# Mode7 rotozoomer: PowerPC 604e guest code submits an affine texture map to # the IE blitter. The CPU calculates signed fixed-point origins and step # vectors, renders into a back buffer, then presents at the next vblank edge. # # Assemble from the repository root with: # make rotozoomer-ppc # # (c) 2024-2026 Zayn Otley - GPLv3 or later .include "ieppc.inc" # ============================================================================ # CONSTANTS # ============================================================================ # The texture occupies 256*256*4 bytes. Each framebuffer occupies 640*480*4 # bytes. These areas are separate from the programme, stack and MMIO. .set TEXTURE_BASE,0x600000 .set BACK_BUFFER_A,0x900000 .set BACK_BUFFER_B,0xB00000 .set RENDER_W,640 .set RENDER_H,480 .set TEX_STRIDE,1024 .set OUTPUT_STRIDE,2560 .set ANGLE_INC,313 .set SCALE_INC,104 # ============================================================================ # ENTRY POINT # ============================================================================ # The flat image loads at PROGRAM_START (0x4000). VASM uses this origin for # label addresses; the file begins with the first instruction, without padding. # # Register allocation throughout the programme: # r20 = angle accumulator r21 = scale accumulator # r22 = draw framebuffer r23 = video register base # r24 = sine table r25 = reciprocal table # r26 = CA r27 = SA # r28 = U0 r29 = V0 # r3-r12 = scratch # # BL saves the return address in LR, unlike the M68K's stack-based BSR. # Each subroutine is a leaf, or ends with a tail branch to another routine, # so no nested call overwrites LR. R1 still receives an aligned stack top. .org PROGRAM_START start: lis r1,STACK_TOP>>16 lis r23,VIDEO_CTRL>>16 li r3,VIDEO_CTRL_ENABLE stw r3,VIDEO_CTRL-VIDEO_CTRL(r23) li r3,0 stw r3,VIDEO_MODE-VIDEO_CTRL(r23) stw r3,BLT_FLAGS-VIDEO_CTRL(r23) lis r3,VRAM_START>>16 stw r3,VIDEO_FB_BASE-VIDEO_CTRL(r23) bl load_texture bl setup_mode7 # --- Initialise animation state --- # The low eight bits retain fractional steps. The high eight bits # select a table entry. Both accumulators wrap after 65536 steps. li r20,0 li r21,0 lis r22,BACK_BUFFER_A>>16 lis r24,sine_table@h ori r24,r24,sine_table@l lis r25,recip_table@h ori r25,r25,recip_table@l # --- Start TED music playback --- # The player consumes the embedded register dump independently of # the CPU. Control value 5 sets START (bit 0) and LOOP (bit 2). lis r3,ted_data@h ori r3,r3,ted_data@l stw r3,TED_PLAY_PTR-VIDEO_CTRL(r23) lis r3,(ted_data_end-ted_data)@h ori r3,r3,(ted_data_end-ted_data)@l stw r3,TED_PLAY_LEN-VIDEO_CTRL(r23) li r3,5 stw r3,TED_PLAY_CTRL-VIDEO_CTRL(r23) # ============================================================================ # MAIN LOOP # ============================================================================ # Render first, wait for a fresh vblank edge, then display the completed # buffer. The next frame writes into the other buffer. main_loop: bl compute_frame bl render_mode7 bl wait_vsync present_frame: stw r22,VIDEO_FB_BASE-VIDEO_CTRL(r23) # The two buffer addresses differ only in bit 21. xoris r22,r22,0x20 bl advance_animation b main_loop # ============================================================================ # LOAD TEXTURE - 256x256 RGBA from Embedded Raw Data # ============================================================================ # The raw image is a byte sequence, copied unchanged by the blitter. PPC # halfword tables are big-endian, but texture bytes need no conversion. # Numeric MMIO registers accept ordinary STW/LWZ accesses: writing 256 # writes the number 256, without a byte-reversing store. load_texture: li r3,BLT_OP_COPY stw r3,BLT_OP-VIDEO_CTRL(r23) lis r3,texture_data@h ori r3,r3,texture_data@l stw r3,BLT_SRC-VIDEO_CTRL(r23) lis r3,TEXTURE_BASE>>16 stw r3,BLT_DST-VIDEO_CTRL(r23) li r3,256 stw r3,BLT_WIDTH-VIDEO_CTRL(r23) stw