AstralTransRocketries/ClassiCube
1
0
Fork 0
mirror of https://github.com/ClassiCube/ClassiCube.git synced 2025-09-07 14:23:29 -04:00
Code Issues Packages Projects Releases Wiki Activity

N64: Save a few cycles here and there (down to 7.5 ms)

Browse Source
This commit is contained in:
UnknownShadow200 2025-07-22 20:35:03 +10:00
parent 21b4fe79aa
commit 06cb87f773
3 changed files with 65 additions and 65 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
misc/n64/gpu.c
View File

@ -166,7 +166,7 @@ static int F2I(float value, int scale) {
e = (raw.i & FLT_EXPONENT_MASK) >> FLT_EXPONENT_SHIFT; e = (raw.i & FLT_EXPONENT_MASK) >> FLT_EXPONENT_SHIFT;
// Ignore denormal, infinity, or large exponents // Ignore denormal, infinity, or large exponents
if (e <= 0 || e >= 160) return 0; if (e <= 0 || e >= 146) return 0;
return value * scale; return value * scale;
} }

106
misc/n64/rsp_gpu.S
View File

@ -123,8 +123,9 @@ GPUCmd_PushRDP:
.func GPUCmd_MatrixLoad .func GPUCmd_MatrixLoad
GPUCmd_MatrixLoad: GPUCmd_MatrixLoad:
#define src t0 #define src s6
#define dst t1 #define dst s7
#define vmat0_i $v02 #define vmat0_i $v02
#define vmat1_i $v03 #define vmat1_i $v03
#define vmat2_i $v04 #define vmat2_i $v04
@ -180,14 +181,14 @@ GPUCmd_MatrixLoad:
# GL_CalcScreenSpace # GL_CalcScreenSpace
# #
# Args: # Args:
# s3 = Destination vertex address # a0 = Destination vertex address
# $v02 = Clip space position (fractional part) # $v02 = Clip space position (fractional part)
# $v03 = Clip space position (integer part) # $v03 = Clip space position (integer part)
# #
################################################################ ################################################################
.func GL_CalcScreenSpace .func GL_CalcScreenSpace
GL_CalcScreenSpace: GL_CalcScreenSpace:
#define dst s3 #define dst a0
#define vcspos_f $v02 #define vcspos_f $v02
#define vcspos_i $v03 #define vcspos_i $v03
#define vinvw_f $v23 #define vinvw_f $v23
@ -197,6 +198,13 @@ GL_CalcScreenSpace:
#define v___ $v29 #define v___ $v29
#define w e3 #define w e3
ldv vcspos_i, SCREEN_VTX_CS_POSi, dst
ldv vcspos_f, SCREEN_VTX_CS_POSf, dst
li t0, %lo(GL_VIEWPORT_SCALE)
ldv vviewscale.e0, 0, t0
ldv vviewoff.e0, 8, t0
# Calculate 32-bit inverse W # Calculate 32-bit inverse W
# TODO: NR? # TODO: NR?
vrcph vinvw_i.w, vcspos_i.w vrcph vinvw_i.w, vcspos_i.w
@ -208,17 +216,17 @@ GL_CalcScreenSpace:
vmadn vscreenpos_f, vcspos_f, vinvw_i.w vmadn vscreenpos_f, vcspos_f, vinvw_i.w
vmadh vscreenpos_i, vcspos_i, vinvw_i.w vmadh vscreenpos_i, vcspos_i, vinvw_i.w
vmudn vscreenpos_f, vscreenpos_f, vviewscale li t0, 0x3F
vmadh vscreenpos_i, vscreenpos_i, vviewscale vmudn v___, vscreenpos_f, vviewscale
vadd vscreenpos_i, vviewoff vmadh v___, vscreenpos_i, vviewscale
vmadh vscreenpos_i, vviewoff, K1
sdv vscreenpos_i, SCREEN_VTX_X ,dst
ssv vcspos_i.w, SCREEN_VTX_W+0 ,dst ssv vcspos_i.w, SCREEN_VTX_W+0 ,dst
ssv vcspos_f.w, SCREEN_VTX_W+2 ,dst ssv vcspos_f.w, SCREEN_VTX_W+2 ,dst
ssv vinvw_i.w, SCREEN_VTX_INVW+0,dst ssv vinvw_i.w, SCREEN_VTX_INVW+0,dst
ssv vinvw_f.w, SCREEN_VTX_INVW+2,dst ssv vinvw_f.w, SCREEN_VTX_INVW+2,dst
sdv vscreenpos_i, SCREEN_VTX_X ,dst
li t0, 0x3F
jr ra jr ra
sb t0, SCREEN_VTX_PADDING(dst) sb t0, SCREEN_VTX_PADDING(dst)
@ -238,14 +246,14 @@ GL_CalcScreenSpace:
