mirror of
https://github.com/ClassiCube/ClassiCube.git
synced 2025-09-09 15:28:21 -04:00
Dreamcast: Use aliases for registers in VertexTransform.S, only load fr0 once
This commit is contained in:
UnknownShadow200
parent
7e656d278e
commit
485098d4fc
@ -1,77 +1,44 @@
|
||||
! =========================================================
|
||||
! ======================== PROCESSOR INFO =================
|
||||
! =========================================================
|
||||
! The SH4 can dual issue (i.e. parallel execution) two instructions
|
||||
! as long as the groups of the two instructions are different:
|
||||
! * LS - most ALU and FPU register load/stores
|
||||
! * EX - most ALU arithmetic instructions
|
||||
! * MT - TST, CMP, NOP, MOV Rm,Rn
|
||||
! * FE - most FPU arithmetic instructions
|
||||
! * CO - other instructions (NOTE: Cannot be exeucted in parallel)
|
||||
#define FLG r0 // clip flags
|
||||
#define TMP r1 // temp
|
||||
#define VTX r2 // PVR_CMD_VERTEX
|
||||
#define EOS r3 // PVR_CMD_VERTEX_EOL
|
||||
#define SRC r4 // src pointer ARG
|
||||
#define DST r5 // dst pointer ARG
|
||||
#define CNT r6 // quads count ARG
|
||||
#define PFT r7 // prefetch address
|
||||
|
||||
! Thee following general aspects of instructions are important to note per the SH4 manual:
|
||||
! * Issue rate: Interval between the issue of an instruction and that of the next instruction
|
||||
! * Latency: Interval between the issue of an instruction and the generation of its result (completion)
|
||||
! * Latency is also the interval between the execution of two instructions with an interdependent relationship.
|
||||
! (although different cases may either increase or decrease Latency)
|
||||
#define ZERO fr0 // 0.0
|
||||
#define F_U fr1 // vertex.u
|
||||
#define F_V fr2 // vertex.v
|
||||
#define F_C fr3 // vertex.colour
|
||||
#define F_X fr4 // vertex.x
|
||||
#define F_Y fr5 // vertex.y
|
||||
#define F_Z fr6 // vertex.z
|
||||
#define F_W fr7 // vertex.w
|
||||
|
||||
|
||||
! =========================================================
|
||||
! ======================== REGISTER USAGES ================
|
||||
! =========================================================
|
||||
! SH4 C ABI:
|
||||
! - R0 to R3 are return values (can be overwritten)
|
||||
! - R4 to R7 are input arguments (can be overwritten)
|
||||
! - R8 to R13 are non-volatile (must be restored at end)
|
||||
! - R14 is the frame pointer (must be restored at end)
|
||||
! - R15 is the stack pointer (must be restored at end)
|
||||
! - FR0 to FR3 are return values (can be overwritten)
|
||||
! - FR4 to FR11 are input arguments (can be overwritten)
|
||||
! - FR12 to FR13 are non-volatile (must be restored at end)
|
||||
|
||||