r3,BLT_HEIGHT-VIDEO_CTRL(r23) li r3,TEX_STRIDE stw r3,BLT_SRC_STRIDE-VIDEO_CTRL(r23) stw r3,BLT_DST_STRIDE-VIDEO_CTRL(r23) b submit_blit # ============================================================================ # SET UP MODE7 - Parameters Shared by Every Frame # ============================================================================ # Source stride is 256 pixels * 4 bytes. Destination stride is 640 * 4. # Masks of 255 wrap the texture coordinates at each 256-texel boundary. setup_mode7: li r3,BLT_OP_MODE7 stw r3,BLT_OP-VIDEO_CTRL(r23) lis r3,TEXTURE_BASE>>16 stw r3,BLT_SRC-VIDEO_CTRL(r23) li r3,RENDER_W stw r3,BLT_WIDTH-VIDEO_CTRL(r23) li r3,RENDER_H stw r3,BLT_HEIGHT-VIDEO_CTRL(r23) li r3,OUTPUT_STRIDE stw r3,BLT_DST_STRIDE-VIDEO_CTRL(r23) li r3,255 stw r3,BLT_MODE7_TEX_W-VIDEO_CTRL(r23) stw r3,BLT_MODE7_TEX_H-VIDEO_CTRL(r23) blr # ============================================================================ # COMPUTE FRAME - Rotation and Zoom in Signed 16.16 Fixed-Point # ============================================================================ # Sine entries represent -1.0 to +1.0 as -256 to +256. The reciprocal # table holds round(256 / (0.5 + sin(i * 2*pi / 256) * 0.3)). Multiplying # these two 8.8 values gives a 16.16 result directly, without a shift. # # LHAX sign-extends the sine halfword; LHZX zero-extends the reciprocal. # MULLW keeps the low 32 product bits, which contain the complete result: # the largest magnitude is 256 * 1280 = 327680. compute_frame: srwi r3,r20,8 slwi r4,r3,1 lhax r5,r24,r4 # sin(angle) addi r3,r3,64 # cosine is sine a quarter turn ahead andi. r3,r3,255 slwi r4,r3,1 lhax r6,r24,r4 # cos(angle) srwi r3,r21,8 slwi r4,r3,1 lhzx r7,r25,r4 # reciprocal zoom factor mullw r26,r6,r7 # CA = cos * reciprocal mullw r27,r5,r7 # SA = sin * reciprocal # --- Centre the transformation --- # The screen centre (320,240) maps to texture centre (128,128). # U0 = (128 << 16) - CA*320 + SA*240 # V0 = (128 << 16) - SA*320 - CA*240 # MULLI accepts a full 32-bit register and a signed immediate. # All intermediate values fit in signed 32 bits. mulli r3,r26,320 mulli r4,r27,240 lis r28,128 subf r28,r3,r28 add r28,r28,r4 mulli r3,r27,320 mulli r4,r26,240 lis r29,128 subf r29,r3,r29 subf r29,r4,r29 blr # ============================================================================ # RENDER MODE7 - Submit the Six Affine Parameters # ============================================================================ # For screen pixel (x,y), the blitter samples: # U = U0 + CA*x - SA*y # V = V0 + SA*x + CA*y # The integer parts select a texel; the masks provide repeating tiles. render_mode7: stw r22,BLT_DST-VIDEO_CTRL(r23) stw r28,BLT_MODE7_U0-VIDEO_CTRL(r23) stw r29,BLT_MODE7_V0-VIDEO_CTRL(r23) stw r26,BLT_MODE7_DU_COL-VIDEO_CTRL(r23) stw r27,BLT_MODE7_DV_COL-VIDEO_CTRL(r23) neg r3,r27 stw r3,BLT_MODE7_DU_ROW-VIDEO_CTRL(r23) stw r26,BLT_MODE7_DV_ROW-VIDEO_CTRL(r23) submit_blit: li r3,1 stw r3,BLT_CTRL-VIDEO_CTRL(r23) wait_blit: # BLT_CTRL bit 1 is BUSY. BLT_STATUS bit 1 is DONE, not BUSY. lwz r3,BLT_CTRL-VIDEO_CTRL(r23) andi. r3,r3,2 bne wait_blit lwz r3,BLT_STATUS-VIDEO_CTRL(r23) andi. r3,r3,1 # ERR: leave the last displayed frame bne blit_error blr blit_error: b blit_error # ============================================================================ # WAIT FOR VSYNC - Observe a Fresh Vertical Blank Edge # ============================================================================ # Wait for any current blank to end, then wait for the next one to begin. # This prevents two frames from being presented within the same blank. wait_vsync: lwz r3,VIDEO_STATUS-VIDEO_CTRL(r23) andi. r3,r3,STATUS_VBLANK bne wait_vsync wait_vsync_start: lwz r3,VIDEO_STATUS-VIDEO_CTRL(r23) andi. r3,r3,STATUS_VBLANK beq wait_vsync_start blr # ============================================================================ # ADVANCE ANIMATION - Update Fractional