ed/pikuma_ps1
1
0
Fork 0
mirror of https://github.com/Ed94/pikuma_ps1.git synced 2026-08-25 02:20:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add RegUse_ support to the lua metaprogram. Ideated further on type mapping atom comonents to their base component op (math distinctions annotated in the asm).

Browse Source
This commit is contained in:
ed
2026-08-15 15:51:06 -04:00
parent f8b28be02e
commit 8a825a59c7
13 changed files with 551 additions and 317 deletions
Download Patch File Download Diff File Expand all files Collapse all files
code/duffle/dsl.h
+10 -9
View File
@@ -140,16 +140,17 @@ enum { false = 0, true = 1, true_overflow, };
typedef void Proc_(VoidFn) (void);
#define Kilo_(n) (C_(U4, n) << 10)
#define Mega_(n) (C_(U4, n) << 20)
#define Giga_(n) (C_(U4, n) << 30)
#define Tera_(n) (C_(U4, n) << 40)
#define Kilo_(n) (C_(U4, n) << 10)
#define Mega_(n) (C_(U4, n) << 20)
#define Giga_(n) (C_(U4, n) << 30)
#define Tera_(n) (C_(U4, n) << 40)
#define null C_(U4, 0)
#define nullptr C_(void*, 0)
#define O_(type, field) C_(U4, & C_(type*,0)->field)
#define OT_(field) O_(typeof_ptr(& field), field))
#define S_(data) C_(U4, sizeof(data))
#define null C_(U4, 0)
#define nullptr C_(void*, 0)
#define O_(type, field) C_(U4, & C_(type*,0)->field)
#define OA_(type, aexpr) C_(U4, & C_(type*,0) aexpr)
#define OT_(field) O_(typeof_ptr(& field), field))
#define S_(data) C_(U4, sizeof(data))
#define sop_1(op,a,b) C_(U1, s1_(a) op s1_(b))
#define sop_2(op,a,b) C_(U2, s2_(a) op s2_(b))
code/duffle/gen/macs.h
+27 -11
View File
@@ -17,7 +17,7 @@
// source: C:\projects\Pikuma\ps1\code\duffle\bios.h
// source: C:\projects\Pikuma\ps1\code\duffle\psyq.h
// source: C:\projects\Pikuma\ps1\code\duffle\pad.c
// source: C:\projects\Pikuma\ps1\code\duffle\math.atom.c
// source: C:\projects\Pikuma\ps1\code\duffle\math.atom.h
// source: C:\projects\Pikuma\ps1\code\duffle\mips.atom.c
// source: C:\projects\Pikuma\ps1\code\duffle\gte.atom.c
// source: C:\projects\Pikuma\ps1\code\duffle\gp.atom.c
@@ -71,19 +71,35 @@ WORD_COUNT(mac_load_v2s2, 2)
WORD_COUNT(mac_store_v2s2, 2)
/* atom_dbg_skip */
#define mac_load_v3s4(rs_x, rs_y, rs_z, r_base, offset) \
load_word( rs_x, r_base, offset + O_(V3_S4,x)) \
, load_word( rs_y, r_base, offset + O_(V3_S4,y)) \
, load_word( rs_z, r_base, offset + O_(V3_S4,z))
#define mac_load_word_v3(tx, ty, tz, base, offset) \
load_word(tx, base, offset + OA_(U4,[0])) \
, load_word(ty, base, offset + OA_(U4,[1])) \
, load_word(tz, base, offset + OA_(U4,[2]))
WORD_COUNT(mac_load_word_v3, 3)
#define mac_load_v3s4(transfer, base, offset) \
mac_load_word_v3(transfer.x, transfer.y, transfer.z, base, offset)
