ed/pikuma_ps1
1
0
Fork 0
mirror of https://github.com/Ed94/pikuma_ps1.git synced 2026-08-14 11:38:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

fixes to the reg file allocator, exploring...

Browse Source
This commit is contained in:
ed
2026-08-14 00:43:19 -04:00
parent d88e0d0487
commit c447bfa877
4 changed files with 126 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files
code/duffle/dsl.h
+5 -5
View File
@@ -218,16 +218,16 @@ def_signed_ops(le, <=)
typedef Span_(S4); typedef Span_(S4);
typedef Span_(U4); typedef Span_(U4);
#if 0
#pragma region Debug #pragma region Debug
#define debug_trap() __builtin_debugtrap() #define debug_trap() __builtin_trap()
#if BUILD_DEBUG #if BUILD_DEBUG
IA_ void assert(U8 cond) { if(cond){return;} else{debug_trap(); ms_exit_process(1);} } #define assert(cond) if(cond == false){debug_trap();}
#else #else
#define assert(cond) # ifndef assert
# include <assert.h>
# endif
#endif #endif
#pragma endregion Debug #pragma endregion Debug
#endif
#define GCC_OPTIMIZATION_DISABLE _Pragma("GCC push_options") _Pragma("GCC optimize(\"O0\")") #define GCC_OPTIMIZATION_DISABLE _Pragma("GCC push_options") _Pragma("GCC optimize(\"O0\")")
#define GCC_OPTIMIZATION_ENABLE _Pragma("GCC pop_options") #define GCC_OPTIMIZATION_ENABLE _Pragma("GCC pop_options")
code/duffle/lottes_tape.h
+22 -12
View File
@@ -307,11 +307,7 @@ typedef Struct_(RegFile) {
A2_U2 GPR; A2_U2 GPR;
A2_U2 GTE; A2_U2 GTE;
}; };
typedef Struct_(RegFile_RInfo) { #define regfile(pin_mask) {.GPR={u4_lo(pin_mask), u4_hi(pin_mask)} }
U2_R section;
U2 mask;
B2 occupied;
};
FI_ void regfile_init(RegFile_R rf) { FI_ void regfile_init(RegFile_R rf) {
/* pack the 32-bit ABI mask into the two U2s */ /* pack the 32-bit ABI mask into the two U2s */
rf->GPR[0] = u4_lo(regfile_abi_mask); rf->GPR[0] = u4_lo(regfile_abi_mask);
@@ -320,11 +316,16 @@ FI_ void regfile_init(RegFile_R rf) {
} }
FI_ RegFile regfile_make(void) { RegFile rf; regfile_init(& rf); return rf; } FI_ RegFile regfile_make(void) { RegFile rf; regfile_init(& rf); return rf; }
typedef Struct_(RegFile_RInfo) {
U2_R section;
U2 mask;
B2 occupied;
};
FI_ RegFile_RInfo regfile_rinfo(A2_U2 file, Reg r_id) { FI_ RegFile_RInfo regfile_rinfo(A2_U2 file, Reg r_id) {
U2 s_id = r_id >> 4; U2 s_id = r_id >> 4;
U2_R section = & file[s_id]; U2_R section = & file[s_id];
U2 mask = (1u << (r_id & 15)); U2 mask = u2_(1u << (r_id & 15));
B2 occupied = section[0] & mask != 0; B2 occupied = (section[0] & mask) != 0;
return (RegFile_RInfo){section, mask, occupied}; return (RegFile_RInfo){section, mask, occupied};