Accumulators # ============================================================================ # 313 gives about 0.03 radians per frame; 104 gives about 0.01. Masking # to 16 bits wraps the table index while retaining the fractional part. advance_animation: addi r20,r20,ANGLE_INC andi. r20,r20,65535 addi r21,r21,SCALE_INC andi. r21,r21,65535 blr # ============================================================================ # SINE TABLE - 256 Entries, Signed 16-bit # ============================================================================ # round(sin(i * 2*pi / 256) * 256), stored as big-endian halfwords. # Entries 0,64,128,192 are 0,+256,0,-256 respectively. .balign 4 sine_table: .short 0,6,13,19,25,31,38,44,50,56,62,68,74,80,86,92 .short 98,104,109,115,121,126,132,137,142,147,152,157,162,167,172,177 .short 181,185,190,194,198,202,206,209,213,216,220,223,226,229,231,234 .short 237,239,241,243,245,247,248,250,251,252,253,254,255,255,256,256 .short 256,256,256,255,255,254,253,252,251,250,248,247,245,243,241,239 .short 237,234,231,229,226,223,220,216,213,209,206,202,198,194,190,185 .short 181,177,172,167,162,157,152,147,142,137,132,126,121,115,109,104 .short 98,92,86,80,74,68,62,56,50,44,38,31,25,19,13,6 .short 0,-6,-13,-19,-25,-31,-38,-44,-50,-56,-62,-68,-74,-80,-86,-92 .short -98,-104,-109,-115,-121,-126,-132,-137,-142,-147,-152,-157,-162,-167,-172,-177 .short -181,-185,-190,-194,-198,-202,-206,-209,-213,-216,-220,-223,-226,-229,-231,-234 .short -237,-239,-241,-243,-245,-247,-248,-250,-251,-252,-253,-254,-255,-255,-256,-256 .short -256,-256,-256,-255,-255,-254,-253,-252,-251,-250,-248,-247,-245,-243,-241,-239 .short -237,-234,-231,-229,-226,-223,-220,-216,-213,-209,-206,-202,-198,-194,-190,-185 .short -181,-177,-172,-167,-162,-157,-152,-147,-142,-137,-132,-126,-121,-115,-109,-104 .short -98,-92,-86,-80,-74,-68,-62,-56,-50,-44,-38,-31,-25,-19,-13,-6 # ============================================================================ # RECIPROCAL TABLE - 256 Entries, Unsigned 16-bit # ============================================================================ # round(256 / (0.5 + sin(i * 2*pi / 256) * 0.3)), range 320 to 1280. # The denominator stays between 0.2 and 0.8, so zoom remains positive. recip_table: .short 512,505,497,490,484,477,471,464,458,453,447,441,436,431,426,421 .short 416,412,407,403,399,395,391,388,384,381,377,374,371,368,365,362 .short 359,357,354,352,350,348,345,343,342,340,338,336,335,333,332,331 .short 329,328,327,326,325,324,324,323,322,322,321,321,321,320,320,320 .short 320,320,320,320,321,321,321,322,322,323,324,324,325,326,327,328 .short 329,331,332,333,335,336,338,340,342,343,345,348,350,352,354,357 .short 359,362,365,368,371,374,377,381,384,388,391,395,399,403,407,412 .short 416,421,426,431,436,441,447,453,458,464,471,477,484,490,497,505 .short 512,520,528,536,544,553,561,571,580,589,599,610,620,631,642,653 .short 665,676,689,701,714,727,740,754,768,782,797,812,827,842,858,873 .short 889,905,922,938,955,972,988,1005,1022,1038,1055,1071,1087,1103,1119,1134 .short 1149,1163,1177,1190,1202,1214,1225,1235,1244,1252,1260,1266,1271,1275,1278,1279 .short 1280,1279,1278,1275,1271,1266,1260,1252,1244,1235,1225,1214,1202,1190,1177,1163 .short 1149,1134,1119,1103,1087,1071,1055,1038,1022,1005,988,972,955,938,922,905 .short 889,873,858,842,827,812,797,782,768,754,740,727,714,701,689,676 .short 665,653,642,631,620,610,599,589,580,571,561,553,544,536,528,520 # ============================================================================ # TEXTURE AND MUSIC DATA # ============================================================================ # Keep the RGBA image and TED register dump as raw bytes. Neither asset # contains PPC words, so the assembler must not change their byte order. .balign 4 texture_data: .incbin "../assets/rotozoomtexture_ppc.raw" .balign 4 ted_data: .incbin "../assets/music/chromatic_admiration.ted" ted_data_end: