From 19fd014e88d20b637691e5103c98661149c4b1df Mon Sep 17 00:00:00 2001 From: Connor Olding Date: Sat, 1 Sep 2018 05:20:28 +0200 Subject: [PATCH] add dummy constants for direct DMEM accesses --- F3DZEX.asm | 293 ++++++++++++++++++++++++++++++++--------------------- 1 file changed, 176 insertions(+), 117 deletions(-) diff --git a/F3DZEX.asm b/F3DZEX.asm index 8c53347..271128c 100644 --- a/F3DZEX.asm +++ b/F3DZEX.asm @@ -59,11 +59,11 @@ constant FUL($0008) // DMA Full constant IOF($0010) // IO Full constant STP($0020) // Single Step constant IOB($0040) // Interrupt On Break -constant SG0($0080) // Signal 0 Set -constant SG1($0100) // Signal 1 Set -constant SG2($0200) // Signal 2 Set -constant SG3($0400) // Signal 3 Set -constant SG4($0800) // Signal 4 Set +constant SG0($0080) // Signal 0 Set (Yield) +constant SG1($0100) // Signal 1 Set (Yielded) +constant SG2($0200) // Signal 2 Set (Task Done) +constant SG3($0400) // Signal 3 Set (RSP Signal) +constant SG4($0800) // Signal 4 Set (CPU Signal) constant SG5($1000) // Signal 5 Set constant SG6($2000) // Signal 6 Set constant SG7($4000) // Signal 7 Set @@ -115,6 +115,64 @@ constant TASK_DATA_SIZE(0xFF4) constant TASK_YIELD_DATA_PTR(0xFF8) constant TASK_YIELD_DATA_SIZE(0xFFC) +// various unknown offsets +constant DMEM_000(0x000) +constant DMEM_002(0x002) +constant DMEM_004(0x004) +constant DMEM_008(0x008) +constant DMEM_00A(0x00A) +constant DMEM_00C(0x00C) +constant DMEM_010(0x010) +constant DMEM_012(0x012) +constant DMEM_014(0x014) +constant DMEM_020(0x020) +constant DMEM_022(0x022) +constant DMEM_024(0x024) +constant DMEM_028(0x028) +constant DMEM_02A(0x02A) +constant DMEM_02C(0x02C) +constant DMEM_030(0x030) +constant DMEM_032(0x032) +constant DMEM_034(0x034) +constant DMEM_080(0x080) +constant DMEM_088(0x088) +constant DMEM_090(0x090) +constant DMEM_098(0x098) +constant DMEM_0A0(0x0A0) +constant DMEM_0A8(0x0A8) +constant DMEM_0B0(0x0B0) +constant DMEM_0B8(0x0B8) +constant DMEM_0C0(0x0C0) +constant DMEM_0C4(0x0C4) +constant DMEM_0C8(0x0C8) +constant DMEM_0CC(0x0CC) +constant DMEM_0D0(0x0D0) +constant DMEM_0D8(0x0D8) +constant DMEM_0DC(0x0DC) +constant DMEM_0DE(0x0DE) +constant DMEM_0E0(0x0E0) +constant DMEM_0E8(0x0E8) +constant DMEM_0F0(0x0F0) +constant DMEM_0F4(0x0F4) +constant DMEM_1B0(0x1B0) +constant DMEM_1C0(0x1C0) +constant DMEM_1D0(0x1D0) +constant DMEM_1D9(0x1D9) +constant DMEM_1DC(0x1DC) +constant DMEM_1DD(0x1DD) +constant DMEM_1E6(0x1E6) +constant DMEM_1E7(0x1E7) +constant