Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[DYNAREC] Fix x87cache issues #1025

Merged
merged 6 commits into from
Oct 14, 2023
Merged

[DYNAREC] Fix x87cache issues #1025

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
ce2d6d9
[DYNAREC][WIP] Fix x87cache issues
ksco Oct 13, 2023
62031bc
Check x87cache is full before push, is it working?
ksco Oct 14, 2023
cc328d6
Fixed a typo
ksco Oct 14, 2023
1580791
Update dynarec_arm64_helper.h
ksco Oct 14, 2023
95764af
Review
ksco Oct 14, 2023
59d7d54
Update test25
ksco Oct 14, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1001,6 +1001,11 @@ add_test(bswap ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
-D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref24.txt
-P ${CMAKE_SOURCE_DIR}/runTest.cmake )

add_test(x87cache ${CMAKE_COMMAND} -D TEST_PROGRAM=${CMAKE_BINARY_DIR}/${BOX64}
-D TEST_ARGS=${CMAKE_SOURCE_DIR}/tests/test25 -D TEST_OUTPUT=tmpfile25.txt
-D TEST_REFERENCE=${CMAKE_SOURCE_DIR}/tests/ref25.txt
-P ${CMAKE_SOURCE_DIR}/runTest.cmake )


file(GLOB extension_tests "${CMAKE_SOURCE_DIR}/tests/extensions/*.c")
foreach(file ${extension_tests})
Expand Down
6 changes: 3 additions & 3 deletions src/dynarec/arm64/dynarec_arm64_d8.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -108,7 +108,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FCMPD(v1, v2);
}
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xE0:
case 0xE1:
Expand Down Expand Up @@ -178,7 +178,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FDIVD(v1, v2, v1);
}
break;

default:
switch((nextop>>3)&7) {
case 0:
Expand Down Expand Up @@ -234,7 +234,7 @@ uintptr_t dynarec64_D8(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FCMPD(v1, s0);
}
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 4:
INST_NAME("FSUB ST0, float[ED]");
Expand Down
40 changes: 20 additions & 20 deletions src/dynarec/arm64/dynarec_arm64_d9.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,7 +51,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
case 0xC6:
case 0xC7:
INST_NAME("FLD STx");
v2 = x87_do_push(dyn, ninst, x1, X87_ST(nextop&7));
X87_PUSH_OR_FAIL(v2, dyn, ninst, x1, X87_ST(nextop&7));
v1 = x87_get_st(dyn, ninst, x1, x2, (nextop&7)+1, X87_COMBINE(0, (nextop&7)+1));
if(ST_IS_F(0)) {
FMOVS(v2, v1);
Expand Down Expand Up @@ -84,7 +84,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin

case 0xD8:
INST_NAME("FSTPNCE ST0, ST0");
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xD9:
case 0xDA:
Expand All @@ -98,7 +98,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xE0:
INST_NAME("FCHS");
Expand Down Expand Up @@ -138,7 +138,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin

case 0xE8:
INST_NAME("FLD1");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
if(ST_IS_F(0)) {
FMOVS_8(v1, 0b01110000);
} else {
Expand All @@ -147,32 +147,32 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
break;
case 0xE9:
INST_NAME("FLDL2T");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
FTABLE64(v1, L2T);
break;
case 0xEA:
case 0xEA:
INST_NAME("FLDL2E");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
FTABLE64(v1, L2E);
break;
case 0xEB:
INST_NAME("FLDPI");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
FTABLE64(v1, PI);
break;
case 0xEC:
INST_NAME("FLDLG2");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
FTABLE64(v1, LG2);
break;
case 0xED:
INST_NAME("FLDLN2");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
FTABLE64(v1, LN2);
break;
case 0xEE:
INST_NAME("FLDZ");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
VEOR(v1, v1, v1);
break;

