODE #if CODE
@ -3414,7 +3414,7 @@ _subi(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
jit_int32_t reg; jit_int32_t reg;
reg = jit_get_reg(jit_class_gpr); reg = jit_get_reg(jit_class_gpr);
movi(rn(reg), i0); movi(rn(reg), i0);
addr(r0, r1, rn(reg)); subr(r0, r1, rn(reg));
jit_unget_reg(reg); jit_unget_reg(reg);
} }
@ -3607,8 +3607,7 @@ _remi_u(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_word_t i0)
return; return;
} }
else if (i0 > 0 && !(i0 & (i0 - 1))) { else if (i0 > 0 && !(i0 & (i0 - 1))) {
movr(r0, r1); andi(r0, r1, i0 - 1);
andi(r0, r0, i0 - 1);
return; return;
} }
reg = jit_get_reg(jit_class_gpr); reg = jit_get_reg(jit_class_gpr);
@ -4926,6 +4925,10 @@ _prolog(jit_state_t *_jit, jit_node_t *node)
} }
rout = reg - _OUT0; rout = reg - _OUT0;
/* Do not know if will call div/mod functions (sqrt) */
if (rout < 2)
rout = 2;
/* Match gcc prolog */ /* Match gcc prolog */
ALLOC(_jitc->breg + 1, ruse, rout); ALLOC(_jitc->breg + 1, ruse, rout);
MOV(_jitc->breg + 2, GR_12); MOV(_jitc->breg + 2, GR_12);
@ -4999,7 +5002,7 @@ _epilog(jit_state_t *_jit, jit_node_t *node)
} }
static void static void
_patch_at(jit_state_t *_jit, jit_node_t *node, _patch_at(jit_state_t *_jit, jit_code_t code,
jit_word_t instr, jit_word_t label) jit_word_t instr, jit_word_t label)
{ {
inst_lo_t l; inst_lo_t l;
@ -5013,7 +5016,7 @@ _patch_at(jit_state_t *_jit, jit_node_t *node,
c.w = instr; l.w = c.p[0]; h.w = c.p[1]; c.w = instr; l.w = c.p[0]; h.w = c.p[1];
get_tm(tm); get_s0(s0); get_s1(s1); get_s2(s2); get_tm(tm); get_s0(s0); get_s1(s1); get_s2(s2);
switch (node->code) { switch (code) {
case jit_code_movi: case jit_code_movi:
case jit_code_calli: case jit_code_calli:
i1 = (label >> 63) & 0x1L; i1 = (label >> 63) & 0x1L;

58
lib/jit_ia64-fpu.c
View file

@ -157,12 +157,12 @@ static void F16_(jit_state_t*,jit_word_t,
/* fcvt.fxuf */ /* fcvt.fxuf */
#define FCVT_XUF(f1,f3) FMA(f1,f3,1,0) #define FCVT_XUF(f1,f3) FMA(f1,f3,1,0)
/* fma */ /* fma */
#define FMA(f1,f3,f4,f2) F1(0x8,0,SF_S0,f4,f3,f3,f1) #define FMA(f1,f3,f4,f2) F1(0x8,0,SF_S0,f4,f3,f2,f1)
#define FMA_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0x8,0,sf,f4,f3,f3,f1) #define FMA_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0x8,0,sf,f4,f3,f2,f1)
#define FMA_S(f1,f3,f4,f2) F1(0x8,1,SF_S0,f4,f3,f3,f1) #define FMA_S(f1,f3,f4,f2) F1(0x8,1,SF_S0,f4,f3,f2,f1)
#define FMA_S_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0x8,1,sf,f4,f3,f3,f1) #define FMA_S_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0x8,1,sf,f4,f3,f2,f1)
#define FMA_D(f1,f3,f4,f2) F1(0x9,0,SF_S0,f4,f3,f3,f1) #define FMA_D(f1,f3,f4,f2) F1(0x9,0,SF_S0,f4,f3,f2,f1)