# GL_TnL # GL_TnL
# #
# Args: # Args:
# a1 = address of the vertex in DMEM (usually within VERTEX_CACHE)
# a2 = address of the vertex in DMEM (usually within VERTEX_CACHE) # a2 = address of the vertex in DMEM (usually within VERTEX_CACHE)
# a3 = address of the vertex in DMEM (usually within VERTEX_CACHE)
# #
################################################################ ################################################################
.func GL_TnL .func GL_TnL
GL_TnL: GL_TnL:
#define vtx1 a1 #define vtx1 a2
#define vtx2 a2 #define vtx2 a3
#define w e3 #define w e3
#define W e7 #define W e7
@ -260,12 +268,15 @@ GL_TnL:
#define vscreenpos_i $v27 #define vscreenpos_i $v27
#define vscreenpos_f $v28 #define vscreenpos_f $v28
//emux_trace_start
ldv vcspos_i.e0, SCREEN_VTX_CS_POSi,vtx1 ldv vcspos_i.e0, SCREEN_VTX_CS_POSi,vtx1
ldv vcspos_i.e4, SCREEN_VTX_CS_POSi,vtx2 ldv vcspos_i.e4, SCREEN_VTX_CS_POSi,vtx2
ldv vcspos_f.e0, SCREEN_VTX_CS_POSf,vtx1 ldv vcspos_f.e0, SCREEN_VTX_CS_POSf,vtx1
ldv vcspos_f.e4, SCREEN_VTX_CS_POSf,vtx2 ldv vcspos_f.e4, SCREEN_VTX_CS_POSf,vtx2
li t1, 0x3F
# Calculate 32-bit inverse W // Calculate 32-bit inverse W for vertex 1
vrcph vinvw_i.w, vcspos_i.w vrcph vinvw_i.w, vcspos_i.w
vrcpl vinvw_f.w, vcspos_f.w vrcpl vinvw_f.w, vcspos_f.w
vrcph vinvw_i.w, vzero.e0 vrcph vinvw_i.w, vzero.e0
@ -273,11 +284,16 @@ GL_TnL:
vmudn vguard_f, vcspos_f, vguardscale vmudn vguard_f, vcspos_f, vguardscale
vmadh vguard_i, vcspos_i, vguardscale vmadh vguard_i, vcspos_i, vguardscale
# Calculate 32-bit inverse W // Calculate 32-bit inverse W for vertex 2
vrcph vinvw_i.W, vcspos_i.W vrcph vinvw_i.W, vcspos_i.W
vrcpl vinvw_f.W, vcspos_f.W vrcpl vinvw_f.W, vcspos_f.W
vrcph vinvw_i.W, vzero.e0 vrcph vinvw_i.W, vzero.e0
ssv vcspos_i.w, SCREEN_VTX_W+0, vtx1
ssv vcspos_f.w, SCREEN_VTX_W+2, vtx1
ssv vcspos_i.W, SCREEN_VTX_W+0, vtx2
ssv vcspos_f.W, SCREEN_VTX_W+2, vtx2
vmudl v___, vcspos_f, vinvw_f.wwwwWWWW vmudl v___, vcspos_f, vinvw_f.wwwwWWWW
vmadm v___, vcspos_i, vinvw_f.wwwwWWWW vmadm v___, vcspos_i, vinvw_f.wwwwWWWW
vmadn vscreenpos_f, vcspos_f, vinvw_i.wwwwWWWW vmadn vscreenpos_f, vcspos_f, vinvw_i.wwwwWWWW
@ -286,34 +302,31 @@ GL_TnL:
vch v___, vguard_i, vguard_i.wwwwWWWW vch v___, vguard_i, vguard_i.wwwwWWWW
vcl v___, vguard_f, vguard_f.wwwwWWWW vcl v___, vguard_f, vguard_f.wwwwWWWW
vmudn vscreenpos_f, vscreenpos_f, vviewscale vmudn v___, vscreenpos_f, vviewscale
vmadh vscreenpos_i, vscreenpos_i, vviewscale vmadh v___, vscreenpos_i, vviewscale
vadd vscreenpos_i, vviewoff vmadh vscreenpos_i, vviewoff, K1
li t0, 0x3F
sdv vscreenpos_i.e0, SCREEN_VTX_X ,vtx1
ssv vcspos_i.w, SCREEN_VTX_W+0 ,vtx1
ssv vcspos_f.w, SCREEN_VTX_W+2 ,vtx1
ssv vinvw_i.w, SCREEN_VTX_INVW+0,vtx1 ssv vinvw_i.w, SCREEN_VTX_INVW+0,vtx1
ssv vinvw_f.w, SCREEN_VTX_INVW+2,vtx1 ssv vinvw_f.w, SCREEN_VTX_INVW+2,vtx1
sb t0, SCREEN_VTX_PADDING(vtx1)
sdv vscreenpos_i.e4, SCREEN_VTX_X ,vtx2
ssv vcspos_i.W, SCREEN_VTX_W+0 ,vtx2
ssv vcspos_f.W, SCREEN_VTX_W+2 ,vtx2
ssv vinvw_i.W, SCREEN_VTX_INVW+0,vtx2 ssv vinvw_i.W, SCREEN_VTX_INVW+0,vtx2
ssv vinvw_f.W, SCREEN_VTX_INVW+2,vtx2 ssv vinvw_f.W, SCREEN_VTX_INVW+2,vtx2
sb t0, SCREEN_VTX_PADDING(vtx2) cfc2 t0, COP2_CTRL_VCC
cfc2 t0, COP2_CTRL_VCC sdv vscreenpos_i.e0, SCREEN_VTX_X ,vtx1
compressClipCodes sdv vscreenpos_i.e4, SCREEN_VTX_X ,vtx2