DMEM_1EC(0x1EC) +constant DMEM_1EE(0x1EE) +constant DMEM_2E0(0x2E0) +constant DMEM_2E8(0x2E8) +constant DMEM_3CC(0x3CC) +constant DMEM_410(0x410) +constant DMEM_418(0x418) +constant DMEM_920(0x920) +constant DMEM_BF8(0xBF8) +constant DMEM_BFC(0xBFC) + output "bin/F3DZEX2.boot.bin", create fill 0xD0 @@ -126,8 +184,8 @@ base 0x1000 label_1008: lw v0, TASK_UCODE-TASK_START(at) - addi v1, r0, 0x0F7F // copy 0xF80 bytes - addi a3, r0, 0x1080 // to 0xA4001080 + addi v1, r0, 0x2000 - label_1080 - 1 // copy 0xF80 bytes + addi a3, r0, label_1080 // to 0xA4001080 mtc0 a3, MEM_ADDR mtc0 v0, DRAM_ADDR mtc0 v1, RD_LEN // start the DMA @@ -166,6 +224,7 @@ label_1054: mfc0 v0, RDP_STATUS // note: this marks 0x80, meaning everything below gets overwritten later. +label_1080: andi v0, v0, 0x0100 bgtz v0,func_103C nop @@ -200,14 +259,14 @@ base 0x1080 vxor vec0,vec0,vec0 // clear vector 0 label_1084: - lqv vec31[e0], 0x1B0(r0) // read some? data from DMEM+1B0 + lqv vec31[e0], DMEM_1B0(r0) func_1088: - lqv vec30[e0], 0x1C0(r0) // read the next row, too + lqv vec30[e0], DMEM_1C0(r0) addi s7, r0, 0xBA8 vadd vec1,vec0,vec0 // multiply vector 0 by 2 addi s6, r0, 0xD00 vsub vec1,vec0,vec31[e8] - lw t3, 0x0F0(r0) // TASK_DRAM_STACK gets written here? + lw t3, DMEM_0F0(r0) // TASK_DRAM_STACK gets written here? lw t4, TASK_FLAGS(r0) addi at, r0, SG1_CLR | SG2_CLR beqz t3,+ @@ -218,7 +277,7 @@ func_1088: sw r0, TASK_FLAGS(r0) j label_1168 - lw k0, 0xBF8(r0) + lw k0, DMEM_BF8(r0) + mfc0 t3, RDP_STATUS @@ -252,27 +311,27 @@ func_1088: mtc0 v0, COMMAND_START mtc0 v0, COMMAND_END + - sw v0, 0x0F0(r0) - lw t3, 0x0F4(r0) + sw v0, DMEM_0F0(r0) + lw t3, DMEM_0F4(r0) bnez t3, label_1130 lw t3, TASK_DRAM_STACK(r0) - sw t3, 0x0F4(r0) + sw t3, DMEM_0F4(r0) label_1130: lw at, TASK_UCODE(r0) - lw v0, 0x2E0(r0) - lw v1, 0x2E8(r0) - lw a0, 0x410(r0) - lw a1, 0x418(r0) + lw v0, DMEM_2E0(r0) + lw v1, DMEM_2E8(r0) + lw a0, DMEM_410(r0) + lw a1, DMEM_418(r0) add v0, v0, at add v1, v1, at - sw v0, 0x2E0(r0) - sw v1, 0x2E8(r0) + sw v0, DMEM_2E0(r0) + sw v1, DMEM_2E8(r0) func_1154: add a0, a0, at add a1, a1, at - sw a0, 0x410(r0) - sw a1, 0x418(r0) + sw a0, DMEM_410(r0) + sw a1, DMEM_418(r0) lw k0, TASK_DATA_PTR(r0) label_1168: addi t3, r0, 0x2E8 @@ -285,7 +344,7 @@ label_1178: ori t8, k0, 0 jal func_1FD8 // load in the DList from TASK_DATA_PTR? - addiu s4, r0, 0x0920 // DMA destination: DMEM+$920 + addiu s4, r0, DMEM_920 // DMA destination label_1188: addiu k0, k0, 0x00A8 @@ -316,22 +375,22 @@ label_11A4: j func_1FD8 addi ra, r0, label_1190 - lw t3, 0x1EC(r0) + lw t3, DMEM_1EC(r0) and t3, t3, t9 or t3, t3, t8 j label_1194 - sw t3, 0x1EC(r0) + sw t3, DMEM_1EC(r0) label_11EC: - lbu at, 0x0DE(r0) + lbu at, DMEM_0DE(r0) beqz at, label_1FAC addi at, at, 0xFFFC j label_1020 lw k0, 0x0138(at) - ldv vec29[e0], 0x0D0(r0) - lw t9, 0x0D8(r0) + ldv vec29[e0], DMEM_0D0(r0) + lw t9, DMEM_0D8(r0) addi s7, s7, COMMAND_START sdv vec29[e0], 0x3F8(s7) label_1210: @@ -349,13 +408,13 @@ func_1224: srl t8, t8, 8 jr ra add t8, t8, t3 - sw t9, 0x0C8(r0) + sw t9, DMEM_0C8(r0) j label_1210 - sw t8, 0x0CC(r0) + sw t8, DMEM_0CC(r0) - sw t9, 0x0C0(r0) + sw t9, DMEM_0C0(r0) j label_1210 - sw t8, 0x0C4(r0) + sw t8, DMEM_0C4(r0) label_1258: addi ra, r0, label_1194 @@ -365,7 +424,7 @@ label_125C: label_1264: mfc0 t4, DMA_BUSY - lw t8, 0x0F0(r0) + lw t8, DMEM_0F0(r0) addiu s3, t3, 0x0158 bnez t4,label_1264 lw t4, TASK_OUTPUT_BUFF_SIZE(r0) @@ -397,7 +456,7 @@ label_1264: + add t3, t8, s3 - sw t3, 0x0F0(r0) + sw t3, DMEM_0F0(r0) addi s3, s3, 0xFFFF addi s4, s6, 0xDEA8 xori s6, s6, 0x0208 @@ -418,7 +477,7 @@ label_12E4: sh v0, 0x03CC(s2) sh v1, 0x03CE(s2) sh r0, 0x03D0(s2) - lw sp, 0x3CC(r0) + lw sp, DMEM_3CC(r0) label_1308: lw t1, 0x03F8(a1) lw s0, 0x0024(v1) @@ -455,7 +514,7 @@ label_1320: vmadn vec10,vec6,vec3[e0] vmadh vec11,vec7,vec3[e0] vaddc vec8,vec8,vec8[e2] - lqv vec25[e0], 0x1D0(r0) + lqv vec25[e0], DMEM_1D0(r0) vadd vec9,vec9,vec9[e2] vaddc vec10,vec10,vec10[e2] vadd vec11,vec11,vec11[e2] @@ -545,7 +604,7 @@ label_14A8: lhu v1, 0x03CE(s5) bnez a1, label_1308 addi a1, a1, 0xFFFC - sw r0, 0x3CC(r0) + sw r0, DMEM_3CC(r0) - lhu at, 0x03CA(s2) @@ -561,7 +620,7 @@ label_14A8: + jr fp - sw sp, 0x3CC(r0) + sw sp, DMEM_3CC(r0) nops(0x1780) @@ -573,12 +632,12 @@ label_14A8: jal func_1FD8 addi s3, at, 0xFFFF - lhu a1, 0x1EC(r0) + lhu a1, DMEM_1EC(r0) srl at, at, 3 sub t7, t9, at lhu t7, 0x0380(t7) ori t6, s4, 0x0 - lbu t0, 0x1D9(r0) + lbu t0, DMEM_1D9(r0) andi a2, a1, 0x2 bnez a2, label_12D8 andi a3, a1, 0x1 @@ -587,32 +646,32 @@ label_17BC: bnez t0,+ sll a3, a3, 3 - sb t9, 0x1D9(r0) + sb t9, DMEM_1D9(r0) addi s5, r0, 0x0040 addi s4, r0, 0x0 jal func_1088 addi s3, r0, 0x0080 + - lqv vec8[e0], 0x80(r0) - lqv vec10[e0], 0x90(r0) - lqv vec12[e0], 0xA0(r0) - lqv vec14[e0], 0xB0(r0) + lqv vec8[e0], DMEM_080(r0) + lqv vec10[e0], DMEM_090(r0) + lqv vec12[e0], DMEM_0A0(r0) + lqv vec14[e0], DMEM_0B0(r0) vadd vec9,vec8,vec0[e8] - ldv vec9[e0], 0x88(r0) + ldv