} }
FI_ Reg regfile__alloc_helper(A2_U2 file, Reg r_id) { FI_ Reg regfile__alloc_helper(A2_U2 file, Reg r_id) {
@@ -337,12 +338,12 @@ FI_ Reg regfile__alloc_helper(A2_U2 file, Reg r_id) {
} }
I_ Reg regfile_alloc(RegFile_R rf) { I_ Reg regfile_alloc(RegFile_R rf) {
U2 allocated = 0; U2 allocated = 0;
for range_iter(r_id, <=, r1u2(R_T0, R_T7)) { for index_iter(Reg, r_id, R_T0, <=, R_T7) {
allocated = regfile__alloc_helper(rf->GPR, r_id); Jmp_nZero_(allocated,resolved);
}
for range_iter(r_id, <=, r1u2(R_V0, R_V1)) {
allocated = regfile__alloc_helper(rf->GPR, r_id); Jmp_nZero_(allocated,resolved); allocated = regfile__alloc_helper(rf->GPR, r_id); Jmp_nZero_(allocated,resolved);
} }
allocated = regfile__alloc_helper(rf->GPR, R_V0); Jmp_nZero_(allocated,resolved);
allocated = regfile__alloc_helper(rf->GPR, R_V1);
assert(allocated != 0);
resolved: return allocated; resolved: return allocated;
} }
FI_ Reg regfile_pin(RegFile_R rf, Reg r_id) { FI_ Reg regfile_pin(RegFile_R rf, Reg r_id) {
@@ -351,9 +352,14 @@ FI_ Reg regfile_pin(RegFile_R rf, Reg r_id) {
info.section[0] |= info.mask; info.section[0] |= info.mask;
return r_id; return r_id;
} }
FI_ void regfile_pin_mask(RegFile_R rf, U4 mask) {
B4 occupied = u4_r(rf->GPR)[0] & mask;
assert(occupied == false);
u4_r(rf->GPR)[0] |= mask;
}
FI_ void regfile_free_mask(RegFile_R rf, U4 mask) { FI_ void regfile_free_mask(RegFile_R rf, U4 mask) {
if (regfile_abi_mask & mask) return; if (regfile_abi_mask & mask) return;
u4_r(rf->GPR)[0] &= mask; u4_r(rf->GPR)[0] &= ~mask;
} }
FI_ void regfile_free_reg(RegFile_R rf, Reg r_id) { FI_ void regfile_free_reg(RegFile_R rf, Reg r_id) {
/* never free the ABI set */ /* never free the ABI set */
@@ -365,6 +371,10 @@ FI_ void regfile_reset(RegFile_R rf) {
rf->GPR[0] = u4_lo(regfile_abi_mask); rf->GPR[0] = u4_lo(regfile_abi_mask);
rf->GPR[1] = u4_hi(regfile_abi_mask); rf->GPR[1] = u4_hi(regfile_abi_mask);
} }
FI_ void regfile_reset_mask(RegFile_R rf, U4 mask) {
rf->GPR[0] = u4_lo(mask);
rf->GPR[1] = u4_hi(mask);
}
#pragma endregion RegFileArena (Register File Allocator) #pragma endregion RegFileArena (Register File Allocator)
#pragma region Mips Atom Procs #pragma region Mips Atom Procs
code/hello_camera/hello_camera.c
+98 -80
View File
@@ -1,7 +1,7 @@
#pragma region Vendors #pragma region Vendors
#include <stdio.h> #include <stdio.h>
#include <stdlib.h> #include <stdlib.h>
#include <assert.h> // #include <assert.h>
// #include "libgpu.h" // #include "libgpu.h"
// #include "libetc.h" // #include "libetc.h"