!r0 = clip flags
|
||||
!r1 = GPU command
|
||||
!r2 = temp
|
||||
!r3 = prefetch address
|
||||
!r4 = src pointer ARG
|
||||
!r5 = dst pointer ARG
|
||||
!r6 = quads count ARG
|
||||
!r7 = PVR_CMD_VERTEX
|
||||
!r11 = PVR_CMD_VERTEX_EOL
|
||||
|
||||
!fr0 = temp
|
||||
!fr1 = u
|
||||
!fr2 = v
|
||||
!fr3 = c
|
||||
!fr4 = x
|
||||
!fr5 = y
|
||||
!fr6 = z
|
||||
!fr7 = w
|
||||
!fv4 = XYZW
|
||||
#define XYZW fv4 // vertex.xyzw
|
||||
|
||||
|
||||
! =========================================================
|
||||
! ========================= TRANSFORM SETUP ===============
|
||||
! =========================================================
|
||||
.macro TransformSetup
|
||||
mov r4,r3 ! MT, r3 = src
|
||||
add #-32, r5 ! EX, r5 -= sizeof(VERTEX)
|
||||
mov.l r11, @-r15 ! LS, push(r11)
|
||||
mov #0xE0, r7 ! EX, r7 = 0x00 00 00 E0
|
||||
pref @r3 ! LS, PREFETCH r3 (first vertex)
|
||||
shll16 r7 ! EX, r7 = 0x00 E0 00 00
|
||||
shll8 r7 ! EX, r7 = 0xE0 00 00 00 (PVR_CMD_VERTEX)
|
||||
mov #0xF0, r11 ! EX, r11 = 0x00 00 00 F0
|
||||
shll16 r11 ! EX, r11 = 0x00 F0 00 00
|
||||
shll8 r11 ! EX, r11 = 0xF0 00 00 00 (PVR_CMD_VERTEX_EOL)
|
||||
mov SRC, PFT ! MT, pft = src
|
||||
add #-32, DST ! EX, dst -= sizeof(VERTEX)
|
||||
mov #0xE0, VTX ! EX, VTX = 0x00 00 00 E0
|
||||
pref @PFT ! LS, PREFETCH pft (first vertex)
|
||||
shll16 VTX ! EX, VTX = 0x00 E0 00 00
|
||||
shll8 VTX ! EX, VTX = 0xE0 00 00 00 (PVR_CMD_VERTEX)
|
||||
mov #0xF0, EOS ! EX, EOS = 0x00 00 00 F0
|
||||
shll16 EOS ! EX, EOS = 0x00 F0 00 00
|
||||
shll8 EOS ! EX, EOS = 0xF0 00 00 00 (PVR_CMD_VERTEX_EOL)
|
||||
fldi0 ZERO ! LS, fr0 = 0.0
|
||||
.endm
|
||||
|
||||
.macro TransformEnd
|
||||
mov.l @r15+, r11 ! LS, pop(r11)
|
||||
|
||||
add #32, r5 ! EX, r5 += sizeof(VERTEX)
|
||||
rts ! CO, return after executing instruction in delay slot
|
||||
mov r5,r0 ! MT, r0 = r5
|
||||
add #32, DST ! EX, DST += sizeof(VERTEX)
|
||||
rts ! CO, return after executing instruction in delay slot
|
||||
mov DST, r0 ! MT, r0 = DST
|
||||
.endm
|
||||
|
||||
|
||||
@ -80,36 +47,36 @@
|
||||
! =========================================================
|
||||
.macro LoadColouredVertex
|
||||
! LOAD XYZ
|
||||
fmov @r4+, fr4 ! LS, X = src->x
|
||||
fmov @r4+, fr5 ! LS, Y = src->y
|
||||
fmov @r4+, fr6 ! LS, Z = src->z
|
||||
fldi1 fr7 ! LS, W = 1.0
|
||||
fmov @SRC+, F_X ! LS, X = src->x
|
||||
fmov @SRC+, F_Y ! LS, Y = src->y
|
||||
fmov @SRC+, F_Z ! LS, Z = src->z
|
||||
fldi1 F_W ! LS, W = 1.0
|
||||
! PREPARE NEXT VERTEX
|
||||
add #16, r3 ! EX, r3 += VERTEX_STRIDE
|
||||
pref @r3 ! LS, PREFETCH r3 (next vertex)
|
||||
add #64, r5 ! EX, r5 += 2 * sizeof(VERTEX)
|
||||
add #16, PFT ! EX, pft += VERTEX_STRIDE
|
||||
pref @PFT ! LS, PREFETCH pft (next vertex)
|