#define FMA_D_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0x9,0,sf,f4,f3,f3,f1) #define FMA_D_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0x9,0,sf,f4,f3,f2,f1)
/* fmax */ /* fmax */
#define FMAX(f1,f2,f3) F8(0,SF_S0,0x15,f3,f2,f1) #define FMAX(f1,f2,f3) F8(0,SF_S0,0x15,f3,f2,f1)
/* fmerge */ /* fmerge */
@ -183,18 +183,18 @@ static void F16_(jit_state_t*,jit_word_t,
#define FMPY_D(f1,f3,f4) FMA_D(f1,f3,f4,0) #define FMPY_D(f1,f3,f4) FMA_D(f1,f3,f4,0)
#define FMPY_D_p(f1,f3,f4,sf,_p) FMA_D_p(f1,f3,f4,0,sf,_p) #define FMPY_D_p(f1,f3,f4,sf,_p) FMA_D_p(f1,f3,f4,0,sf,_p)
/* fms */ /* fms */
#define FMS(f1,f3,f4,f2) F1(0xa,0,SF_S0,f4,f3,f3,f1) #define FMS(f1,f3,f4,f2) F1(0xa,0,SF_S0,f4,f3,f2,f1)
#define FMS_S(f1,f3,f4,f2) F1(0xa,1,SF_S0,f4,f3,f3,f1) #define FMS_S(f1,f3,f4,f2) F1(0xa,1,SF_S0,f4,f3,f2,f1)
#define FMS_D(f1,f3,f4,f2) F1(0xb,0,SF_S0,f4,f3,f3,f1) #define FMS_D(f1,f3,f4,f2) F1(0xb,0,SF_S0,f4,f3,f2,f1)
/* fneg */ /* fneg */
#define FNEG(f1,f3) FMERGE_NS(f1,f3,f3) #define FNEG(f1,f3) FMERGE_NS(f1,f3,f3)
/* fnegabs */ /* fnegabs */
#define FNEGABS(f1,f3) FMERGE_NS(f1,0,f3) #define FNEGABS(f1,f3) FMERGE_NS(f1,0,f3)
/* fnma */ /* fnma */
#define FNMA(f1,f3,f4,f2) F1(0xc,0,SF_S0,f4,f3,f3,f1) #define FNMA(f1,f3,f4,f2) F1(0xc,0,SF_S0,f4,f3,f2,f1)
#define FNMA_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0xc,0,sf,f4,f3,f3,f1) #define FNMA_p(f1,f3,f4,f2,sf,_p) F1_(_jit,_p,0xc,0,sf,f4,f3,f2,f1)
#define FNMA_S(f1,f3,f4,f2) F1(0xc,1,SF_S0,f4,f3,f3,f1) #define FNMA_S(f1,f3,f4,f2) F1(0xc,1,SF_S0,f4,f3,f2,f1)
#define FNMA_D(f1,f3,f4,f2) F1(0xd,0,SF_S0,f4,f3,f3,f1) #define FNMA_D(f1,f3,f4,f2) F1(0xd,0,SF_S0,f4,f3,f2,f1)
/* fnmpy */ /* fnmpy */
#define FNMPY(f1,f3,f4) FNMA(f1,f3,f4,0) #define FNMPY(f1,f3,f4) FNMA(f1,f3,f4,0)
/* fnorm */ /* fnorm */
@ -496,7 +496,7 @@ static void _gei_f(jit_state_t*,jit_int32_t,jit_int32_t,jit_float32_t*);
#define gei_d(r0,r1,i0) _gei_d(_jit,r0,r1,i0) #define gei_d(r0,r1,i0) _gei_d(_jit,r0,r1,i0)
static void _gei_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_float64_t*); static void _gei_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_float64_t*);
#define gtr_f(r0,r1,r2) gtr_d(r0,r1,r2) #define gtr_f(r0,r1,r2) gtr_d(r0,r1,r2)
#define gtr_d(r0,r1,r2) _gtr_d(_jit,r0,r2,r1) #define gtr_d(r0,r1,r2) _gtr_d(_jit,r0,r1,r2)
static void _gtr_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t); static void _gtr_d(jit_state_t*,jit_int32_t,jit_int32_t,jit_int32_t);
#define gti_f(r0,r1,i0) _gti_f(_jit,r0,r1,i0) #define gti_f(r0,r1,i0) _gti_f(_jit,r0,r1,i0)
static void _gti_f(jit_state_t*,jit_int32_t,jit_int32_t,jit_float32_t*); static void _gti_f(jit_state_t*,jit_int32_t,jit_int32_t,jit_float32_t*);