Expand All @@ -188,14 +188,14 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
x87_forget(dyn, ninst, x1, x2, 0);
x87_forget(dyn, ninst, x1, x2, 1);
CALL(native_fyl2x, -1);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xF2:
INST_NAME("FPTAN");
MESSAGE(LOG_DUMP, "Need Optimization\n");
x87_forget(dyn, ninst, x1, x2, 0);
CALL(native_ftan, -1);
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_F);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_F);
if(ST_IS_F(0)) {
FMOVS_8(v1, 0b01110000);
} else {
Expand All @@ -208,12 +208,12 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
x87_forget(dyn, ninst, x1, x2, 0);
x87_forget(dyn, ninst, x1, x2, 1);
CALL(native_fpatan, -1);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xF4:
INST_NAME("FXTRACT");
MESSAGE(LOG_DUMP, "Need Optimization\n");
x87_do_push_empty(dyn, ninst, 0);
X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
x87_forget(dyn, ninst, x1, x2, 1);
CALL(native_fxtract, -1);
break;
Expand Down Expand Up @@ -253,7 +253,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
x87_forget(dyn, ninst, x1, x2, 0);
x87_forget(dyn, ninst, x1, x2, 1);
CALL(native_fyl2xp1, -1);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xFA:
INST_NAME("FSQRT");
Expand All @@ -267,7 +267,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
case 0xFB:
INST_NAME("FSINCOS");
MESSAGE(LOG_DUMP, "Need Optimization\n");
x87_do_push_empty(dyn, ninst, 0);
X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, 0);
x87_forget(dyn, ninst, x1, x2, 1);
CALL(native_fsincos, -1);
break;
Expand Down Expand Up @@ -322,12 +322,12 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
case 0xEF:
DEFAULT;
break;

default:
switch((nextop>>3)&7) {
case 0:
INST_NAME("FLD ST0, float[ED]");
v1 = x87_do_push(dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, box64_dynarec_x87double?NEON_CACHE_ST_D:NEON_CACHE_ST_F);
addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
VLD32(v1, ed, fixedaddress);
if(!ST_IS_F(0)) {
Expand All @@ -354,7 +354,7 @@ uintptr_t dynarec64_D9(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FCVT_S_D(v1, v1);
}
VST32(v1, ed, fixedaddress);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 4:
INST_NAME("FLDENV Ed");
Expand Down
8 changes: 4 additions & 4 deletions src/dynarec/arm64/dynarec_arm64_da.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -119,7 +119,7 @@ uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
} else {
FCSELD(v1, v2, v1, cNE); // F_PF==0
}
break;
break;
case 0xE9:
INST_NAME("FUCOMPP ST0, ST1");
v1 = x87_get_st(dyn, ninst, x1, x2, 0, X87_COMBINE(0, nextop&7));
Expand All @@ -130,8 +130,8 @@ uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FCMPD(v1, v2);
}
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;

