Mesa (master): gallivm: fix cube map array (and cube map shadow with bias) handling
Roland Scheidegger
sroland at kemper.freedesktop.org
Tue Aug 5 02:13:53 UTC 2014
Module: Mesa
Branch: master
Commit: c3c33756ff828ba7ae065c224943aeb48b8ed2ba
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=c3c33756ff828ba7ae065c224943aeb48b8ed2ba
Author: Roland Scheidegger <sroland at vmware.com>
Date: Sat Aug 2 01:49:39 2014 +0200
gallivm: fix cube map array (and cube map shadow with bias) handling
In particular need to handle TEX2/TXB2/TXL2 opcodes.
cube map shadow with bias already used TXB2 which didn't work before
at all, despite that there's by default no piglit change (but using
no_quad_lod and no_rho_opt indeed passes some more tex-miplevel-selection
tests).
The actual sampling code still won't handle cube map arrays.
Reviewed-by: Brian Paul <brianp at vmware.com>
---
src/gallium/auxiliary/gallivm/lp_bld_tgsi_info.c | 20 ++++
src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c | 134 +++++++++++++++++-----
2 files changed, 125 insertions(+), 29 deletions(-)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_info.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_info.c
index 48d1c52..3bbf260 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_info.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_info.c
@@ -137,10 +137,21 @@ analyse_tex(struct analysis_context *ctx,
case TGSI_TEXTURE_SHADOWCUBE:
readmask = TGSI_WRITEMASK_XYZW;
break;
+ case TGSI_TEXTURE_CUBE_ARRAY:
+ readmask = TGSI_WRITEMASK_XYZW;
+ break;
+ case TGSI_TEXTURE_SHADOWCUBE_ARRAY:
+ readmask = TGSI_WRITEMASK_XYZW;
+ break;
default:
assert(0);
return;
}
+ /* XXX
+ * For cube map arrays, this will not analyze lod or shadow argument.
+ * For shadow cube, this will not analyze lod bias argument.
+ * "Indirect" really has no meaning for such textures anyway though.
+ */
if (modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV) {
/* We don't track explicit derivatives, although we could */
@@ -296,6 +307,15 @@ analyse_instruction(struct analysis_context *ctx,
case TGSI_OPCODE_TXP:
analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_PROJECTED);
break;
+ case TGSI_OPCODE_TEX2:
+ analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_NONE);
+ break;
+ case TGSI_OPCODE_TXB2:
+ analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_LOD_BIAS);
+ break;
+ case TGSI_OPCODE_TXL2:
+ analyse_tex(ctx, inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD);
+ break;
case TGSI_OPCODE_SAMPLE:
analyse_sample(ctx, inst, LP_BLD_TEX_MODIFIER_NONE, FALSE);
break;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 3d7df3e..02a4a52 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -1938,9 +1938,10 @@ static void
emit_tex( struct lp_build_tgsi_soa_context *bld,
const struct tgsi_full_instruction *inst,
enum lp_build_tex_modifier modifier,
- LLVMValueRef *texel)
+ LLVMValueRef *texel,
+ unsigned sampler_reg)
{
- unsigned unit;
+ unsigned unit = inst->Src[sampler_reg].Register.Index;
LLVMValueRef lod_bias, explicit_lod;
LLVMValueRef oow = NULL;
LLVMValueRef coords[5];
@@ -2010,7 +2011,16 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
num_derivs = 3;
break;
case TGSI_TEXTURE_CUBE_ARRAY:
+ num_offsets = 2;
+ num_derivs = 3;