// #include "libgte.h" // #include "libgte.h"
@@ -144,64 +144,82 @@ FI_ void camera_look_at_c11(Camera* c, P3_S4* target, V3_S4* up_in) { resolve_lo
*/ */
internal void resolve_look_at_init(void) { internal void resolve_look_at_init(void) {
/* Wrap the static arena in a MipsAtomBuilder. */ /* Wrap the static arena in a MipsAtomBuilder. */
AtomArena ab = atomarena_make(slice_ut_arr(smem.resolve_look_at_mem)); AtomArena ab = atomarena_make(slice_ut_arr(smem.resolve_look_at_mem));
TapeBuilder tb = tb_make(slice_ut_arr(smem.resolve_look_at_atom_addrs));
/* === ATOM 0: input_and_sub === */ U4 pin_mask = regfile_abi_mask | (1 << R_ResolveScratch);
U4 const r_target_ptr = R_T0; /* tape pop → target */ RegFile rf = regfile(pin_mask);
U4 const r_eye_ptr = R_T1; /* tape pop → eye */
U4 const r_up_in_ptr = R_T2; /* tape pop → up_in */ // defer(regfile_reset_mask(& rf, pin_mask)) {
U4 const r_tmp0_0 = R_T3; // U4 r_target_ptr = regfile_alloc(& rf);
U4 const r_tmp1_0 = R_T5; // U4 r_eye_ptr = regfile_alloc(& rf);
U4 const r_tmp2_0 = R_T6; // U4 r_up_in_ptr = regfile_alloc(& rf);
U4 const r_tmp3_0 = R_T7; // U4 r_tmp0 = regfile_alloc(& rf);
// U4 r_tmp1 = regfile_alloc(& rf);
// U4 r_tmp2 = regfile_alloc(& rf);
// U4 r_tmp3 = regfile_alloc(& rf);
// tb_emit_(resolve_look_at__input_and_sub_proc(& ab,
// R_ResolveScratch,
// r_target_ptr, r_eye_ptr, r_up_in_ptr,
// r_tmp0, r_tmp1, r_tmp2, r_tmp3));
// }
U4 r_target_ptr = regfile_alloc(& rf);
U4 r_eye_ptr = regfile_alloc(& rf);
U4 r_up_in_ptr = regfile_alloc(& rf);
U4 r_tmp0 = regfile_alloc(& rf);
U4 r_tmp1 = regfile_alloc(& rf);
U4 r_tmp2 = regfile_alloc(& rf);
U4 r_tmp3 = regfile_alloc(& rf);
smem.resolve_look_at_atom_addrs[0] = resolve_look_at__input_and_sub_proc(& ab, smem.resolve_look_at_atom_addrs[0] = resolve_look_at__input_and_sub_proc(& ab,
R_ResolveScratch, R_ResolveScratch,
r_target_ptr, r_eye_ptr, r_up_in_ptr, r_target_ptr, r_eye_ptr, r_up_in_ptr,
r_tmp0_0, r_tmp1_0, r_tmp2_0, r_tmp3_0); r_tmp0, r_tmp1, r_tmp2, r_tmp3);
/* === ATOM 1: normalize fwd→uz === */ /* === ATOM 1: normalize fwd→uz === */
U4 const r_src_offset_1 = O_(ResolveLookAtScratch, fwd); U4 r_src_offset = O_(ResolveLookAtScratch, fwd);
U4 const r_dst_offset_1 = O_(ResolveLookAtScratch, uz); U4 r_dst_offset = O_(ResolveLookAtScratch, uz);
U4 const r_src_ptr_1 = R_T0; U4 r_src_ptr = R_T0;
U4 const r_dst_ptr_1 = R_T1; U4 r_dst_ptr = R_T1;
U4 const r_tmp_1 = R_T2; U4 r_tmp = R_T2;
U4 const r_mac1_1 = R_T3; U4 r_mac1 = R_T3;
U4 const r_mac2_1 = R_T5; U4 r_mac2 = R_T5;
U4 const r_recip_1 = R_T6; U4 r_recip = R_T6;
