Mesa (master): gallivm: implement indirect addressing over temporaries
Zack Rusin
zack at kemper.freedesktop.org
Thu Apr 22 22:32:14 UTC 2010
Module: Mesa
Branch: master
Commit: 021e0dc78b15fab29e761012860276c2597c8d8f
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=021e0dc78b15fab29e761012860276c2597c8d8f
Author: Zack Rusin <zackr at vmware.com>
Date: Thu Apr 22 18:36:07 2010 -0400
gallivm: implement indirect addressing over temporaries
a bit more involved than indirect addressing over consts, but still
fairly reasonable. we allocate an array instead of individual alloca's,
and we do it only if the shader does indirect addressing.
---
src/gallium/auxiliary/draw/draw_llvm.c | 3 +-
src/gallium/auxiliary/gallivm/lp_bld_tgsi.h | 4 +-
src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c | 105 +++++++++++++++--------
src/gallium/drivers/llvmpipe/lp_state_fs.c | 2 +-
4 files changed, 74 insertions(+), 40 deletions(-)
diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index 9d11031..5e8bef4 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -252,7 +252,8 @@ generate_vs(struct draw_llvm *llvm,
NULL /*pos*/,
inputs,
outputs,
- NULL/*sampler*/);
+ NULL/*sampler*/,
+ &llvm->draw->vs.vertex_shader->info);
}
#if DEBUG_STORE
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
index 63b938b..2eac5da 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h
@@ -39,6 +39,7 @@
struct tgsi_token;
+struct tgsi_shader_info;
struct lp_type;
struct lp_build_context;
struct lp_build_mask_context;
@@ -78,7 +79,8 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
const LLVMValueRef *pos,
const LLVMValueRef (*inputs)[4],
LLVMValueRef (*outputs)[4],
- struct lp_build_sampler_soa *sampler);
+ struct lp_build_sampler_soa *sampler,
+ struct tgsi_shader_info *info);
#endif /* LP_BLD_TGSI_H */
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 72ed8c0..df2e24c 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -46,6 +46,7 @@
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_util.h"
#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_scan.h"
#include "lp_bld_type.h"
#include "lp_bld_const.h"
#include "lp_bld_arit.h"
@@ -127,6 +128,11 @@ struct lp_build_tgsi_soa_context
LLVMValueRef temps[LP_MAX_TEMPS][NUM_CHANNELS];
LLVMValueRef addr[LP_MAX_TEMPS][NUM_CHANNELS];
+ /* we allocate an array of temps if we have indirect
+ * addressing and then the temps above is unused */
+ LLVMValueRef temps_array;
+ boolean has_indirect_addressing;
+
struct lp_build_mask_context *mask;
struct lp_exec_mask exec_mask;
};
@@ -358,6 +364,23 @@ emit_ddy(struct lp_build_tgsi_soa_context *bld,
return lp_build_sub(&bld->base, src_top, src_bottom);
}
+static LLVMValueRef
+get_temp_ptr(struct lp_build_tgsi_soa_context *bld,
+ unsigned index,
+ unsigned swizzle,
+ boolean is_indirect,
+ LLVMValueRef addr)
+{
+ if (!bld->has_indirect_addressing) {
+ return bld->temps[index][swizzle];
+ } else {
+ LLVMValueRef lindex =
+ LLVMConstInt(LLVMInt32Type(), index*4 + swizzle, 0);
+ if (is_indirect)
+ lindex = lp_build_add(&bld->base, lindex, addr);
+ return LLVMBuildGEP(bld->base.builder, bld->temps_array, &lindex, 1, "");
+ }
+}
/**
* Register fetch.
@@ -392,6 +415,7 @@ emit_fetch(
addr = LLVMBuildExtractElement(bld->base.builder,
addr, LLVMConstInt(LLVMInt32Type(), 0, 0),
"");
+ addr = lp_build_mul(&bld->base, addr, LLVMConstInt(LLVMInt32Type(), 4, 0));
}
switch (reg->Register.File) {
@@ -402,7 +426,6 @@ emit_fetch(
if (reg->Register.Indirect) {
/*lp_build_printf(bld->base.builder,
"\taddr = %d\n", addr);*/
- addr = lp_build_mul(&bld->base, addr, LLVMConstInt(LLVMInt32Type(), 4, 0));
index = lp_build_add(&bld->base, index, addr);
}
scalar_ptr = LLVMBuildGEP(bld->base.builder, bld->consts_ptr, &index, 1, "");
@@ -422,11 +445,16 @@ emit_fetch(
assert(res);
break;
- case TGSI_FILE_TEMPORARY:
- res = LLVMBuildLoad(bld->base.builder, bld->temps[reg->Register.Index][swizzle], "");
+ case TGSI_FILE_TEMPORARY: {
+ LLVMValueRef temp_ptr = get_temp_ptr(bld, reg->Register.Index,
+ swizzle,
+ reg->Register.Indirect,
+ addr);
+ res = LLVMBuildLoad(bld->base.builder, temp_ptr, "");
if(!res)
return bld->base.undef;
break;
+ }
default:
assert( 0 );
@@ -504,6 +532,7 @@ emit_store(
LLVMValueRef value)