+ layer_coord = 3;
+ break;
case TGSI_TEXTURE_SHADOWCUBE_ARRAY:
+ num_offsets = 2;
+ num_derivs = 3;
+ layer_coord = 3;
+ shadow_coord = 4; /* shadow coord special different reg */
+ break;
case TGSI_TEXTURE_2D_MSAA:
case TGSI_TEXTURE_2D_ARRAY_MSAA:
default:
@@ -2021,7 +2031,15 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
/* Note lod and especially projected are illegal in a LOT of cases */
if (modifier == LP_BLD_TEX_MODIFIER_LOD_BIAS ||
modifier == LP_BLD_TEX_MODIFIER_EXPLICIT_LOD) {
- LLVMValueRef lod = lp_build_emit_fetch(&bld->bld_base, inst, 0, 3);
+ LLVMValueRef lod;
+ if (inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE ||
+ inst->Texture.Texture == TGSI_TEXTURE_CUBE_ARRAY) {
+ /* note that shadow cube array with bias/explicit lod does not exist */
+ lod = lp_build_emit_fetch(&bld->bld_base, inst, 1, 0);
+ }
+ else {
+ lod = lp_build_emit_fetch(&bld->bld_base, inst, 0, 3);
+ }
if (modifier == LP_BLD_TEX_MODIFIER_LOD_BIAS) {
lod_bias = lod;
explicit_lod = NULL;
@@ -2053,13 +2071,23 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
/* Layer coord always goes into 3rd slot, except for cube map arrays */
if (layer_coord) {
- coords[2] = lp_build_emit_fetch(&bld->bld_base, inst, 0, layer_coord);
+ if (layer_coord == 3) {
+ coords[3] = lp_build_emit_fetch(&bld->bld_base, inst, 0, layer_coord);
+ }
+ else {
+ coords[2] = lp_build_emit_fetch(&bld->bld_base, inst, 0, layer_coord);
+ }
if (modifier == LP_BLD_TEX_MODIFIER_PROJECTED)
coords[2] = lp_build_mul(&bld->bld_base.base, coords[2], oow);
}
/* Shadow coord occupies always 5th slot. */
if (shadow_coord) {
- coords[4] = lp_build_emit_fetch(&bld->bld_base, inst, 0, shadow_coord);
+ if (shadow_coord == 4) {
+ coords[4] = lp_build_emit_fetch(&bld->bld_base, inst, 1, 0);
+ }
+ else {
+ coords[4] = lp_build_emit_fetch(&bld->bld_base, inst, 0, shadow_coord);
+ }
if (modifier == LP_BLD_TEX_MODIFIER_PROJECTED)
coords[4] = lp_build_mul(&bld->bld_base.base, coords[4], oow);
}
@@ -2071,7 +2099,6 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
derivs.ddy[dim] = lp_build_emit_fetch(&bld->bld_base, inst, 2, dim);
}
deriv_ptr = &derivs;
- unit = inst->Src[3].Register.Index;
/*
* could also check all src regs if constant but I doubt such
* cases exist in practice.
@@ -2087,8 +2114,6 @@ emit_tex( struct lp_build_tgsi_soa_context *bld,
else {
lod_property = LP_SAMPLER_LOD_PER_ELEMENT;
}
- } else {
- unit = inst->Src[1].Register.Index;
}
/* some advanced gather instructions (txgo) would require 4 offsets */
@@ -2453,7 +2478,7 @@ emit_size_query( struct lp_build_tgsi_soa_context *bld,
static boolean
near_end_of_shader(struct lp_build_tgsi_soa_context *bld,
- int pc)
+ int pc)
{
int i;
@@ -2461,27 +2486,38 @@ near_end_of_shader(struct lp_build_tgsi_soa_context *bld,
unsigned opcode;
if (pc + i >= bld->bld_base.info->num_instructions)
- return TRUE;
+ return TRUE;
opcode = bld->bld_base.instructions[pc + i].Instruction.Opcode;
if (opcode == TGSI_OPCODE_END)
- return TRUE;
+ return TRUE;
if (opcode == TGSI_OPCODE_TEX ||
- opcode == TGSI_OPCODE_TXP ||
- opcode == TGSI_OPCODE_TXD ||
- opcode == TGSI_OPCODE_TXB ||
- opcode == TGSI_OPCODE_TXL ||
- opcode == TGSI_OPCODE_TXF ||
- opcode == TGSI_OPCODE_TXQ ||
- opcode == TGSI_OPCODE_CAL ||
