From: Marek Olšák <marek.ol...@amd.com>
---
src/gallium/drivers/radeonsi/si_shader.c | 82 ++++++++++++++------------------
1 file changed, 35 insertions(+), 47 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_shader.c
b/src/gallium/drivers/radeonsi/si_shader.c
index 7d92425..67ab16b 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -343,56 +343,60 @@ get_tcs_out_current_patch_data_offset(struct
si_shader_context *ctx)
get_tcs_out_patch0_patch_data_offset(ctx);
LLVMValueRef patch_stride = get_tcs_out_patch_stride(ctx);
LLVMValueRef rel_patch_id = get_rel_patch_id(ctx);
return LLVMBuildAdd(gallivm->builder, patch0_patch_data_offset,
LLVMBuildMul(gallivm->builder, patch_stride,
rel_patch_id, ""),
"");
}
+static LLVMValueRef build_gep0(struct si_shader_context *ctx,
+ LLVMValueRef base_ptr, LLVMValueRef index)
+{
+ LLVMValueRef indices[2] = {
+ LLVMConstInt(ctx->i32, 0, 0),
+ index,
+ };
+ return LLVMBuildGEP(ctx->radeon_bld.gallivm.builder, base_ptr,
+ indices, 2, "");
+}
+
static void build_indexed_store(struct si_shader_context *ctx,
LLVMValueRef base_ptr, LLVMValueRef index,
LLVMValueRef value)
{
struct lp_build_tgsi_context *bld_base = &ctx->radeon_bld.soa.bld_base;
struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMValueRef indices[2], pointer;
-
- indices[0] = bld_base->uint_bld.zero;
- indices[1] = index;
- pointer = LLVMBuildGEP(gallivm->builder, base_ptr, indices, 2, "");
- LLVMBuildStore(gallivm->builder, value, pointer);
+ LLVMBuildStore(gallivm->builder, value,
+ build_gep0(ctx, base_ptr, index));
}
/**
* Build an LLVM bytecode indexed load using LLVMBuildGEP + LLVMBuildLoad.
* It's equivalent to doing a load from &base_ptr[index].
*
* \param base_ptr Where the array starts.
* \param index The element index into the array.
* \param uniform Whether the base_ptr and index can be assumed to be
* dynamically uniform
*/
static LLVMValueRef build_indexed_load(struct si_shader_context *ctx,
LLVMValueRef base_ptr, LLVMValueRef
index,
bool uniform)
{
struct lp_build_tgsi_context *bld_base = &ctx->radeon_bld.soa.bld_base;
struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMValueRef indices[2], pointer;
-
- indices[0] = bld_base->uint_bld.zero;
- indices[1] = index;
+ LLVMValueRef pointer;
- pointer = LLVMBuildGEP(gallivm->builder, base_ptr, indices, 2, "");
+ pointer = build_gep0(ctx, base_ptr, index);
if (uniform)
LLVMSetMetadata(pointer, ctx->uniform_md_kind, ctx->empty_md);
return LLVMBuildLoad(gallivm->builder, pointer, "");
}
/**
* Do a load from &base_ptr[index], but also add a flag that it's loading
* a constant from a dynamically uniform index.