WORD_COUNT(mac_load_v3s4, 3)
#define mac_load_p3s4(transfer, base, offset) \
mac_load_word_v3(transfer.x, transfer.y, transfer.z, base, offset)
WORD_COUNT(mac_load_p3s4, 3)
/* atom_dbg_skip */
#define mac_store_v3s4(rt_x, rt_y, rt_z, base, offset) \
store_word(rt_x, base, offset + O_(V3_S4,x)) \
, store_word(rt_y, base, offset + O_(V3_S4,y)) \
, store_word(rt_z, base, offset + O_(V3_S4,z))
#define mac_store_word_v3(tx, ty, tz, base, offset) \
store_word(tx, base, offset + OA_(U4,[0])) \
, store_word(ty, base, offset + OA_(U4,[1])) \
, store_word(tz, base, offset + OA_(U4,[2]))
WORD_COUNT(mac_store_word_v3, 3)
#define mac_store_v3s4(transfer, base, offset) \
mac_store_word_v3(transfer.x, transfer.y, transfer.z, base, offset)
WORD_COUNT(mac_store_v3s4, 3)
#define mac_store_p3s4(transfer, base, offset) \
mac_store_word_v3(transfer.x, transfer.y, transfer.z, base, offset)
WORD_COUNT(mac_store_p3s4, 3)
/* atom_dbg_skip */
#define mac_add_si_v3s4(rt_x, rt_y, rt_z, base, offset) \
add_si(rt_x, base, O_(V3_S4,x)) \
@@ -234,8 +250,8 @@ WORD_COUNT(mac_lzcr_round_even_half_shift, 5)
, gte_mv_to_data_r(to_ir1, C2_IR1) /* IR1 = src.x (preserved in r_tmp — r_mac2_scratch was clobbered to MAC2 in stage 1.5) */ \
, gte_mv_to_data_r(to_ir2, C2_IR2) \
, gte_mv_to_data_r(to_ir3, C2_IR3) /* IR3 = src.z (reloaded) */ \
, LdSlot_ nop_slot1 \
, LdSlot_ nop_slot2 \
, DmaSlot_ nop_slot1 \
, DmaSlot_ nop_slot2 \
, gte_cmdw_gpf \
, gte_mv_from_data_r(fr_mac1, C2_MAC1) \
, gte_mv_from_data_r(fr_mac2, C2_MAC2) \
code/duffle/gen/offsets.h
+1 -1
View File
@@ -14,7 +14,7 @@
// source: C:\projects\Pikuma\ps1\code\duffle\bios.h
// source: C:\projects\Pikuma\ps1\code\duffle\psyq.h
// source: C:\projects\Pikuma\ps1\code\duffle\pad.c
// source: C:\projects\Pikuma\ps1\code\duffle\math.atom.c
// source: C:\projects\Pikuma\ps1\code\duffle\math.atom.h
// source: C:\projects\Pikuma\ps1\code\duffle\mips.atom.c
// source: C:\projects\Pikuma\ps1\code\duffle\gte.atom.c
// source: C:\projects\Pikuma\ps1\code\duffle\gp.atom.c
code/duffle/gte.atom.c
+8 -8
View File
@@ -150,8 +150,8 @@ MipsAtomComp_Proc_(ab, {
gte_mv_to_data_r(to_ir1, C2_IR1), /* IR1 = src.x (preserved in r_tmp — r_mac2_scratch was clobbered to MAC2 in stage 1.5) */
gte_mv_to_data_r(to_ir2, C2_IR2),
gte_mv_to_data_r(to_ir3, C2_IR3), /* IR3 = src.z (reloaded) */
LdSlot_ nop_slot1,
LdSlot_ nop_slot2,
DmaSlot_ nop_slot1,
DmaSlot_ nop_slot2,
gte_cmdw_gpf,
gte_mv_from_data_r(fr_mac1, C2_MAC1),
gte_mv_from_data_r(fr_mac2, C2_MAC2),
@@ -284,7 +284,7 @@ MipsAtom_Proc_(aa, {
/* Load src.x/y/z from r_src_ptr (caller-determined address) into r_tmp/r_recip_est/r_branch_tmp.