U4 const r_lzcr_1 = R_T7; U4 r_lzcr = R_T7;
U4 const r_shift_1 = R_V0; U4 r_shift = R_V0;
U4 const r_branch_1 = R_V1; U4 r_branch = R_V1;
// tb_emit_(
smem.resolve_look_at_atom_addrs[1] = normalize_v3s4_proc(& ab, smem.resolve_look_at_atom_addrs[1] = normalize_v3s4_proc(& ab,
R_ResolveScratch, R_ResolveScratch,
r_src_offset_1, r_dst_offset_1, r_src_offset, r_dst_offset,
r_src_ptr_1, r_dst_ptr_1, r_tmp_1, r_src_ptr, r_dst_ptr, r_tmp,
r_mac1_1, r_mac2_1, r_recip_1, r_lzcr_1, r_mac1, r_mac2, r_recip, r_lzcr,
r_shift_1, r_branch_1); r_shift, r_branch);
// );
/* === ATOM 2: cross uz×up_in→right === */ /* === ATOM 2: cross uz×up_in→right === */
U4 const r_a_2 = R_T0; U4 r_a_2 = R_T0;
U4 const r_b_2 = R_T1; U4 r_b_2 = R_T1;
U4 const r_c_2 = R_T2; U4 r_c_2 = R_T2;
U4 const r_d_2 = R_T3; U4 r_d_2 = R_T3;
U4 const r_f_2 = R_T5; /* out ptr (HARDCODED in body: scratch+32) */ U4 r_f_2 = R_T5; /* out ptr (HARDCODED in body: scratch+32) */
U4 const r_g_2 = R_T6; /* a ptr = scratch+16 */ U4 r_g_2 = R_T6; /* a ptr = scratch+16 */
U4 const r_h_2 = R_T7; /* b ptr = scratch+128 */ U4 r_h_2 = R_T7; /* b ptr = scratch+128 */
smem.resolve_look_at_atom_addrs[2] = resolve_look_at__cross_uz_up_in_to_right_proc(& ab, smem.resolve_look_at_atom_addrs[2] = resolve_look_at__cross_uz_up_in_to_right_proc(& ab,
R_ResolveScratch, R_ResolveScratch,
r_a_2, r_b_2, r_c_2, r_d_2, r_f_2, r_g_2, r_h_2); r_a_2, r_b_2, r_c_2, r_d_2, r_f_2, r_g_2, r_h_2);
/* === ATOM 3: normalize right→ux === */ /* === ATOM 3: normalize right→ux === */
U4 const r_src_offset_3 = O_(ResolveLookAtScratch, right); U4 r_src_offset_3 = O_(ResolveLookAtScratch, right);
U4 const r_dst_offset_3 = O_(ResolveLookAtScratch, ux); U4 r_dst_offset_3 = O_(ResolveLookAtScratch, ux);
U4 const r_src_ptr_3 = R_T0; U4 r_src_ptr_3 = R_T0;
U4 const r_dst_ptr_3 = R_T1; U4 r_dst_ptr_3 = R_T1;
U4 const r_tmp_3 = R_T2; U4 r_tmp_3 = R_T2;
U4 const r_mac1_3 = R_T3; U4 r_mac1_3 = R_T3;
U4 const r_mac2_3 = R_T5; U4 r_mac2_3 = R_T5;
U4 const r_recip_3 = R_T6; U4 r_recip_3 = R_T6;
U4 const r_lzcr_3 = R_T7; U4 r_lzcr_3 = R_T7;
U4 const r_shift_3 = R_V0; U4 r_shift_3 = R_V0;
U4 const r_branch_3 = R_V1; U4 r_branch_3 = R_V1;
smem.resolve_look_at_atom_addrs[3] = normalize_v3s4_proc(& ab, smem.resolve_look_at_atom_addrs[3] = normalize_v3s4_proc(& ab,
R_ResolveScratch, R_ResolveScratch,
r_src_offset_3, r_dst_offset_3, r_src_offset_3, r_dst_offset_3,
@@ -210,29 +228,29 @@ internal void resolve_look_at_init(void) {
r_shift_3, r_branch_3); r_shift_3, r_branch_3);