vec9[e0], DMEM_088(r0) vadd vec11,vec10,vec0[e8] - ldv vec11[e0], 0x98(r0) + ldv vec11[e0], DMEM_098(r0) vadd vec13,vec12,vec0[e8] - ldv vec13[e0], 0xA8(r0) + ldv vec13[e0], DMEM_0A8(r0) vadd vec15,vec14,vec0[e8] - ldv vec15[e0], 0xB8(r0) - ldv vec8[e8], 0x80(r0) - ldv vec10[e8], 0x90(r0) + ldv vec15[e0], DMEM_0B8(r0) + ldv vec8[e8], DMEM_080(r0) + ldv vec10[e8], DMEM_090(r0) jal func_19F4 - ldv vec12[e8], 0xA0(r0) + ldv vec12[e8], DMEM_0A0(r0) jal func_1FC8 - ldv vec14[e8], 0xB0(r0) + ldv vec14[e8], DMEM_0B0(r0) ldv vec20[e0], 0x0(t6) vmov vec16[e13],vec21[e9] @@ -740,18 +799,18 @@ label_1870: func_19F4: addi t5, r0, 0x0180 - ldv vec16[e0], 0xE0(r0) - ldv vec16[e8], 0xE0(r0) + ldv vec16[e0], DMEM_0E0(r0) + ldv vec16[e8], DMEM_0E0(r0) llv vec29[e0], 0x60(t5) - ldv vec17[e0], 0xE8(r0) - ldv vec17[e8], 0xE8(r0) + ldv vec17[e0], DMEM_0E8(r0) + ldv vec17[e8], DMEM_0E8(r0) vlt vec19,vec31,vec31[e11] vsub vec21,vec0,vec16[e0] llv vec18[e4], 0x68(t5) vmrg vec16,vec16,vec29[e8] llv vec18[e12], 0x68(t5) vmrg vec19,vec0,vec1[e8] - llv vec18[e8], 0xDC(r0) + llv vec18[e8], DMEM_0DC(r0) vmrg vec17,vec17,vec29[e9] lsv vec18[e10], 0x6(t5) vmov vec16[e9],vec21[e9] @@ -798,10 +857,10 @@ func_1A7C: vlt vec13,vec2,vec4[e9] vmrg vec14,vec6,vec4[e0] bnez t3, label_1FD4 - lbu t3, 0x1EE(r0) + lbu t3, DMEM_1EE(r0) vmudh vec29,vec10,vec12[e9] - lw t4, 0x3CC(r0) + lw t4, DMEM_3CC(r0) vmadh vec29,vec12,vec11[e9] or a1, a1, a2 vge vec2,vec2,vec4[e9] @@ -831,7 +890,7 @@ func_1A7C: vsub vec8,vec10,vec14[e0] mfc2 v1,vec10[e12] vsub vec11,vec14,vec2[e0] - lw a2, 0x1EC(r0) + lw a2, DMEM_1EC(r0) vsub vec12,vec14,vec10[e0] llv vec13[e0], 0x20(at) vsub vec15,vec10,vec2[e0] @@ -875,7 +934,7 @@ label_1BC0: vrcph vec22[e11],vec8[e9] lw t0, 0x0020(v1) vmudl vec18,vec18,vec30[e11] - lbu t1, 0x1E7(r0) + lbu t1, DMEM_1E7(r0) vmudl vec19,vec19,vec30[e11] sub t3, a1, a3 vmudl vec21,vec21,vec30[e11] @@ -899,7 +958,7 @@ label_1BC0: vmudm vec29,vec25,vec20[e0] mfc2 a1,vec17[e1] vmadl vec29,vec15,vec20[e0] - lbu a3, 0x1E6(r0) + lbu a3, DMEM_1E6(r0) vmadn vec20,vec15,vec22[e0] lsv vec19[e14], 0x1C(v0) vmadh vec15,vec25,vec22[e0] @@ -1085,7 +1144,7 @@ label_1BC0: lh t9, 6(t9) sub v0, t9, t8 bgez v0, label_1194 - lw t8, 0x0D8(r0) + lw t8, DMEM_0D8(r0) j label_1008 lbu at, 0x09C1(k1) @@ -1133,18 +1192,18 @@ base 0x1000 bgezal t4, label_1264 nop jal func_1FC8 - lw t8, 0x0F0(r0) + lw t8, DMEM_0F0(r0) bltz at, label_1084 mtc0 t8, COMMAND_END bnez at, label_1060 add k0, k0, k1 lw t8, 0x09C4(k1) - sw k0, 