* r.rt1_src_x holds src.x throughout stages 1-2 — r_mac2_scratch is clobbered to MAC2 in stage 1.5 (line below). */
mac_load_v3s4(r.src_x, r.recip_est, r.t5.src_z, r.src_ptr, 0),
mac_load_word_v3(r.src_x, r.recip_est, r.t5.src_z, r.src_ptr, 0),
/* Stage 1: mtc2 src → IR1/2/3, SQR fires. */
LdSlot_ mac_gte_sqr_v3s4(r.src_x, r.recip_est, r.t5.src_z, LdSlot_ nop),
@@ -293,8 +293,8 @@ MipsAtom_Proc_(aa, {
mac_gte_mv_from_data_r_mac123(r.t3.mac1_scratch, r.t4.mac2_scratch, r.norm), LdSlot_ nop,
add_u_self( r.norm, r.t3.mac1_scratch),
add_u_self( r.norm, r.t4.mac2_scratch),
gte_mv_to_data_r( r.norm, C2_LZCS), LdSlot_ nop2,
gte_mv_from_data_r(r.shift, C2_LZCR), LdSlot_ nop,
gte_mv_to_data_r( r.norm, C2_LZCS), DmaSlot_ nop2,
gte_mv_from_data_r(r.shift, C2_LZCR), DmaSlot_ nop,
/* Stage 3: round LZCR to even, compute half-shift, align |v|² to bit 24.
* r_norm holds |v|² sum; r_shift holds the LZCR count from mfc2.
@@ -328,13 +328,13 @@ MipsAtom_Proc_(aa, {
r.recip_est,
r.t5.src_z, /* IR3 = src.z (reloaded) */
r.t4.mac2_scratch, r.recip_est, r.t5.src_z,
LdSlot_ add_si(r.dst_ptr, r.scratch, dst_offset), // pre-laoding destination to register here.
LdSlot_ nop
DmaSlot_ add_si(r.dst_ptr, r.scratch, dst_offset), // pre-laoding destination to register here.
DmaSlot_ nop
),
/* sra by r_shift = (31-LZCR)/2 (saved before sqrtbl lookup) */
mac_shift_aright_var_v3_self(r.t4.mac2_scratch, r.recip_est, r.t5.src_z, r.shift),
/* Store result.x/y/z to r_dst_ptr (caller-determined dst address). */
mac_store_v3s4(r.t4.mac2_scratch, r.recip_est, r.t5.src_z, r.dst_ptr, 0),
mac_store_word_v3(r.t4.mac2_scratch, r.recip_est, r.t5.src_z, r.dst_ptr, 0),
mac_yield()
})
code/duffle/lottes_tape.h
+4 -1
View File
@@ -101,6 +101,7 @@ enum {
};
typedef U2 Reg; // Register parameter used with atom or atom component procedures
#define Reg_(type) tmpl(Reg,type) // Just a way to template register allocations of C-struct types.
typedef U4 const MipsCode; // Underlying type to mips asm words.
typedef Slice_(MipsCode);
@@ -143,6 +144,8 @@ typedef Slice_(MipsAtom);
// Inline-only callers (the generated `mac_<name>` aliases) skip the `ab` arg via metaprogram filtering; escape callers (ac_<name> invoked as a function) pass a long-lived builder.
#define MipsAtomComp_Proc_(ab, ...) { MipsCode atom_comp_code[] align_(4) = __VA_ARGS__; atombuilder_push(ab, slice_from_array(MipsCode, atom_comp_code)); }
#define MipsAtomComp_ProcMap_(ab, base_command) atom_dbg_skip MipsAtomComp_Proc_(ab, {base_command })
/* Line-table anchor: gcc only adds a file to the .debug_line file table when the contains line-numbered content.
Files containing only atoms and atom components.
Place `ATOM_FILE_LINE_MARKER();` once at file scope in any `.atom.c` that defines atoms.
