GTA 3 Port by SKMP

[TheSaviorDavidBowie]

The Savior Has Arised

[Arnold.D]

I might have to make my own thread to get help lol

[lozz]

i cant get dreamSDK to download from GITHUB i only see on the bottom of the page the SDK R3 setup iso? is that what i need? what program do i open it with my CPU does not know. I purchased the GTA3 from rockstar and want to make a CDI but found a CDI someone made online in the archives.....that one is old and freezes on the car crasher stuff (been playing the game a few weeks. HATE the radio situation but GTA 2 was like that too so im not crying...) Im trying to make the latest and greatest CDI but am having issues even starting.....

If I recall correctly, I believe you can unzip the ISO. Once you've done that, you'll find there is an installer in the folder which you just run.

[Arnold.D]

ok groovy ill do that this weekend. thank you kindly! all the missions im on right now around the city are pretty much car crushing involved. This weekend i may be able to progress a little more.

[mistamontiel]

Ever since Vista (or 7?) Windoze just mounts the ISO if not it's WinCDEmu

[grouik1er]

I ask Grok to Analyse the 10 last post of @falco_girgis on Xr relative to GTA3.
To optimize with full sh4 asm simd and mathematical approximation for speed and cycle count.
I don't think this code works but ican't test it so maybe it can inspire

; Optimized Vertex Transformation for GTA III on Dreamcast
; Includes reciprocal table generation and SIMD matrix multiplication
; Author: Grok (inspired by @falco_girgis's work)
; Date: March 06, 2025

; Inputs:
; r1 = Pointer to vertex buffer (x1, y1, z1, x2, y2, z2, ...), 4 vertices at a time
; r2 = Pointer to 3x3 matrix (m00-m22, aligned for SIMD)
; r3 = Pointer to translation vector (tx, ty, tz)
; r4 = Pointer to output buffer for transformed vertices (x', y', z')
; r15 = End address of vertex buffer (for loop termination)
; Outputs:
; Transformed vertices written to r4

.section .data
.align 4
recip_table: ; Reciprocal lookup table (4096 entries, 4 KB, filled by init_recip_table)
.space 4096*4 ; Reserve 4 KB for 4096 single-precision floats (1/w values)

.section .text
.align 4 ; Align code for optimal instruction fetch

; Subroutine to initialize the reciprocal table (call once at startup)
init_recip_table:
MOV #0x3F000000, r5 ; r5 = 0.5 (start of range, IEEE 754)
MOV #4096, r6 ; r6 = number of entries
MOV #recip_table, r7; r7 = pointer to table
MOV #0x3A4, r8 ; r8 = step size (~0.000366, for 0.5 to 2.0 range)
.init_loop:
MOV.L r5, @r7 ; Store current value (w)
FCNVSD r5, fr0 ; Convert w to float in fr0
FSRRA fr0, fr1 ; fr1 = 1/sqrt(w) (approx), then refine
FMUL fr1, fr1, fr1 ; fr1 = 1/w (square the reciprocal sqrt)
MOV.L fr1, @(r7,4) ; Store 1/w in table
FADD r8, r5, r5 ; Increment w by step size
ADD #4, r7 ; Move to next table entry
DT r6 ; Decrement counter
BF .init_loop ; Loop until all 4096 entries are filled
RTS ; Return
NOP ; Delay slot

; Main vertex transformation routine
transform_vertices:
MOV #recip_table, r10 ; r10 = pointer to precomputed reciprocal table
.loop:
PREF @r1 ; Preload vertex data into SH-4's 16 KB cache

; Load 3x3 matrix into vector registers (128-bit, 4 floats each)
FMOV.DR @r2+, DR4 ; DR4 = {m00, m01, m02, m10} - First row + part of second
FMOV.DR @r2+, DR6 ; DR6 = {m11, m12, m20, m21} - Rest of second + third
FMOV.DR @r2, DR8 ; DR8 = {m22, pad, pad, pad} - Last element (padded)

; Load 4 vertices (x, y, z) into vector registers
FMOV.DR @r1+, DR0 ; DR0 = {x1, x2, x3, x4} - 4 x-coordinates
FMOV.DR @r1+, DR2 ; DR2 = {y1, y2, y3, y4} - 4 y-coordinates
FMOV.DR @r1+, DR10 ; DR10 = {z1, z2, z3, z4} - 4 z-coordinates