- opcode == TGSI_OPCODE_CALLNZ ||
- opcode == TGSI_OPCODE_IF ||
- opcode == TGSI_OPCODE_UIF ||
- opcode == TGSI_OPCODE_BGNLOOP ||
- opcode == TGSI_OPCODE_SWITCH)
- return FALSE;
+ opcode == TGSI_OPCODE_TXP ||
+ opcode == TGSI_OPCODE_TXD ||
+ opcode == TGSI_OPCODE_TXB ||
+ opcode == TGSI_OPCODE_TXL ||
+ opcode == TGSI_OPCODE_TXF ||
+ opcode == TGSI_OPCODE_TXQ ||
+ opcode == TGSI_OPCODE_TEX2 ||
+ opcode == TGSI_OPCODE_TXB2 ||
+ opcode == TGSI_OPCODE_TXL2 ||
+ opcode == TGSI_OPCODE_SAMPLE ||
+ opcode == TGSI_OPCODE_SAMPLE_B ||
+ opcode == TGSI_OPCODE_SAMPLE_C ||
+ opcode == TGSI_OPCODE_SAMPLE_C_LZ ||
+ opcode == TGSI_OPCODE_SAMPLE_D ||
+ opcode == TGSI_OPCODE_SAMPLE_I ||
+ opcode == TGSI_OPCODE_SAMPLE_L ||
+ opcode == TGSI_OPCODE_SVIEWINFO ||
+ opcode == TGSI_OPCODE_CAL ||
+ opcode == TGSI_OPCODE_CALLNZ ||
+ opcode == TGSI_OPCODE_IF ||
+ opcode == TGSI_OPCODE_UIF ||
+ opcode == TGSI_OPCODE_BGNLOOP ||
+ opcode == TGSI_OPCODE_SWITCH)
+ return FALSE;
}
return TRUE;
@@ -2888,7 +2924,20 @@ tex_emit(
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
- emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE, emit_data->output);
+ emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
+ emit_data->output, 1);
+}
+
+static void
+tex2_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+ emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_NONE,
+ emit_data->output, 2);
}
static void
@@ -2900,7 +2949,19 @@ txb_emit(
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_LOD_BIAS,
- emit_data->output);
+ emit_data->output, 1);
+}
+
+static void
+txb2_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+ emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_LOD_BIAS,
+ emit_data->output, 2);
}
static void
@@ -2912,7 +2973,7 @@ txd_emit(
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_EXPLICIT_DERIV,
- emit_data->output);
+ emit_data->output, 3);
}
static void
@@ -2924,7 +2985,19 @@ txl_emit(
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD,
- emit_data->output);
+ emit_data->output, 1);
+}
+
+static void
+txl2_emit(
+ const struct lp_build_tgsi_action * action,
+ struct lp_build_tgsi_context * bld_base,
+ struct lp_build_emit_data * emit_data)
+{
+ struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
+
+ emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_EXPLICIT_LOD,
+ emit_data->output, 2);
}
static void
@@ -2936,7 +3009,7 @@ txp_emit(
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
emit_tex(bld, emit_data->inst, LP_BLD_TEX_MODIFIER_PROJECTED,
- emit_data->output);
+ emit_data->output, 1);
}
static void
@@ -3745,6 +3818,9 @@ lp_build_tgsi_soa(struct gallivm_state *gallivm,
bld.bld_base.op_actions[TGSI_OPCODE_TXP].emit = txp_emit;
bld.bld_base.op_actions[TGSI_OPCODE_TXQ].emit = txq_emit;
bld.bld_base.op_actions[TGSI_OPCODE_TXF].emit = txf_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_TEX2].emit = tex2_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_TXB2].emit = txb2_emit;
+ bld.bld_base.op_actions[TGSI_OPCODE_TXL2].emit = txl2_emit;
/* DX10 sampling ops */
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE].emit = sample_emit;
bld.bld_base.op_actions[TGSI_OPCODE_SAMPLE_B].emit = sample_b_emit;
More information about the mesa-commit
mailing list