From 3f66b72fdb4834c5211305698d22806eac80aa35 Mon Sep 17 00:00:00 2001 From: Ben Skeggs Date: Mon, 5 Jan 2009 13:38:47 +1100 Subject: nv50: ensure we actually get contiguous regs for TEX insn. Still many more horrible things to fix here... --- src/gallium/drivers/nv50/nv50_program.c | 59 +++++++++++++++++++++++---------- 1 file changed, 42 insertions(+), 17 deletions(-) (limited to 'src/gallium/drivers/nv50') diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index d6fbdd1824..566d18cef4 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -179,6 +179,38 @@ free_temp(struct nv50_pc *pc, struct nv50_reg *r) } } +static int +alloc_temp4(struct nv50_pc *pc, struct nv50_reg *dst[4], int idx) +{ + int i; + + if ((idx + 4) >= NV50_SU_MAX_TEMP) + return 1; + + if (pc->r_temp[idx] || pc->r_temp[idx + 1] || + pc->r_temp[idx + 2] || pc->r_temp[idx + 3]) + return alloc_temp4(pc, dst, idx + 1); + + for (i = 0; i < 4; i++) { + dst[i] = CALLOC_STRUCT(nv50_reg); + dst[i]->type = P_TEMP; + dst[i]->index = -1; + dst[i]->hw = idx + i; + pc->r_temp[idx + i] = dst[i]; + } + + return 0; +} + +static void +free_temp4(struct nv50_pc *pc, struct nv50_reg *reg[4]) +{ + int i; + + for (i = 0; i < 4; i++) + free_temp(pc, reg[i]); +} + static struct nv50_reg * temp_temp(struct nv50_pc *pc) { @@ -1156,33 +1188,26 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) break; case TGSI_OPCODE_TEX: { - struct nv50_reg *t0, *t1, *t2, *t3; + struct nv50_reg *t[4]; struct nv50_program_exec *e; - t0 = alloc_temp(pc, NULL); - t0 = alloc_temp(pc, NULL); - t1 = alloc_temp(pc, NULL); - t2 = alloc_temp(pc, NULL); - t3 = alloc_temp(pc, NULL); - emit_mov(pc, t0, src[0][0]); - emit_mov(pc, t1, src[0][1]); + alloc_temp4(pc, t, 0); + emit_mov(pc, t[0], src[0][0]); + emit_mov(pc, t[1], src[0][1]); e = exec(pc); e->inst[0] = 0xf6400000; set_long(pc, e); e->inst[1] |= 0x0000c004; - set_dst(pc, t0, e); + set_dst(pc, t[0], e); emit(pc, e); - if (mask & (1 << 0)) emit_mov(pc, dst[0], t0); - if (mask & (1 << 1)) emit_mov(pc, dst[1], t1); - if (mask & (1 << 2)) emit_mov(pc, dst[2], t2); - if (mask & (1 << 3)) emit_mov(pc, dst[3], t3); + if (mask & (1 << 0)) emit_mov(pc, dst[0], t[0]); + if (mask & (1 << 1)) emit_mov(pc, dst[1], t[1]); + if (mask & (1 << 2)) emit_mov(pc, dst[2], t[2]); + if (mask & (1 << 3)) emit_mov(pc, dst[3], t[3]); - free_temp(pc, t0); - free_temp(pc, t1); - free_temp(pc, t2); - free_temp(pc, t3); + free_temp4(pc, t); } break; case TGSI_OPCODE_XPD: -- cgit v1.2.3