@ -1191,11 +1191,11 @@ dopi(ungt)
static void static void
_ltgtr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2) _ltgtr_d(jit_state_t *_jit, jit_int32_t r0, jit_int32_t r1, jit_int32_t r2)
{ {
MOV(r0, GR_0); MOVI(r0, 1);
FCMP_LE(PR_8, PR_9, r1, r2); FCMP_EQ(PR_8, PR_9, r1, r2);
FCMP_LE(PR_6, PR_7, r2, r1); FCMP_UNORD(PR_6, PR_7, r1, r2);
MOVI_p(r0, 1, PR_9); /* !(r1 >= r2) || !(r2 >= r1) */ MOV_p(r0, GR_0, PR_8);
MOVI_p(r0, 1, PR_7); MOV_p(r0, GR_0, PR_6);
} }
fopi(ltgt) fopi(ltgt)
dopi(ltgt) dopi(ltgt)
@ -1444,20 +1444,25 @@ static jit_word_t
_buneqr_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1) _buneqr_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
{ {
jit_word_t w; jit_word_t w;
jit_word_t junord, jne;
FCMP_UNORD(PR_8, PR_9, r0, r1); FCMP_UNORD(PR_8, PR_9, r0, r1);
/* junord L1 */
sync(); sync();
/* junord L1 */
junord = _jit->pc.w;
BRI_COND(3, PR_8); BRI_COND(3, PR_8);
FCMP_EQ(PR_6, PR_7, r0, r1); FCMP_EQ(PR_6, PR_7, r0, r1);
/* jne L2 */
sync(); sync();
/* jne L2 */
jne = _jit->pc.w;
BRI_COND(2, PR_7); BRI_COND(2, PR_7);
sync(); sync();
w = _jit->pc.w; w = _jit->pc.w;
/* L1: */ /* L1: */
patch_at(jit_code_bunordr_d, junord, _jit->pc.w);
BRI((i0 - w) >> 4); /* unconditional jump to patch */ BRI((i0 - w) >> 4); /* unconditional jump to patch */
sync(); sync();
/* L2: */ /* L2: */
patch_at(jit_code_bner_d, jne, _jit->pc.w);
return (w); return (w);
} }
fbopi(uneq) fbopi(uneq)
@ -1467,9 +1472,9 @@ static jit_word_t
_bunger_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1) _bunger_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
{ {
jit_word_t w; jit_word_t w;
FCMP_LT(PR_6, PR_7, r0, r1);
sync(); sync();
w = _jit->pc.w; w = _jit->pc.w;
FCMP_LT(PR_6, PR_7, r0, r1);
BRI_COND((i0 - w) >> 4, PR_7); BRI_COND((i0 - w) >> 4, PR_7);
return (w); return (w);
} }
@ -1493,19 +1498,24 @@ static jit_word_t
_bltgtr_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1) _bltgtr_d(jit_state_t *_jit, jit_word_t i0, jit_int32_t r0, jit_int32_t r1)
{ {
jit_word_t w; jit_word_t w;
jit_word_t jeq, junord;
FCMP_EQ(PR_8, PR_9, r0, r1); FCMP_EQ(PR_8, PR_9, r0, r1);
/* jeq L1 */ /* jeq L1 */
sync(); sync();
jeq = _jit->pc.w;
BRI_COND(4, PR_8); BRI_COND(4, PR_8);
FCMP_UNORD(PR_6, PR_7, r0, r1); FCMP_UNORD(PR_6, PR_7, r0, r1);
/* jord L1 */ /* junord L1 */
sync(); sync();
BRI_COND(2, PR_7); junord = _jit->pc.w;
BRI_COND(2, PR_6);
sync(); sync();
w = _jit->pc.w; w = _jit->pc.w;
BRI((i0 - w) >> 4); /* unconditional jump to patch */ BRI((i0 - w) >> 4); /* unconditional jump to patch */
/* L1 */ /* L1 */
sync(); sync();
patch_at(jit_code_beqr_d, jeq, _jit->pc.w);
patch_at(jit_code_bordr_d, junord, _jit->pc.w);