0xFF0(r0) - sw t8, 0xFD0(r0) + sw k0, TASK_DATA_PTR(r0) + sw t8, TASK_UCODE(r0) addiu s4, r0, 0x1080 jal func_1FD8 addi s3, r0, 0x0F47 - lw t8, 0x0D8(r0) + lw t8, DMEM_0D8(r0) addiu s4, r0, 0x0180 andi s3, t9, 0x0FFF add t8, t8, s4 @@ -1154,11 +1213,11 @@ base 0x1000 addi ra, r0, label_1084 label_1060: - lw t3, 0xFD0(r0) - sw k0, 0xBF8(r0) - sw t3, 0xBFC(r0) + lw t3, TASK_UCODE(r0) + sw k0, DMEM_BF8(r0) + sw t3, DMEM_BFC(r0) addi t4, r0, 0x5000 - lw t8, 0xFF8(r0) + lw t8, TASK_YIELD_DATA_PTR(r0) addi s4, r0, 0x8000 addi s3, r0, 0x0BFF j func_1FD8 @@ -1173,7 +1232,7 @@ label_1060: // this gets loaded by func_1FB4 from DMEM+$2E8 base 0x1000 - lbu at, 0x0DE(r0) + lbu at, DMEM_0DE(r0) sll v0, t9, 15 jal func_1224 add v1, k0, k1 @@ -1182,7 +1241,7 @@ base 0x1000 sw v1, 0x0138(at) addi at, at, 0x0004 j label_1178 - sb at, 0x0DE(r0) + sb at, DMEM_0DE(r0) addi t3, r0, 0x1140 sw t9, 0xF0A4(t3) @@ -1195,8 +1254,8 @@ base 0x1000 j label_1194 sw t8, 0x0000(at) - lw t3, 0x0F4(r0) - lw v0, 0xFE0(r0) + lw t3, DMEM_0F4(r0) + lw v0, TASK_DRAM_STACK(r0) sub t8, t3, t8 sub at, t8, v0 bgez at, label_1068 @@ -1204,9 +1263,9 @@ base 0x1000 ori t8, v0, 0x0000 label_1068: beq t8, t3, label_1194 - sw t8, 0x0F4(r0) + sw t8, DMEM_0F4(r0) j label_1124 - sw r0, 0x1D9(r0) + sw r0, DMEM_1D9(r0) lhu s3, 0x02F2(at) jal func_1FC8 @@ -1244,16 +1303,16 @@ label_10A0: andi t3, t9, 0x0005 bnez t3, label_1118 andi v0, t9, 0x0002 - lw t8, 0x0F4(r0) + lw t8, DMEM_0F4(r0) addi s4, r0, 0xE000 jal func_1FD8 addi s3, r0, 0x003F addi t8, t8, 0x0040 - sw t8, 0x0F4(r0) + sw t8, DMEM_0F4(r0) lw t8, 0x09C4(k1) label_1118: add t4, t4, v0 - sw r0, 0x1D9(r0) + sw r0, DMEM_1D9(r0) jal func_1224 label_1124: andi at, t9, 0x00FE @@ -1272,15 +1331,15 @@ label_1124: and v1, v1, v0 or v1, v1, t8 sw v1, 0xEF8C(t3) - lw t9, 0x0C8(r0) + lw t9, DMEM_0C8(r0) j label_1210 - lw t8, 0x0CC(r0) + lw t8, DMEM_0CC(r0) base 0x12D8 - lbu t3, 0x1DC(r0) + lbu t3, DMEM_1DC(r0) j label_12F4 - lbu a2, 0x1DD(r0) + lbu a2, DMEM_1DD(r0) ori fp, ra, 0x0000 addi t3, r0, 0x0418 @@ -1290,27 +1349,27 @@ base 0x12D8 label_12F4: bnez t3, label_17BC addi a2, a2, 0x0168 - sb t9, 0x1DC(r0) - lqv vec12[e0], 0x20(r0) - lqv vec8[e0], 0x0(r0) - lsv vec13[e2], 0x2A(r0) - lsv vec9[e2], 0xA(r0) + sb t9, DMEM_1DC(r0) + lqv vec12[e0], DMEM_020(r0) + lqv vec8[e0], DMEM_000(r0) + lsv vec13[e2], DMEM_02A(r0) + lsv vec9[e2], DMEM_00A(r0) vmov vec13[e8],vec12[e9] - lsv vec14[e4], 0x34(r0) + lsv vec14[e4], DMEM_034(r0) vmov vec9[e8],vec8[e9] - lsv