; Compute x' for 4 vertices in parallel
FMAC DR0, DR4, DR12 ; DR12 = x * m00 (SIMD multiply-accumulate)
FMAC DR2, DR5, DR12 ; DR12 += y * m01
FMAC DR10, DR6, DR12; DR12 += z * m02
FMOV.DR @r3, DR14 ; DR14 = {tx, tx, tx, tx} - Load translation x
FADD DR12, DR14, DR0; DR0 = x' = x*m00 + y*m01 + z*m02 + tx

; Compute y' for 4 vertices
FMAC DR0, DR4+1, DR12; DR12 = x * m10
FMAC DR2, DR6, DR12 ; DR12 += y * m11
FMAC DR10, DR6+1, DR12; DR12 += z * m12
FMOV.DR @(4,r3), DR14; DR14 = {ty, ty, ty, ty} - Load translation y
FADD DR12, DR14, DR1; DR1 = y' = x*m10 + y*m11 + z*m12 + ty

; Compute z' for 4 vertices
FMAC DR0, DR4+2, DR12; DR12 = x * m20
FMAC DR2, DR6+1, DR12; DR12 += y * m21
FMAC DR10, DR8, DR12 ; DR12 += z * m22
FMOV.DR @(8,r3), DR14; DR14 = {tz, tz, tz, tz} - Load translation z
FADD DR12, DR14, DR2; DR2 = z' = x*m20 + y*m21 + z*m22 + tz

; Fast perspective division (1/w) using lookup table
; Example for first vertex (repeat for others in DR0-DR2)
FMOV FR0, r5 ; r5 = w1 (assuming w from projection or vertex data)
SHLR r5, r6, #9 ; r6 = w >> 9 = index in table (10-bit precision)
AND #0x0080, r7, r5 ; r7 = fraction (7 bits for interpolation)
MOV.L @(r6,r10), r8 ; r8 = 1/w from table
ADD #4, r6 ; Next entry
MOV.L @(r6,r10), r9 ; r9 = 1/w[i+1]
FSUB r9, r8, r11 ; r11 = delta = 1/w[i+1] - 1/w
FMUL r11, r7, r11 ; r11 = delta * fraction
FADD r8, r11, FR0 ; FR0 = 1/w = 1/w + delta * fraction (~8 cycles)

; Apply 1/w to x', y', z' for first vertex (repeat for others)
FMUL FR0, FR0, FR0 ; x' = x' * (1/w)
FMUL FR1, FR0, FR1 ; y' = y' * (1/w)
FMUL FR2, FR0, FR2 ; z' = z' * (1/w)

; Store results without cache latency
MOVCA.L DR0, @r4+ ; Write 4 transformed x' coordinates
MOVCA.L DR1, @r4+ ; Write 4 transformed y' coordinates
MOVCA.L DR2, @r4+ ; Write 4 transformed z' coordinates

; Loop control
ADD #12, r1 ; Next group of 4 vertices (3 floats per vertex)
CMP/HS r1, r15 ; End of vertex buffer?
BT .loop ; Continue if more vertices

.end:
RTS ; Return
NOP ; Delay slot

[Matsilagi]

Hey y'all, i know being new and asking questions suck, but i didn't find it anywhere else.
So, what settings do i need to run this GTA 3 port on DreamShell/Retrodream? I tried both with an ISO and CDI and couldn't get past the loader saying "Executing" no matter what. Does it need any exact settings to run that way? I saw people running via Serial but i can't do that since i dont have a serial cable (I got this from a friend pre-modded with Dreamshell/Retrodream and a SD Card).
ISO images are fine since they worked on another friend's GDEMU and any emulators can load and read them.
Thanks in advance.

[fafadou]

Retrodream loader 14
Last version of deeamshell default preset for me.

[Cool_amigaN]

Could anyone desrcibe the exact process of updating with newer builds? I deleted "dca3-game" folder, I re-cloned the repo from "git clone --branch alpha https://gitlab.com/skmp/dca3-game.git", then downloaded the latest "build-dreamcast-liberty" from here and unpacked to "dca3-game\dreamcast", renamed the newer ELF as "dca3.elf" and rebuilt the image. After booting the cdi under gdemu, I got a crash at the main screen (right before the main menu options). Following the guide for the original build, game loads and plays fine.

[mistamontiel]

No longer is DCA in alpha! First step just don't key in --branch alpha

ELFs mustn't be renamed

I too also got VC going two disc unpatched v1.0 and that takes a bit more attempts, close DreamSDK and resume proceed again oh that instead uses:

Code: Select all

make FOR_DISC=2 cdi-prebuilt

Not for disc by any means just uses that command to cook it a lil north of a gig CDI will spit out