case 0xE4:
Expand Down Expand Up @@ -190,7 +190,7 @@ uintptr_t dynarec64_DA(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
SCVTFDD(v2, v2); // i64 -> double
FCMPD(v1, v2);
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 4:
INST_NAME("FISUB ST0, Ed");
Expand Down
16 changes: 8 additions & 8 deletions src/dynarec/arm64/dynarec_arm64_db.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -156,7 +156,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
}
FCOMI(x1, x2);
break;
case 0xF0:
case 0xF0:
case 0xF1:
case 0xF2:
case 0xF3:
Expand Down Expand Up @@ -188,7 +188,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
switch((nextop>>3)&7) {
case 0:
INST_NAME("FILD ST0, Ed");
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
addr = geted(dyn, addr, ninst, nextop, &ed, x2, &fixedaddress, &unscaled, 0xfff<<2, 3, rex, NULL, 0, 0);
VLD32(v1, ed, fixedaddress);
SXTL_32(v1, v1); // i32 -> i64
Expand Down Expand Up @@ -218,7 +218,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
STW(x5, wback, fixedaddress);
MARK3;
#endif
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 2:
INST_NAME("FIST Ed, ST0");
Expand Down Expand Up @@ -273,7 +273,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
MARK3;
#endif
x87_restoreround(dyn, ninst, u8);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 5:
INST_NAME("FLD tbyte");
Expand All @@ -296,13 +296,13 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
STRH_U12(x6, ed, 8);
} else {
if(box64_x87_no80bits) {
v1 = x87_do_push(dyn, ninst, x1, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x1, NEON_CACHE_ST_D);
VLDR64_U12(v1, ed, fixedaddress);
} else {
if(ed!=x1) {
MOVx_REG(x1, ed);
}
x87_do_push_empty(dyn, ninst, x3);
X87_PUSH_EMPTY_OR_FAIL(dyn, ninst, x3);
CALL(native_fld, -1);
}
}
Expand Down Expand Up @@ -344,7 +344,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
CMPSw_REG(x3, x4);
B_MARK2(cNE);
// NaN and Infinite
ORRw_mask(x3, x5, 0, 0b1110); //x3 = sign | 0x7fff
ORRw_mask(x3, x5, 0, 0b1110); //x3 = sign | 0x7fff
TSTx_mask(x1, 1, 0, 0b110011); //0x000fffffffffffffL
ORRx_mask(x5, xZR, 1, 1, 0); //0x8000000000000000
ORRx_mask(x4, xZR, 1, 0b10, 0b01); //0xc000000000000000
Expand Down Expand Up @@ -372,7 +372,7 @@ uintptr_t dynarec64_DB(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
STRH_U12(x3, wback, 8);
#endif
}
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
default:
DEFAULT;
Expand Down
6 changes: 3 additions & 3 deletions src/dynarec/arm64/dynarec_arm64_dc.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,7 +106,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FCMPD(v1, v2);
}
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xE0:
case 0xE1:
Expand Down Expand Up @@ -158,7 +158,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
} else {
FDIVD(v1, v2, v1);
}
break;
break;
case 0xF8:
case 0xF9:
case 0xFA:
Expand Down Expand Up @@ -211,7 +211,7 @@ uintptr_t dynarec64_DC(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
VLD64(v2, wback, fixedaddress);
FCMPD(v1, v2);
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 4:
INST_NAME("FSUB ST0, double[ED]");
Expand Down
16 changes: 8 additions & 8 deletions src/dynarec/arm64/dynarec_arm64_dd.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
break;
case 0xD8:
INST_NAME("FSTP ST0, ST0");
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 0xD9:
case 0xDA:
Expand All @@ -88,7 +88,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
x87_get_st_empty(dyn, ninst, x1, x2, nextop&7, X87_ST(nextop&7));
x87_get_st(dyn, ninst, x1, x2, 0, X87_ST0);
x87_swapreg(dyn, ninst, x1, x2, 0, nextop&7);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;

case 0xE0:
Expand Down Expand Up @@ -126,7 +126,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
FCMPD(v1, v2);
}
FCOM(x1, x2, x3);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;

case 0xC8:
Expand Down Expand Up @@ -160,7 +160,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
switch((nextop>>3)&7) {
case 0:
INST_NAME("FLD double");
v1 = x87_do_push(dyn, ninst, x3, NEON_CACHE_ST_D);
X87_PUSH_OR_FAIL(v1, dyn, ninst, x3, NEON_CACHE_ST_D);
addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
VLD64(v1, ed, fixedaddress);
break;
Expand All @@ -187,7 +187,7 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
STx(x5, ed, fixedaddress);
MARK3;
#endif
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 2:
INST_NAME("FST double");
Expand All @@ -200,17 +200,17 @@ uintptr_t dynarec64_DD(dynarec_arm_t* dyn, uintptr_t addr, uintptr_t ip, int nin
v1 = x87_get_st(dyn, ninst, x1, x2, 0, NEON_CACHE_ST_D);
addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, &unscaled, 0xfff<<3, 7, rex, NULL, 0, 0);
VST64(v1, ed, fixedaddress);
x87_do_pop(dyn, ninst, x3);
X87_POP_OR_FAIL(dyn, ninst, x3);
break;
case 4:
case 4:
INST_NAME("FRSTOR m108byte");
MESSAGE(LOG_DUMP, "Need Optimization\n");
fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
addr = geted(dyn, addr, ninst, nextop, &ed, x1, &fixedaddress, NULL, 0, 0, rex, NULL, 0, 0);
if(ed!=x1) {MOVx_REG(x1, ed);}
CALL(native_frstor, -1);
break;
case 6:
case 6:
INST_NAME("FSAVE m108byte");
MESSAGE(LOG_DUMP, "Need Optimization\n");
fpu_purgecache(dyn, ninst, 0, x1, x2, x3);
Expand Down
Loading