sb t1, SCREEN_VTX_PADDING(vtx1)
sb t1, SCREEN_VTX_PADDING(vtx2)
compressClipCodes # TODO move to overlap with vector ops
sb t2, SCREEN_VTX_CLIP_CODE(vtx1) sb t2, SCREEN_VTX_CLIP_CODE(vtx1)
srl t0, t0, 4 srl t0, t0, 4
compressClipCodes compressClipCodes # TODO move to overlap with vector ops
//emux_trace_stop
jr ra jr ra
sb t2, SCREEN_VTX_CLIP_CODE(vtx2) sb t2, SCREEN_VTX_CLIP_CODE(vtx2)
#undef vinvw_f #undef vinvw_f
#undef vinvw_i #undef vinvw_i
#undef vscreenpos_i #undef vscreenpos_i
@ -334,6 +347,10 @@ GL_TnL:
.align 3 .align 3
.func GPUCmd_DrawQuad .func GPUCmd_DrawQuad
GPUCmd_DrawQuad: GPUCmd_DrawQuad:
#define vtx_ptr a0
#define mtx_ptr s2
#define src_ptr s3
#define v___ $v01 #define v___ $v01
#define vst_i $v12 #define vst_i $v12
@ -361,9 +378,7 @@ GPUCmd_DrawQuad:
#define v1_cflags t2 #define v1_cflags t2
#define v2_cflags t3 #define v2_cflags t3
#define v3_cflags t4 #define v3_cflags t4
#define mtx_ptr t5 // t5 is also used by GL_ClipTriangle // t5 is used by GL_ClipTriangle
#define vtx_ptr t6
#define src_ptr t7
addi src_ptr, rspq_dmem_buf_ptr, %lo(RSPQ_DMEM_BUFFER) - 64 addi src_ptr, rspq_dmem_buf_ptr, %lo(RSPQ_DMEM_BUFFER) - 64
li vtx_ptr, %lo(VERTEX_CACHE) li vtx_ptr, %lo(VERTEX_CACHE)
@ -380,7 +395,6 @@ GPUCmd_DrawQuad:
lqv vmtx1_f, 0x50,mtx_ptr lqv vmtx1_f, 0x50,mtx_ptr
lqv vmtx2_f, 0x60,mtx_ptr lqv vmtx2_f, 0x60,mtx_ptr
lqv vmtx3_f, 0x70,mtx_ptr lqv vmtx3_f, 0x70,mtx_ptr
#undef mtx_ptr
// ######################## // ########################
// Vertex 0 and 1 transform // Vertex 0 and 1 transform
@ -405,10 +419,10 @@ GPUCmd_DrawQuad:
vmudm vcspos_i, vcspos_i, K2048 vmudm vcspos_i, vcspos_i, K2048
vmadl vcspos_f, vcspos_f, K2048 vmadl vcspos_f, vcspos_f, K2048
li tmp, %lo(GL_STATE_TEX_SIZE) li t6, %lo(GL_STATE_TEX_SIZE)
lqv vtexsize, 0x00, tmp lqv vtexsize, 0x00, t6
slv vcol.e0, SCREEN_VTX_RGBA + V0_OFFSET, vtx_ptr slv vcol.e0, SCREEN_VTX_RGBA + V0_OFFSET, vtx_ptr
lqv vtexoffset, 0x10, tmp lqv vtexoffset, 0x10, t6
slv vcol.e2, SCREEN_VTX_RGBA + V1_OFFSET, vtx_ptr slv vcol.e2, SCREEN_VTX_RGBA + V1_OFFSET, vtx_ptr
// Calculate and store clipping flags against CS.W. // Calculate and store clipping flags against CS.W.
@ -527,13 +541,13 @@ GPUCmd_DrawQuad:
ldv vguardscale.e0, 0, t0 ldv vguardscale.e0, 0, t0
ldv vguardscale.e4, 0, t0 ldv vguardscale.e4, 0, t0
li a1, %lo(VERTEX_CACHE) + V0_OFFSET li a2, %lo(VERTEX_CACHE) + V0_OFFSET
jal GL_TnL jal GL_TnL
li a2, %lo(VERTEX_CACHE) + V1_OFFSET li a3, %lo(VERTEX_CACHE) + V1_OFFSET
li a1, %lo(VERTEX_CACHE) + V2_OFFSET li a2, %lo(VERTEX_CACHE) + V2_OFFSET
jal GL_TnL jal GL_TnL
li a2, %lo(VERTEX_CACHE) + V3_OFFSET li a3, %lo(VERTEX_CACHE) + V3_OFFSET
// ######################## // ########################
// Guardband check // Guardband check
@ -580,11 +594,19 @@ GPUCmd_DrawQuad:
beqz v1, gl_draw_triangle_end beqz v1, gl_draw_triangle_end
addi s2, -6 addi s2, -6
lhu s5, 0(s1) lhu s5, 0(s1)
jal GL_CalcScreenSpace
lhu a0, 0(s1)
jal GL_CalcScreenSpace
lhu a0, 2(s1)
gl_draw_clipped_triangles_loop: gl_draw_clipped_triangles_loop:
move vtx1, s5 move vtx1, s5
lhu vtx2, 2(s1) lhu vtx2, 2(s1)
lhu vtx3, 4(s1) lhu vtx3, 4(s1)
# TODO do VP transform here
jal GL_CalcScreenSpace
move a0, vtx3
gl_draw_single_triangle: gl_draw_single_triangle:
addi vtx1, SCREEN_VTX_X addi vtx1, SCREEN_VTX_X