*/
static LLVMValueRef build_indexed_load_const(
@@ -4996,54 +5000,47 @@ static void si_llvm_emit_txqs(
static void si_llvm_emit_ddxy(
const struct lp_build_tgsi_action *action,
struct lp_build_tgsi_context *bld_base,
struct lp_build_emit_data *emit_data)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
const struct tgsi_full_instruction *inst = emit_data->inst;
unsigned opcode = inst->Instruction.Opcode;
- LLVMValueRef indices[2];
- LLVMValueRef store_ptr, load_ptr0, load_ptr1;
+ LLVMValueRef store_ptr, load_ptr0, load_ptr1, thread_id;
LLVMValueRef tl, trbl, result[4];
LLVMValueRef tl_tid, trbl_tid;
unsigned swizzle[4];
unsigned c;
int idx;
unsigned mask;
- indices[0] = bld_base->uint_bld.zero;
- indices[1] = get_thread_id(ctx);
- store_ptr = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ thread_id = get_thread_id(ctx);;
+ store_ptr = build_gep0(ctx, ctx->lds, thread_id);
if (opcode == TGSI_OPCODE_DDX_FINE)
mask = TID_MASK_LEFT;
else if (opcode == TGSI_OPCODE_DDY_FINE)
mask = TID_MASK_TOP;
else
mask = TID_MASK_TOP_LEFT;
- tl_tid = LLVMBuildAnd(gallivm->builder, indices[1],
+ tl_tid = LLVMBuildAnd(gallivm->builder, thread_id,
lp_build_const_int32(gallivm, mask), "");
- indices[1] = tl_tid;
- load_ptr0 = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ load_ptr0 = build_gep0(ctx, ctx->lds, tl_tid);
/* for DDX we want to next X pixel, DDY next Y pixel. */
idx = (opcode == TGSI_OPCODE_DDX || opcode == TGSI_OPCODE_DDX_FINE) ? 1
: 2;
- trbl_tid = LLVMBuildAdd(gallivm->builder, indices[1],
+ trbl_tid = LLVMBuildAdd(gallivm->builder, tl_tid,
lp_build_const_int32(gallivm, idx), "");
- indices[1] = trbl_tid;
- load_ptr1 = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ load_ptr1 = build_gep0(ctx, ctx->lds, trbl_tid);
for (c = 0; c < 4; ++c) {
unsigned i;
LLVMValueRef val;
LLVMValueRef args[2];
swizzle[c] =
tgsi_util_get_full_src_register_swizzle(&inst->Src[0], c);
for (i = 0; i < c; ++i) {
if (swizzle[i] == swizzle[c]) {
result[c] = result[i];
@@ -5088,53 +5085,44 @@ static void si_llvm_emit_ddxy(
* this takes an I,J coordinate pair,
* and works out the X and Y derivatives.
* it returns DDX(I), DDX(J), DDY(I), DDY(J).
*/
static LLVMValueRef si_llvm_emit_ddxy_interp(
struct lp_build_tgsi_context *bld_base,
LLVMValueRef interp_ij)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
- LLVMValueRef indices[2];
LLVMValueRef store_ptr, load_ptr_x, load_ptr_y, load_ptr_ddx,
load_ptr_ddy, temp, temp2;
- LLVMValueRef tl, tr, bl, result[4];
+ LLVMValueRef tl, tr, bl, result[4], thread_id;
unsigned c;
- indices[0] = bld_base->uint_bld.zero;
- indices[1] = get_thread_id(ctx);
- store_ptr = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ thread_id = get_thread_id(ctx);
+ store_ptr = build_gep0(ctx, ctx->lds, thread_id);
- temp = LLVMBuildAnd(gallivm->builder, indices[1],
+ temp = LLVMBuildAnd(gallivm->builder, thread_id,
lp_build_const_int32(gallivm, TID_MASK_LEFT), "");
- temp2 = LLVMBuildAnd(gallivm->builder, indices[1],
+ temp2 = LLVMBuildAnd(gallivm->builder, thread_id,
lp_build_const_int32(gallivm, TID_MASK_TOP), "");
- indices[1] = temp;
- load_ptr_x = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ load_ptr_x = build_gep0(ctx, ctx->lds, temp);
- indices[1] = temp2;
- load_ptr_y = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ load_ptr_y = build_gep0(ctx, ctx->lds, temp2);
- indices[1] = LLVMBuildAdd(gallivm->builder, temp,
- lp_build_const_int32(gallivm, 1), "");
- load_ptr_ddx = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ load_ptr_ddx = build_gep0(ctx, ctx->lds,
+ LLVMBuildAdd(gallivm->builder, temp,
+ lp_build_const_int32(gallivm,
1), ""));
- indices[1] = LLVMBuildAdd(gallivm->builder, temp2,
- lp_build_const_int32(gallivm, 2), "");
- load_ptr_ddy = LLVMBuildGEP(gallivm->builder, ctx->lds,
- indices, 2, "");
+ load_ptr_ddy = build_gep0(ctx, ctx->lds,
+ LLVMBuildAdd(gallivm->builder, temp2,
+ lp_build_const_int32(gallivm,
2), ""));
for (c = 0; c < 2; ++c) {
LLVMValueRef store_val;
LLVMValueRef c_ll = lp_build_const_int32(gallivm, c);
store_val = LLVMBuildExtractElement(gallivm->builder,
interp_ij, c_ll, "");
LLVMBuildStore(gallivm->builder,
store_val,
store_ptr);
--
2.7.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