||||
add #64, DST ! EX, dst += 2 * sizeof(VERTEX)
|
||||
! TRANSFORM VERTEX
|
||||
ftrv xmtrx, fv4 ! FE, TRANSFORM(XYZW)
|
||||
ftrv xmtrx, XYZW ! FE, TRANSFORM(XYZW)
|
||||
! LOAD ATTRIBUTES
|
||||
fmov @r4+,fr3 ! LS, C = src->color
|
||||
fmov @SRC+, F_C ! LS, C = src->color
|
||||
.endm
|
||||
|
||||
.macro LoadTexturedVertex
|
||||
! LOAD XYZ
|
||||
fmov @r4+, fr4 ! LS, X = src->x
|
||||
fmov @r4+, fr5 ! LS, Y = src->y
|
||||
fmov @r4+, fr6 ! LS, Z = src->z
|
||||
fldi1 fr7 ! LS, W = 1.0
|
||||
fmov @SRC+, F_X ! LS, X = src->x
|
||||
fmov @SRC+, F_Y ! LS, Y = src->y
|
||||
fmov @SRC+, F_Z ! LS, Z = src->z
|
||||
fldi1 F_W ! LS, W = 1.0
|
||||
! PREPARE NEXT VERTEX
|
||||
add #24, r3 ! EX, r3 += VERTEX_STRIDE
|
||||
pref @r3 ! LS, PREFETCH r3 (next vertex)
|
||||
add #64, r5 ! EX, r5 += 2 * sizeof(VERTEX)
|
||||
add #24, PFT ! EX, pft += VERTEX_STRIDE
|
||||
pref @PFT ! LS, PREFETCH pft (next vertex)
|
||||
add #64, DST ! EX, dst += 2 * sizeof(VERTEX)
|
||||
! TRANSFORM VERTEX
|
||||
ftrv xmtrx, fv4 ! FE, TRANSFORM(XYZW)
|
||||
ftrv xmtrx, XYZW ! FE, TRANSFORM(XYZW)
|
||||
! LOAD ATTRIBUTES
|
||||
fmov @r4+,fr3 ! LS, C = src->color
|
||||
fmov @r4+,fr1 ! LS, U = src->u
|
||||
fmov @r4+,fr2 ! LS, V = src->v
|
||||
fmov @SRC+, F_C ! LS, C = src->color
|
||||
fmov @SRC+, F_U ! LS, U = src->u
|
||||
fmov @SRC+, F_V ! LS, V = src->v
|
||||
.endm
|
||||
|
||||
! =========================================================
|
||||
@ -118,67 +85,63 @@
|
||||
! To take advantage of SH4 dual instruction processing,
|
||||
! clipflag calculation and vertex output are interleaved
|
||||
.macro ProcessVertex1
|
||||
fmov.s fr7,@-r5 ! LS, dst->w = W
|
||||
fmov.s fr3,@-r5 ! LS, dst->c = C
|
||||
fldi0 fr0 ! LS, fr0 = 0.0
|
||||
fmov.s fr2,@-r5 ! LS, dst->v = V
|
||||
fcmp/gt fr0,fr6 ! FE, T = Z > 0
|
||||
fmov.s fr1,@-r5 ! LS, dst->u = U
|
||||
movt r0 ! EX, CLIPFLAGS = T
|
||||
fmov.s fr6,@-r5 ! LS, dst->z = Z
|
||||
fmov.s fr5,@-r5 ! LS, dst->y = Y
|
||||
fmov.s fr4,@-r5 ! LS, dst->x = X
|
||||
mov.l r7,@-r5 ! LS, dst->flags = PVR_CMD_VERTEX
|
||||
fmov.s F_W,@-DST ! LS, dst->w = W
|
||||
fmov.s F_C,@-DST ! LS, dst->c = C
|
||||
fmov.s F_V,@-DST ! LS, dst->v = V
|
||||
fcmp/gt ZERO, F_Z ! FE, T = Z > 0
|
||||
fmov.s F_U,@-DST ! LS, dst->u = U
|
||||
movt FLG ! EX, CLIPFLAGS = T
|
||||
fmov.s F_Z,@-DST ! LS, dst->z = Z
|
||||
fmov.s F_Y,@-DST ! LS, dst->y = Y
|
||||
fmov.s F_X,@-DST ! LS, dst->x = X
|
||||
mov.l VTX,@-DST ! LS, dst->flags = PVR_CMD_VERTEX
|
||||
.endm
|
||||
|
||||
.macro ProcessVertex2
|
||||
fmov.s fr7,@-r5 ! LS, dst->w = W
|
||||
fmov.s fr3,@-r5 ! LS, dst->c = C
|
||||
fldi0 fr0 ! LS, fr0 = 0.0
|
||||
fmov.s fr2,@-r5 ! LS, dst->v = V
|
||||
fcmp/gt fr0,fr6 ! FE, T = Z > 0
|
||||