22
misc/n64/rsp_gpu_clipping.inc
View File

@ -18,7 +18,6 @@ CACHE_OFFSETS: .half 2,4,6,8, 10,12,14,16, 18,20
.section .bss.gl_clipping .section .bss.gl_clipping
.align 4
CLIP_CACHE: .dcb.b SCREEN_VTX_SIZE * CLIPPING_CACHE_SIZE CLIP_CACHE: .dcb.b SCREEN_VTX_SIZE * CLIPPING_CACHE_SIZE
CLIP_CACHE_END: CLIP_CACHE_END:
@ -342,32 +341,11 @@ gl_clip_plane_loop_end:
blt plane_flag, (1<<CLIPPING_PLANE_COUNT), gl_clip_plane_loop blt plane_flag, (1<<CLIPPING_PLANE_COUNT), gl_clip_plane_loop
addi plane, CLIPPING_PLANE_SIZE addi plane, CLIPPING_PLANE_SIZE
#define cache_vtx s3
#define cache_end s5
# Calculate screen space values for new vertices (in the clip cache)
# TODO: maybe iterate over out_list instead
li cache_vtx, %lo(CLIP_CACHE)
li cache_end, %lo(CLIP_CACHE_END) - SCREEN_VTX_SIZE
gl_clip_finalize_loop:
lbu t0, SCREEN_VTX_PADDING(cache_vtx)
neg t0
# Only calculate screen space values if the vertex is actually used
ldv vint_i, SCREEN_VTX_CS_POSi,cache_vtx
bltzal t0, GL_CalcScreenSpace
ldv vint_f, SCREEN_VTX_CS_POSf,cache_vtx
blt cache_vtx, cache_end, gl_clip_finalize_loop
addi cache_vtx, SCREEN_VTX_SIZE
gl_clip_return: gl_clip_return:
# Done! # Done!
jr ra2 jr ra2
add s2, out_list, out_count add s2, out_list, out_count
#undef cache_vtx
#undef cache_end
#undef clip_flags #undef clip_flags
#undef plane_flag #undef plane_flag
#undef in_count #undef in_count