{
const struct tgsi_full_dst_register *reg = &inst->Dst[index];
+ LLVMValueRef addr;
switch( inst->Instruction.Saturate ) {
case TGSI_SAT_NONE:
@@ -523,16 +552,35 @@ emit_store(
assert(0);
}
+ if (reg->Register.Indirect) {
+ LLVMTypeRef int_vec_type = lp_build_int_vec_type(bld->base.type);
+ unsigned swizzle = tgsi_util_get_src_register_swizzle( ®->Indirect, chan_index );
+ addr = LLVMBuildLoad(bld->base.builder,
+ bld->addr[reg->Indirect.Index][swizzle],
+ "");
+ /* for indexing we want integers */
+ addr = LLVMBuildFPToSI(bld->base.builder, addr,
+ int_vec_type, "");
+ addr = LLVMBuildExtractElement(bld->base.builder,
+ addr, LLVMConstInt(LLVMInt32Type(), 0, 0),
+ "");
+ addr = lp_build_mul(&bld->base, addr, LLVMConstInt(LLVMInt32Type(), 4, 0));
+ }
+
switch( reg->Register.File ) {
case TGSI_FILE_OUTPUT:
lp_exec_mask_store(&bld->exec_mask, value,
bld->outputs[reg->Register.Index][chan_index]);
break;
- case TGSI_FILE_TEMPORARY:
- lp_exec_mask_store(&bld->exec_mask, value,
- bld->temps[reg->Register.Index][chan_index]);
+ case TGSI_FILE_TEMPORARY: {
+ LLVMValueRef temp_ptr = get_temp_ptr(bld, reg->Register.Index,
+ chan_index,
+ reg->Register.Indirect,
+ addr);
+ lp_exec_mask_store(&bld->exec_mask, value, temp_ptr);
break;
+ }
case TGSI_FILE_ADDRESS:
lp_exec_mask_store(&bld->exec_mask, value,
@@ -691,30 +739,6 @@ emit_kilp(struct lp_build_tgsi_soa_context *bld,
lp_build_mask_update(bld->mask, mask);
}
-
-/**
- * Check if inst src/dest regs use indirect addressing into temporary
- * register file.
- */
-static boolean
-indirect_temp_reference(const struct tgsi_full_instruction *inst)
-{
- uint i;
- for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
- const struct tgsi_full_src_register *reg = &inst->Src[i];
- if (reg->Register.File == TGSI_FILE_TEMPORARY &&
- reg->Register.Indirect)
- return TRUE;
- }
- for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
- const struct tgsi_full_dst_register *reg = &inst->Dst[i];
- if (reg->Register.File == TGSI_FILE_TEMPORARY &&
- reg->Register.Indirect)
- return TRUE;
- }
- return FALSE;
-}
-
static int
emit_declaration(
struct lp_build_tgsi_soa_context *bld,
@@ -740,8 +764,16 @@ emit_declaration(
for (idx = first; idx <= last; ++idx) {
switch (decl->Declaration.File) {
case TGSI_FILE_TEMPORARY:
- for (i = 0; i < NUM_CHANNELS; i++)
- bld->temps[idx][i] = lp_build_alloca(&bld->base);
+ if (bld->has_indirect_addressing) {
+ LLVMValueRef val = LLVMConstInt(LLVMInt32Type(),
+ last*4 + 4, 0);
+ bld->temps_array = LLVMBuildArrayAlloca(bld->base.builder,
+ lp_build_vec_type(bld->base.type),
+ val, "");
+ } else {
+ for (i = 0; i < NUM_CHANNELS; i++)
+ bld->temps[idx][i] = lp_build_alloca(&bld->base);
+ }
break;
case TGSI_FILE_OUTPUT:
@@ -787,10 +819,6 @@ emit_instruction(
LLVMValueRef res;
LLVMValueRef dst0[NUM_CHANNELS];
- /* we can't handle indirect addressing into temp register file yet */
- if (indirect_temp_reference(inst))
- return FALSE;
-
/*
* Stores and write masks are handled in a general fashion after the long
* instruction opcode switch statement.
@@ -1742,7 +1770,8 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
const LLVMValueRef *pos,
const LLVMValueRef (*inputs)[NUM_CHANNELS],
LLVMValueRef (*outputs)[NUM_CHANNELS],
- struct lp_build_sampler_soa *sampler)
+ struct lp_build_sampler_soa *sampler,
+ struct tgsi_shader_info *info)
{
struct lp_build_tgsi_soa_context bld;
struct tgsi_parse_context parse;
@@ -1758,6 +1787,8 @@ lp_build_tgsi_soa(LLVMBuilderRef builder,
bld.outputs = outputs;
bld.consts_ptr = consts_ptr;
bld.sampler = sampler;
+ bld.has_indirect_addressing = info->opcode_count[TGSI_OPCODE_ARR] > 0 ||
+ info->opcode_count[TGSI_OPCODE_ARL] > 0;
lp_exec_mask_init(&bld.exec_mask, &bld.base);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 551c375..2c4303a 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -474,7 +474,7 @@ generate_fs(struct llvmpipe_context *lp,
lp_build_tgsi_soa(builder, tokens, type, &mask,
consts_ptr, interp->pos, interp->inputs,
- outputs, sampler);
+ outputs, sampler, &shader->info);
for (attrib = 0; attrib < shader->info.num_outputs; ++attrib) {
for(chan = 0; chan < NUM_CHANNELS; ++chan) {
More information about the mesa-commit
mailing list