/* === ATOM 4: cross uz×ux→up === */ /* === ATOM 4: cross uz×ux→up === */
U4 const r_a_4 = R_T0; U4 r_a_4 = R_T0;
U4 const r_b_4 = R_T1; U4 r_b_4 = R_T1;
U4 const r_c_4 = R_T2; U4 r_c_4 = R_T2;
U4 const r_d_4 = R_T3; U4 r_d_4 = R_T3;
U4 const r_f_4 = R_T5; /* out ptr (HARDCODED: scratch+64) */ U4 r_f_4 = R_T5; /* out ptr (HARDCODED: scratch+64) */
U4 const r_g_4 = R_T6; /* a ptr = scratch+16 */ U4 r_g_4 = R_T6; /* a ptr = scratch+16 */
U4 const r_h_4 = R_T7; /* b ptr = scratch+48 */ U4 r_h_4 = R_T7; /* b ptr = scratch+48 */
smem.resolve_look_at_atom_addrs[4] = resolve_look_at__cross_uz_ux_to_up_proc(& ab, smem.resolve_look_at_atom_addrs[4] = resolve_look_at__cross_uz_ux_to_up_proc(& ab,
R_ResolveScratch, R_ResolveScratch,
r_a_4, r_b_4, r_c_4, r_d_4, r_f_4, r_g_4, r_h_4); r_a_4, r_b_4, r_c_4, r_d_4, r_f_4, r_g_4, r_h_4);
/* === ATOM 5: normalize up→uy === */ /* === ATOM 5: normalize up→uy === */
U4 const r_src_offset_5 = O_(ResolveLookAtScratch, up); U4 r_src_offset_5 = O_(ResolveLookAtScratch, up);
U4 const r_dst_offset_5 = O_(ResolveLookAtScratch, uy); U4 r_dst_offset_5 = O_(ResolveLookAtScratch, uy);
U4 const r_src_ptr_5 = R_T0; U4 r_src_ptr_5 = R_T0;
U4 const r_dst_ptr_5 = R_T1; U4 r_dst_ptr_5 = R_T1;
U4 const r_tmp_5 = R_T2; U4 r_tmp_5 = R_T2;
U4 const r_mac1_5 = R_T3; U4 r_mac1_5 = R_T3;
U4 const r_mac2_5 = R_T5; U4 r_mac2_5 = R_T5;
U4 const r_recip_5 = R_T6; U4 r_recip_5 = R_T6;
U4 const r_lzcr_5 = R_T7; U4 r_lzcr_5 = R_T7;
U4 const r_shift_5 = R_V0; U4 r_shift_5 = R_V0;
U4 const r_branch_5 = R_V1; U4 r_branch_5 = R_V1;
smem.resolve_look_at_atom_addrs[5] = normalize_v3s4_proc(& ab, smem.resolve_look_at_atom_addrs[5] = normalize_v3s4_proc(& ab,
R_ResolveScratch, R_ResolveScratch,
r_src_offset_5, r_dst_offset_5, r_src_offset_5, r_dst_offset_5,
@@ -241,14 +259,14 @@ internal void resolve_look_at_init(void) {
r_shift_5, r_branch_5); r_shift_5, r_branch_5);
/* === ATOM 6a: populate (m[][] from ux/uy/uz, t[]=0) === */ /* === ATOM 6a: populate (m[][] from ux/uy/uz, t[]=0) === */
U4 const r_look_at_6a = R_T0; /* tape pop → look_at* */ U4 r_look_at_6a = R_T0; /* tape pop → look_at* */
U4 const r_scratch_6a = R_ResolveScratch; U4 r_scratch_6a = R_ResolveScratch;
U4 const r_pux_6a = R_T1; U4 r_pux_6a = R_T1;
U4 const r_puy_6a = R_T3; U4 r_puy_6a = R_T3;
U4 const r_puz_6a = R_T5; U4 r_puz_6a = R_T5;
U4 const r_tmp0_6a = R_T2; U4 r_tmp0_6a = R_T2;
U4 const r_tmp1_6a = R_T6; U4 r_tmp1_6a = R_T6;
U4 const r_tmp2_6a = R_V0; U4 r_tmp2_6a = R_V0;
smem.resolve_look_at_atom_addrs[6] = resolve_look_at__populate_proc(& ab, smem.resolve_look_at_atom_addrs[6] = resolve_look_at__populate_proc(& ab,