vec10[e4], 0x14(r0) + lsv vec10[e4], DMEM_014(r0) vmov vec14[e8],vec12[e10] addi s4, r0, 0x0150 vmov vec10[e8],vec8[e10] lpv vec7[e0], 0xA8(s4) vmov vec14[e9],vec12[e14] - lsv vec13[e4], 0x32(r0) + lsv vec13[e4], DMEM_032(r0) vmov vec10[e9],vec8[e14] - lsv vec9[e4], 0x12(r0) + lsv vec9[e4], DMEM_012(r0) vmov vec12[e9],vec12[e12] - lsv vec12[e4], 0x30(r0) + lsv vec12[e4], DMEM_030(r0) vmov vec8[e9],vec8[e12] - lsv vec8[e4], 0x10(r0) + lsv vec8[e4], DMEM_010(r0) label_1350: vmudn vec29,vec13,vec7[e9] vmadh vec29,vec9,vec7[e9] @@ -1380,8 +1439,8 @@ label_140C: addi t1, t1, 0xFFE8 label_144C: - lqv vec31[e0], 0x1B0(r0) - lqv vec30[e0], 0x1C0(r0) + lqv vec31[e0], DMEM_1B0(r0) + lqv vec30[e0], DMEM_1C0(r0) llv vec22[e4], 0x18(t6) bgezal t4, func_1480 addi t4, r0, 0x8080 @@ -1414,37 +1473,37 @@ func_1480: ldv vec14[e8], 0x30(t4) func_14C4: - lsv vec4[e0], 0x0(r0) - lsv vec3[e0], 0x20(r0) - lsv vec21[e0], 0x2(r0) - lsv vec28[e0], 0x22(r0) - lsv vec30[e0], 0x4(r0) + lsv vec4[e0], DMEM_000(r0) + lsv vec3[e0], DMEM_020(r0) + lsv vec21[e0], DMEM_002(r0) + lsv vec28[e0], DMEM_022(r0) + lsv vec30[e0], DMEM_004(r0) vmov vec4[e12],vec4[e8] - lsv vec31[e0], 0x24(r0) + lsv vec31[e0], DMEM_024(r0) vmov vec3[e12],vec3[e8] - lsv vec4[e2], 0x8(r0) + lsv vec4[e2], DMEM_008(r0) vmov vec21[e12],vec21[e8] - lsv vec3[e2], 0x28(r0) + lsv vec3[e2], DMEM_028(r0) vmov vec28[e12],vec28[e8] - lsv vec21[e2], 0xA(r0) + lsv vec21[e2], DMEM_00A(r0) vmov vec30[e12],vec30[e8] - lsv vec28[e2], 0x2A(r0) + lsv vec28[e2], DMEM_02A(r0) vmov vec31[e12],vec31[e8] - lsv vec30[e2], 0xC(r0) + lsv vec30[e2], DMEM_00C(r0) vmov vec4[e13],vec4[e9] - lsv vec31[e2], 0x2C(r0) + lsv vec31[e2], DMEM_02C(r0) vmov vec3[e13],vec3[e9] - lsv vec4[e4], 0x10(r0) + lsv vec4[e4], DMEM_010(r0) vmov vec21[e13],vec21[e9] - lsv vec3[e4], 0x30(r0) + lsv vec3[e4], DMEM_030(r0) vmov vec28[e13],vec28[e9] - lsv vec21[e4], 0x12(r0) + lsv vec21[e4], DMEM_012(r0) vmov vec30[e13],vec30[e9] - lsv vec28[e4], 0x32(r0) + lsv vec28[e4], DMEM_032(r0) vmov vec31[e13],vec31[e9] - lsv vec30[e4], 0x14(r0) + lsv vec30[e4], DMEM_014(r0) vmov vec4[e14],vec4[e10] - lsv vec31[e4], 0x34(r0) + lsv vec31[e4], DMEM_034(r0) vmov vec3[e14],vec3[e10] or t4, r0, r0 vmov vec21[e14],vec21[e10] @@ -1573,7 +1632,7 @@ label_1708: vmulf vec28,vec7,vec20[e4] vmacf vec28,vec6,vec20[e5] vmacf vec28,vec5,vec20[e6] - lqv vec2[e0], 0x1D0(r0) + lqv vec2[e0], DMEM_1D0(r0) vmudh vec22,vec1,vec31[e13] vmacf vec22,vec3,vec21[e4] beqz t4, label_1870