return (w); return (w);
} }
fbopi(ltgt) fbopi(ltgt)

12
lib/jit_ia64.c
View file

@ -67,7 +67,6 @@ jit_register_t _rvs[] = {
{ rc(gpr)|23, "r23" }, { rc(gpr)|22, "r22" }, { rc(gpr)|23, "r23" }, { rc(gpr)|22, "r22" },
{ rc(gpr)|21, "r21" }, { rc(gpr)|20, "r20" }, { rc(gpr)|21, "r21" }, { rc(gpr)|20, "r20" },
{ rc(gpr)|19, "r19" }, { rc(gpr)|18, "r18" }, { rc(gpr)|19, "r19" }, { rc(gpr)|18, "r18" },
/* JIT_R4-JIT_R0 */
{ rc(gpr)|17, "r17" }, { rc(gpr)|16, "r16" }, { rc(gpr)|17, "r17" }, { rc(gpr)|16, "r16" },
{ rc(gpr)|15, "r15" }, { rc(gpr)|14, "r14" }, { rc(gpr)|15, "r15" }, { rc(gpr)|14, "r14" },
/* Do not allow allocating r32-r41 as temoraries for the sake of /* Do not allow allocating r32-r41 as temoraries for the sake of
@ -76,12 +75,12 @@ jit_register_t _rvs[] = {
{ rc(arg)|34, "r34" }, { rc(arg)|35, "r35" }, { rc(arg)|34, "r34" }, { rc(arg)|35, "r35" },
{ rc(arg)|36, "r36" }, { rc(arg)|37, "r37" }, { rc(arg)|36, "r36" }, { rc(arg)|37, "r37" },
{ rc(arg)|38, "r38" }, { rc(arg)|39, "r39" }, { rc(arg)|38, "r38" }, { rc(arg)|39, "r39" },
/* JIT_V0-JIT_V3 */ /* JIT_R0-JIT_V3 */
{ rc(gpr)|40, "r40" }, { rc(gpr)|41, "r41" }, { rc(gpr)|40, "r40" }, { rc(gpr)|41, "r41" },
{ rc(gpr)|42, "r42" }, { rc(gpr)|43, "r43" }, { rc(gpr)|42, "r42" }, { rc(gpr)|43, "r43" },
/* Temporaries/locals */
{ rc(gpr)|44, "r44" }, { rc(gpr)|45, "r45" }, { rc(gpr)|44, "r44" }, { rc(gpr)|45, "r45" },
{ rc(gpr)|46, "r46" }, { rc(gpr)|47, "r47" }, { rc(gpr)|46, "r46" }, { rc(gpr)|47, "r47" },
/* Temporaries/locals */
{ rc(gpr)|48, "r48" }, { rc(gpr)|49, "r49" }, { rc(gpr)|48, "r48" }, { rc(gpr)|49, "r49" },
{ rc(gpr)|50, "r50" }, { rc(gpr)|51, "r51" }, { rc(gpr)|50, "r50" }, { rc(gpr)|51, "r51" },
{ rc(gpr)|52, "r52" }, { rc(gpr)|53, "r53" }, { rc(gpr)|52, "r52" }, { rc(gpr)|53, "r53" },
@ -648,6 +647,11 @@ _emit_code(jit_state_t *_jit)
_jitc->function = NULL; _jitc->function = NULL;
/* If did resize the code buffer, these were not reset */
_jitc->ioff = 0;
jit_regset_set_ui(&_jitc->regs, 0);
_jitc->pred = 0;
jit_reglive_setup(); jit_reglive_setup();
undo.word = 0; undo.word = 0;
@ -1229,7 +1233,7 @@ _emit_code(jit_state_t *_jit)
for (offset = 0; offset < _jitc->patches.offset; offset++) { for (offset = 0; offset < _jitc->patches.offset; offset++) {
node = _jitc->patches.ptr[offset].node; node = _jitc->patches.ptr[offset].node;
word = node->code == jit_code_movi ? node->v.n->u.w : node->u.n->u.w; word = node->code == jit_code_movi ? node->v.n->u.w : node->u.n->u.w;
patch_at(node, _jitc->patches.ptr[offset].inst, word); patch_at(node->code, _jitc->patches.ptr[offset].inst, word);
} }
word = sysconf(_SC_PAGE_SIZE); word = sysconf(_SC_PAGE_SIZE);