r_look_at_6a, r_scratch_6a, r_look_at_6a, r_scratch_6a,
r_pux_6a, r_puy_6a, r_puz_6a, r_pux_6a, r_puy_6a, r_puz_6a,
@@ -265,21 +283,21 @@ internal void resolve_look_at_init(void) {
* Uses mac_apply_matrix_lv component macro which internally uses * Uses mac_apply_matrix_lv component macro which internally uses
* r_t0 for the RT matrix load + V0 load, then r_t0/r_t1/r_t2 * r_t0 for the RT matrix load + V0 load, then r_t0/r_t1/r_t2
* for the mfc2/store. We pass our GPRs. */ * for the mfc2/store. We pass our GPRs. */
U4 const r_scratch_6b = R_ResolveScratch; U4 r_scratch_6b = R_ResolveScratch;
U4 const r_peye_6b = R_T1; /* scratch+96 (packed V0 dst, then off dst) */ U4 r_peye_6b = R_T1; /* scratch+96 (packed V0 dst, then off dst) */
U4 const r_look_at_6b = R_T0; /* tape pop → look_at* */ U4 r_look_at_6b = R_T0; /* tape pop → look_at* */
U4 const r_tmp0_6b = R_T2; U4 r_tmp0_6b = R_T2;
U4 const r_tmp1_6b = R_T3; U4 r_tmp1_6b = R_T3;
U4 const r_tmp2_6b = R_T5; U4 r_tmp2_6b = R_T5;
smem.resolve_look_at_atom_addrs[8] = resolve_look_at__matrix_vector_proc(& ab, smem.resolve_look_at_atom_addrs[8] = resolve_look_at__matrix_vector_proc(& ab,
r_scratch_6b, r_peye_6b, r_look_at_6b, r_scratch_6b, r_peye_6b, r_look_at_6b,
r_tmp0_6b, r_tmp1_6b, r_tmp2_6b); r_tmp0_6b, r_tmp1_6b, r_tmp2_6b);
/* === ATOM 6c: trans_matrix (off → look_at->t[]) === */ /* === ATOM 6c: trans_matrix (off → look_at->t[]) === */
U4 const r_look_at_6c = R_T0; /* tape pop → look_at* */ U4 r_look_at_6c = R_T0; /* tape pop → look_at* */
U4 const r_scratch_6c = R_ResolveScratch; U4 r_scratch_6c = R_ResolveScratch;
U4 const r_off_ptr_6c = R_T1; /* &scratch.eye (= off dst) */ U4 r_off_ptr_6c = R_T1; /* &scratch.eye (= off dst) */
U4 const r_tmp0_6c = R_T2; U4 r_tmp0_6c = R_T2;
smem.resolve_look_at_atom_addrs[9] = resolve_look_at__trans_matrix_proc(& ab, smem.resolve_look_at_atom_addrs[9] = resolve_look_at__trans_matrix_proc(& ab,
r_look_at_6c, r_scratch_6c, r_off_ptr_6c, r_tmp0_6c, R_T3, R_T4); r_look_at_6c, r_scratch_6c, r_off_ptr_6c, r_tmp0_6c, R_T3, R_T4);
scripts/build_psyq.ps1
+1
View File
@@ -532,6 +532,7 @@ function build-hello_camera {
$compile_args = @() $compile_args = @()
$compile_args += $f_debug $compile_args += $f_debug
$compile_args += ($f_define + 'BUILD_DEBUG')
$compile_args += $f_optimize_none $compile_args += $f_optimize_none
# $compile_args += $f_optimize_intrinsics # $compile_args += $f_optimize_intrinsics
# $compile_args += $f_optimize_size # $compile_args += $f_optimize_size