fmov.s fr1,@-r5 ! LS, dst->u = U
|
||||
movt r2 ! EX, tmp = T
|
||||
fmov.s fr6,@-r5 ! LS, dst->z = Z
|
||||
add r2,r2 ! EX, tmp = tmp + tmp
|
||||
fmov.s fr5,@-r5 ! LS, dst->y = Y
|
||||
or r2,r0 ! EX, CLIPFLAGS |= tmp (T << 1)
|
||||
fmov.s fr4,@-r5 ! LS, dst->x = X
|
||||
mov.l r7,@-r5 ! LS, dst->flags = PVR_CMD_VERTEX
|
||||
fmov.s F_W,@-DST ! LS, dst->w = W
|
||||
fmov.s F_C,@-DST ! LS, dst->c = C
|
||||
fmov.s F_V,@-DST ! LS, dst->v = V
|
||||
fcmp/gt ZERO,F_Z ! FE, T = Z > 0
|
||||
fmov.s F_U,@-DST ! LS, dst->u = U
|
||||
movt TMP ! EX, tmp = T
|
||||
fmov.s F_Z,@-DST ! LS, dst->z = Z
|
||||
add TMP,TMP ! EX, tmp = tmp + tmp
|
||||
fmov.s F_Y,@-DST ! LS, dst->y = Y
|
||||
or TMP,FLG ! EX, CLIPFLAGS |= tmp (T << 1)
|
||||
fmov.s F_X,@-DST ! LS, dst->x = X
|
||||
mov.l VTX,@-DST ! LS, dst->flags = PVR_CMD_VERTEX
|
||||
.endm
|
||||
|
||||
.macro ProcessVertex3
|
||||
fmov.s fr7,@-r5 ! LS, dst->w = W
|
||||
fmov.s fr3,@-r5 ! LS, dst->c = C
|
||||
fldi0 fr0 ! LS, fr0 = 0.0
|
||||
fmov.s fr2,@-r5 ! LS, dst->v = V
|
||||
fcmp/gt fr0,fr6 ! FE, T = Z > 0
|
||||
fmov.s fr1,@-r5 ! LS, dst->u = U
|
||||
movt r2 ! EX, tmp = T
|
||||
fmov.s fr6,@-r5 ! LS, dst->z = Z
|
||||
fmov.s fr5,@-r5 ! LS, dst->y = Y
|
||||
shll2 r2 ! EX, tmp = tmp << 2
|
||||
fmov.s fr4,@-r5 ! LS, dst->x = X
|
||||
or r2,r0 ! EX, CLIPFLAGS |= tmp (T << 2)
|
||||
mov.l r7,@-r5 ! LS, dst->flags = PVR_CMD_VERTEX
|
||||
fmov.s F_W,@-DST ! LS, dst->w = W
|
||||
fmov.s F_C,@-DST ! LS, dst->c = C
|
||||
fmov.s F_V,@-DST ! LS, dst->v = V
|
||||
fcmp/gt ZERO, F_Z ! FE, T = Z > 0
|
||||
fmov.s F_U,@-DST ! LS, dst->u = U
|
||||
movt TMP ! EX, tmp = T
|
||||
fmov.s F_Z,@-DST ! LS, dst->z = Z
|
||||
fmov.s F_Y,@-DST ! LS, dst->y = Y
|
||||
shll2 TMP ! EX, tmp = tmp << 2
|
||||
fmov.s F_X,@-DST ! LS, dst->x = X
|
||||
or TMP,FLG ! EX, CLIPFLAGS |= tmp (T << 2)
|
||||
mov.l VTX,@-DST ! LS, dst->flags = PVR_CMD_VERTEX
|
||||
.endm
|
||||
|
||||
.macro ProcessVertex4
|
||||
fmov.s fr7,@-r5 ! LS, dst->w = W
|
||||
or r11,r0 ! EX, CLIPFLAGS |= PVR_CMD_VERTEX_EOL
|
||||
fmov.s fr3,@-r5 ! LS, dst->c = C
|
||||
fldi0 fr0 ! LS, fr0 = 0.0
|
||||
fmov.s fr2,@-r5 ! LS, dst->v = V
|
||||
fcmp/gt fr0,fr6 ! FE, T = Z > 0
|
||||
fmov.s fr1,@-r5 ! LS, dst->u = U
|
||||
movt r2 ! EX, tmp = T
|
||||
fmov.s fr6,@-r5 ! LS, dst->z = Z
|
||||
shll2 r2 ! EX, tmp = tmp << 2
|
||||
fmov.s fr5,@-r5 ! LS, dst->y = Y
|
||||
add r2,r2 ! EX, tmp = (tmp << 2) + (tmp << 2) (T << 3)
|
||||
fmov.s fr4,@-r5 ! LS, dst->x = X
|
||||
or r2,r0 ! EX, CLIPFLAGS |= tmp (T << 3)
|
||||
mov.l r0,@-r5 ! LS, dst->flags = PVR_CMD_VERTEX_EOL | CLIPFLAGS
|
||||
fmov.s F_W,@-DST ! LS, dst->w = W
|
||||
or EOS,FLG ! EX, CLIPFLAGS |= PVR_CMD_VERTEX_EOL
|
||||
fmov.s F_C,@-DST ! LS, dst->c = C
|
||||