@@ -372,7 +375,7 @@ FI_ void regfile_reset(RegFile_R rf) {
rf->GPR[0] = u4_lo(regfile_abi_mask);
rf->GPR[1] = u4_hi(regfile_abi_mask);
}
FI_ void regfile_reset_mask(RegFile_R rf, U4 mask) {
FI_ void regfile_reset_to_mask(RegFile_R rf, U4 mask) {
rf->GPR[0] = u4_lo(mask);
rf->GPR[1] = u4_hi(mask);
}
code/duffle/math.atom.c → code/duffle/math.atom.h
+17 -16
View File
@@ -7,9 +7,10 @@
ATOM_FILE_DEBUGGER_LINE_MARKER(math_atom_c);
#pragma region MACs (Mips Atom Component)
typedef Struct_(Reg_V3_S4) { Reg x, y, z; }; // Register allocation of a V3_S4
typedef Struct_(Reg_P3_S4) { Reg x, y, z; }; // Register allocation of a P3_S4
// FI_ Slice_MipsCode ac_load_imm
#pragma region MACs (Mips Atom Component)
FI_ Slice_MipsCode ac_load_v2s2(AtomBuilder_R ab, U4 rs_x, U4 rs_y, U4 r_base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ab, {
load_half( rs_x, r_base, offset + O_(V3_S2,x)),
@@ -21,23 +22,23 @@ FI_ Slice_MipsCode ac_store_v2s2(AtomBuilder_R ab, U4 rt_x, U4 rt_y, U4 base, U4
store_half(rt_y, base, offset + O_(V2_S2,y)),
})
FI_ Slice_MipsCode ac_load_v3s4(AtomBuilder_R ab, U4 rs_x, U4 rs_y, U4 rs_z, U4 r_base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ab, {
load_word( rs_x, r_base, offset + O_(V3_S4,x)),
load_word( rs_y, r_base, offset + O_(V3_S4,y)),
load_word( rs_z, r_base, offset + O_(V3_S4,z)),
FI_ Slice_MipsCode ac_load_word_v3(AtomBuilder_R ab, Reg tx, Reg ty, Reg tz, Reg base, U2 offset) atom_dbg_skip MipsAtomComp_Proc_(ab, {
load_word(tx, base, offset + OA_(U4,[0])),
load_word(ty, base, offset + OA_(U4,[1])),
load_word(tz, base, offset + OA_(U4,[2])),
})
// TODO(Ed): we could generate these mappings properly..
#define ac_load_p3s4 ac_load_v3s4
#define mac_load_p3s4 mac_load_v3s4
FI_ Slice_MipsCode ac_store_v3s4(AtomBuilder_R ab, U4 rt_x, U4 rt_y, U4 rt_z, U4 base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ab, {
store_word(rt_x, base, offset + O_(V3_S4,x)),
store_word(rt_y, base, offset + O_(V3_S4,y)),
store_word(rt_z, base, offset + O_(V3_S4,z)),
FI_ Slice_MipsCode ac_load_v3s4(AtomBuilder_R ab, Reg_(V3_S4) transfer, Reg base, U2 offset) MipsAtomComp_ProcMap_(ab, mac_load_word_v3(transfer.x, transfer.y, transfer.z, base, offset))
FI_ Slice_MipsCode ac_load_p3s4(AtomBuilder_R ab, Reg_(P3_S4) transfer, Reg base, U2 offset) MipsAtomComp_ProcMap_(ab, mac_load_word_v3(transfer.x, transfer.y, transfer.z, base, offset))
FI_ Slice_MipsCode ac_store_word_v3(AtomBuilder_R ab, U4 tx, U4 ty, U4 tz, U4 base, U4 offset) atom_dbg_skip MipsAtomComp_Proc_(ab, {
store_word(tx, base, offset + OA_(U4,[0])),
store_word(ty, base, offset + OA_(U4,[1])),
store_word(tz, base, offset + OA_(U4,[2])),
})
// TODO(Ed): we could generate these mappings properly..
#define ac_store_p3s4 ac_store_v3s4
#define mac_store_p3s4 mac_store_v3s4
FI_ Slice_MipsCode ac_store_v3s4(AtomBuilder_R ab, Reg_(V3_S4) transfer, Reg base, U2 offset) MipsAtomComp_ProcMap_(ab, mac_store_word_v3(transfer.x, transfer.y, transfer.z, base, offset))
FI_ Slice_MipsCode ac_store_p3s4(AtomBuilder_R ab, Reg_(P3_S4) transfer, Reg base, U2 offset) MipsAtomComp_ProcMap_(ab, mac_store_word_v3(transfer.x, transfer.y, transfer.z, base, offset))
FI_ Slice_MipsCode ac_add_si_v3s4(AtomBuilder_R ab, Reg rt_x, Reg rt_y, Reg rt_z, Reg base, U2 offset)
atom_dbg_skip MipsAtomComp_Proc_(ab, {