From 98c87c382d080ff5a048564e942e649fbaf43879 Mon Sep 17 00:00:00 2001 From: Christoph Bumiller Date: Sun, 12 Sep 2010 14:31:01 +0200 Subject: [PATCH] nv50: handle TGSI EXP and LOG again --- src/gallium/drivers/nv50/nv50_pc_optimize.c | 2 + src/gallium/drivers/nv50/nv50_tgsi_to_nc.c | 48 +++++++++++++++++++-- 2 files changed, 47 insertions(+), 3 deletions(-) diff --git a/src/gallium/drivers/nv50/nv50_pc_optimize.c b/src/gallium/drivers/nv50/nv50_pc_optimize.c index 8653bc6e634..ea1da6268da 100644 --- a/src/gallium/drivers/nv50/nv50_pc_optimize.c +++ b/src/gallium/drivers/nv50/nv50_pc_optimize.c @@ -322,6 +322,8 @@ nv_pass_fold_stores(struct nv_pass *ctx, struct nv_basic_block *b) continue; assert(nvi->def[0] == sti->src[0]->value); + if (nvi->opcode == NV_OP_SELECT) + continue; if (nvi->def[0]->refc > 1) continue; diff --git a/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c b/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c index 978bba4d571..b4f5a884c46 100644 --- a/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c +++ b/src/gallium/drivers/nv50/nv50_tgsi_to_nc.c @@ -214,6 +214,7 @@ static INLINE void bld_warn_uninitialized(struct bld_context *bld, int kind, struct bld_value_stack *stk, struct nv_basic_block *b) { +#ifdef NV50_TGSI2NC_DEBUG long i = (stk - &bld->tvs[0][0]) / 4; long c = (stk - &bld->tvs[0][0]) & 3; @@ -222,6 +223,7 @@ bld_warn_uninitialized(struct bld_context *bld, int kind, debug_printf("WARNING: TEMP[%li].%c %s used uninitialized in BB:%i\n", i, (int)('x' + c), kind ? "may be" : "is", b->id); +#endif } static INLINE struct nv_value * @@ -646,7 +648,10 @@ bld_pow(struct bld_context *bld, struct nv_value *x, struct nv_value *e) static INLINE struct nv_value * bld_load_imm_f32(struct bld_context *bld, float f) { - return bld_insn_1(bld, NV_OP_MOV, bld_imm_f32(bld, f)); + struct nv_value *imm = bld_insn_1(bld, NV_OP_MOV, bld_imm_f32(bld, f)); + + SET_TYPE(imm, NV_TYPE_F32); + return imm; } static INLINE struct nv_value * @@ -944,6 +949,8 @@ emit_store(struct bld_context *bld, const struct tgsi_full_instruction *inst, switch (reg->Register.File) { case TGSI_FILE_OUTPUT: + if (!value->insn && (bld->ti->output_file == NV_FILE_OUT)) + value = bld_insn_1(bld, NV_OP_MOV, value); value = bld_insn_1(bld, NV_OP_MOV, value); value->reg.file = bld->ti->output_file; @@ -956,9 +963,9 @@ emit_store(struct bld_context *bld, const struct tgsi_full_instruction *inst, break; case TGSI_FILE_TEMPORARY: assert(reg->Register.Index < BLD_MAX_TEMPS); - value->reg.file = NV_FILE_GPR; - if (value->insn->bb != bld->pc->current_block) + if (!value->insn || (value->insn->bb != bld->pc->current_block)) value = bld_insn_1(bld, NV_OP_MOV, value); + value->reg.file = NV_FILE_GPR; if (bld->ti->store_to_memory) bld_lmem_store(bld, ptr, reg->Register.Index * 4 + chan, value); @@ -1616,6 +1623,23 @@ bld_instruction(struct bld_context *bld, if (insn->Dst[0].Register.WriteMask & 8) dst0[3] = emit_fetch(bld, insn, 1, 3); break; + case TGSI_OPCODE_EXP: + src0 = emit_fetch(bld, insn, 0, 0); + temp = bld_insn_1(bld, NV_OP_FLOOR, src0); + + if (insn->Dst[0].Register.WriteMask & 2) + dst0[1] = bld_insn_2(bld, NV_OP_SUB, src0, temp); + if (insn->Dst[0].Register.WriteMask & 1) { + temp = bld_insn_1(bld, NV_OP_PREEX2, temp); + dst0[0] = bld_insn_1(bld, NV_OP_EX2, temp); + } + if (insn->Dst[0].Register.WriteMask & 4) { + temp = bld_insn_1(bld, NV_OP_PREEX2, src0); + dst0[2] = bld_insn_1(bld, NV_OP_EX2, temp); + } + if (insn->Dst[0].Register.WriteMask & 8) + dst0[3] = bld_imm_f32(bld, 1.0f); + break; case TGSI_OPCODE_EX2: src0 = emit_fetch(bld, insn, 0, 0); temp = bld_insn_1(bld, NV_OP_PREEX2, src0); @@ -1798,6 +1822,24 @@ bld_instruction(struct bld_context *bld, FOR_EACH_DST0_ENABLED_CHANNEL(c, insn) dst0[c] = temp; break; + case TGSI_OPCODE_LOG: + src0 = emit_fetch(bld, insn, 0, 0); + src0 = bld_insn_1(bld, NV_OP_ABS, src0); + temp = bld_insn_1(bld, NV_OP_LG2, src0); + dst0[2] = temp; + if (insn->Dst[0].Register.WriteMask & 3) { + temp = bld_insn_1(bld, NV_OP_FLOOR, temp); + dst0[0] = temp; + } + if (insn->Dst[0].Register.WriteMask & 2) { + temp = bld_insn_1(bld, NV_OP_PREEX2, temp); + temp = bld_insn_1(bld, NV_OP_EX2, temp); + temp = bld_insn_1(bld, NV_OP_RCP, temp); + dst0[1] = bld_insn_2(bld, NV_OP_MUL, src0, temp); + } + if (insn->Dst[0].Register.WriteMask & 8) + dst0[3] = bld_imm_f32(bld, 1.0f); + break; case TGSI_OPCODE_RCP: case TGSI_OPCODE_LG2: src0 = emit_fetch(bld, insn, 0, 0); -- 2.30.2