fmov.s F_V,@-DST ! LS, dst->v = V
|
||||
fcmp/gt ZERO, F_Z ! FE, T = Z > 0
|
||||
fmov.s F_U,@-DST ! LS, dst->u = U
|
||||
movt TMP ! EX, tmp = T
|
||||
fmov.s F_Z,@-DST ! LS, dst->z = Z
|
||||
shll2 TMP ! EX, tmp = tmp << 2
|
||||
fmov.s F_Y,@-DST ! LS, dst->y = Y
|
||||
add TMP,TMP ! EX, tmp = (tmp << 2) + (tmp << 2) (T << 3)
|
||||
fmov.s F_X,@-DST ! LS, dst->x = X
|
||||
or TMP,FLG ! EX, CLIPFLAGS |= tmp (T << 3)
|
||||
mov.l FLG,@-DST ! LS, dst->flags = PVR_CMD_VERTEX_EOL | CLIPFLAGS
|
||||
.endm
|
||||
|
||||
|
||||
@ -206,21 +169,21 @@ _DrawTexturedQuads:
|
||||
ProcessVertex4
|
||||
|
||||
! CLIPFLAGS TESTING
|
||||
and #15,r0
|
||||
cmp/eq #0,r0 ! T = r0 == 0 (all points invisible)
|
||||
bt/s .T_NO_POINTS_VISIBLE ! if T goto NO_POINTS_VISIBLE
|
||||
and #15,FLG
|
||||
cmp/eq #0,FLG ! T = CLIPFLAGS == 0 (all points invisible)
|
||||
bt/s .T_NONE_VISIBLE ! if T goto NONE_VISIBLE
|
||||
nop
|
||||
bra .T_SOME_POINTS_VISIBLE
|
||||
bra .T_SOME_VISIBLE
|
||||
nop
|
||||
|
||||
.T_NO_POINTS_VISIBLE:
|
||||
.T_NONE_VISIBLE:
|
||||
bra .T_LOOP_END ! jump to loop end after executing instruction in delay slot
|
||||
add #-128, r5 ! r5 -= 4 * sizeof(VERTEX), move back to prior quad, so that this invisible quad gets overwritten in next iteration
|
||||
add #-128, DST ! DST -= 4 * sizeof(VERTEX), move back to prior quad, so that this invisible quad gets overwritten in next iteration
|
||||
|
||||
.T_SOME_POINTS_VISIBLE:
|
||||
.T_SOME_VISIBLE:
|
||||
|
||||
.T_LOOP_END:
|
||||
dt r6 ! r6--; T = r6 == 0
|
||||
dt CNT ! count--; T = count == 0
|
||||
bf .T_TRANSFORM_QUAD ! if !T then goto T_TRANSFORM_QUAD
|
||||
nop
|
||||
|
||||
@ -236,8 +199,8 @@ _DrawTexturedQuads:
|
||||
|
||||
_DrawColouredQuads:
|
||||
! Setup
|
||||
fldi0 fr1 ! U = 0
|
||||
fldi0 fr2 ! V = 0
|
||||
fldi0 F_U ! U = 0
|
||||
fldi0 F_V ! V = 0
|
||||
TransformSetup
|
||||
|
||||
.C_TRANSFORM_QUAD:
|
||||
@ -254,21 +217,21 @@ _DrawColouredQuads:
|
||||
ProcessVertex4
|
||||
|
||||
! CLIPFLAGS TESTING
|
||||
and #15,r0
|
||||
cmp/eq #0,r0 ! T = r0 == 0 (all points invisible)
|
||||
bt/s .C_NO_POINTS_VISIBLE ! if T goto NO_POINTS_VISIBLE
|
||||
and #15,FLG
|
||||
cmp/eq #0,FLG ! T = CLIPFLAGS == 0 (all points invisible)
|
||||
bt/s .C_NONE_VISIBLE ! if T goto NONE_VISIBLE
|
||||
nop
|
||||
bra .C_SOME_POINTS_VISIBLE
|
||||
bra .C_SOME_VISIBLE
|
||||
nop
|
||||
|
||||
.C_NO_POINTS_VISIBLE:
|
||||
.C_NONE_VISIBLE:
|
||||
bra .C_LOOP_END ! jump to loop end after executing instruction in delay slot
|
||||
add #-128, r5 ! r5 -= 4 * sizeof(VERTEX), move back to 1 vertex before start of quad
|
||||
add #-128, DST ! dst -= 4 * sizeof(VERTEX), move back to 1 vertex before start of quad
|
||||
|
||||
.C_SOME_POINTS_VISIBLE:
|
||||
.C_SOME_VISIBLE:
|
||||
|
||||
.C_LOOP_END:
|
||||
dt r6 ! r6--; T = r6 == 0
|
||||
dt CNT ! count--; T = count == 0
|
||||
bf .C_TRANSFORM_QUAD ! if !T then goto TRANSFORM_QUAD
|
||||
nop
|
||||
|
||||
|
||||
31
misc/dreamcast/sh4_notes.txt
Normal file
31
misc/dreamcast/sh4_notes.txt
Normal file
@ -0,0 +1,31 @@
|
||||
=========================================================
|
||||
======================== PROCESSOR INFO =================
|
||||
=========================================================
|
||||
The SH4 can dual issue (i.e. parallel execution) two instructions
|
||||
as long as the groups of the two instructions are different:
|
||||
* LS - most ALU and FPU register load/stores
|
||||
* EX - most ALU arithmetic instructions
|
||||
* MT - TST, CMP, NOP, MOV Rm,Rn (NOTE: Can execute in parallel with other MT)
|
||||
* FE - most FPU arithmetic instructions
|
||||
* CO - other instructions (NOTE: Cannot never execute in parallel)
|
||||
|
||||
The following general aspects of instructions are important to note per the SH4 manual:
|
||||
* Issue rate: Interval between the issue of an instruction and that of the next instruction
|
||||
* Latency: Interval between the issue of an instruction and the generation of its result (completion)
|
||||
* Latency is also the interval between the execution of two instructions with an interdependent relationship.
|
||||
(although different cases may either increase or decrease Latency)
|
||||
|
||||
|
||||
=========================================================
|
||||
======================== REGISTER USAGES ================
|
||||
=========================================================
|
||||
SH4 C ABI:
|
||||
- R0 to R3 are return values (can be overwritten)
|
||||
- R4 to R7 are input arguments (can be overwritten)
|
||||
- R8 to R13 are non-volatile (must be restored at end)
|
||||
- R14 is the frame pointer (must be restored at end)
|
||||
- R15 is the stack pointer (must be restored at end)
|
||||
- FR0 to FR3 are return values (can be overwritten)
|
||||
- FR4 to FR11 are input arguments (can be overwritten)
|
||||
- FR12 to FR13 are non-volatile (must be restored at end)
|
||||
|
||||
@ -37,6 +37,14 @@ float sqrtf(float x) {
|
||||
}
|
||||
#elif defined __GNUC__
|
||||
/* Defined in .h using builtins */
|
||||
#elif defined __TINYC__
|
||||
/* Older versions of TinyC don't support fabsf or sqrtf */
|
||||
/* Those can be used though if compiling with newer TinyC */
|
||||
/* versions for a very small performance improvement */
|
||||
#include <math.h>
|
||||
|
||||
float Math_AbsF(float x) { return fabs(x); }
|
||||
float Math_SqrtF(float x) { return sqrt(x); }
|
||||
#else
|
||||
#include <math.h>
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user