fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-27 01:50:10 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

ac: refactor visit_load_buffer

Browse source 
This is so that we can split different types of loads more easily.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
This commit is contained in:
Rhys Perry 2018-12-05 13:42:47 +00:00
parent ed4020fabe
commit 12dc7cb202
2 changed files with 44 additions and 46 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

8
src/amd/common/ac_llvm_build.c
View file

@ -2701,9 +2701,11 @@ LLVMValueRef ac_trim_vector(struct ac_llvm_context *ctx, LLVMValueRef value,
if (count == num_components)
return value;
LLVMValueRef masks[] = {
ctx->i32_0, ctx->i32_1,
LLVMConstInt(ctx->i32, 2, false), LLVMConstInt(ctx->i32, 3, false)};
LLVMValueRef masks[MAX2(count, 2)];
masks[0] = ctx->i32_0;
masks[1] = ctx->i32_1;
for (unsigned i = 2; i < count; i++)
masks[i] = LLVMConstInt(ctx->i32, i, false);
if (count == 1)
return LLVMBuildExtractElement(ctx->builder, value, masks[0],

82
src/amd/common/ac_nir_to_llvm.c
View file

@ -1621,37 +1621,43 @@ static LLVMValueRef visit_atomic_ssbo(struct ac_nir_context *ctx,
static LLVMValueRef visit_load_buffer(struct ac_nir_context *ctx,
const nir_intrinsic_instr *instr)
{
LLVMValueRef results[2];
int load_bytes;
int elem_size_bytes = instr->dest.ssa.bit_size / 8;
int num_components = instr->num_components;
int num_bytes = num_components * elem_size_bytes;
enum gl_access_qualifier access = nir_intrinsic_access(instr);
LLVMValueRef glc = ctx->ac.i1false;
if (access & (ACCESS_VOLATILE | ACCESS_COHERENT))
glc = ctx->ac.i1true;
for (int i = 0; i < num_bytes; i += load_bytes) {
load_bytes = MIN2(num_bytes - i, 16);
const char *load_name;
LLVMTypeRef data_type;
LLVMValueRef offset = get_src(ctx, instr->src[1]);
LLVMValueRef immoffset = LLVMConstInt(ctx->ac.i32, i, false);
LLVMValueRef rsrc = ctx->abi->load_ssbo(ctx->abi,
get_src(ctx, instr->src[0]), false);
LLVMValueRef vindex = ctx->ac.i32_0;
LLVMValueRef offset = get_src(ctx, instr->src[1]);
LLVMValueRef rsrc = ctx->abi->load_ssbo(ctx->abi,
get_src(ctx, instr->src[0]), false);
LLVMValueRef vindex = ctx->ac.i32_0;
int idx = i ? 1 : 0;
LLVMTypeRef def_type = get_def_type(ctx, &instr->dest.ssa);
LLVMTypeRef def_elem_type = num_components > 1 ? LLVMGetElementType(def_type) : def_type;
LLVMValueRef results[4];
for (int i = 0; i < num_components;) {
int num_elems = num_components - i;
if (num_elems * elem_size_bytes > 16)
num_elems = 16 / elem_size_bytes;
int load_bytes = num_elems * elem_size_bytes;
LLVMValueRef immoffset = LLVMConstInt(ctx->ac.i32, i * elem_size_bytes, false);
LLVMValueRef ret;
if (load_bytes == 2) {
results[idx] = ac_build_tbuffer_load_short(&ctx->ac,
rsrc,
vindex,
offset,
ctx->ac.i32_0,
immoffset,
glc);
ret = ac_build_tbuffer_load_short(&ctx->ac,
rsrc,
vindex,
offset,
ctx->ac.i32_0,
immoffset,
glc);
} else {
const char *load_name;
LLVMTypeRef data_type;
switch (load_bytes) {
case 16:
case 12:
@ -1677,33 +1683,23 @@ static LLVMValueRef visit_load_buffer(struct ac_nir_context *ctx,
glc,
ctx->ac.i1false,
};
results[idx] = ac_build_intrinsic(&ctx->ac, load_name, data_type, params, 5, 0);
unsigned num_elems = ac_get_type_size(data_type) / elem_size_bytes;
LLVMTypeRef resTy = LLVMVectorType(LLVMIntTypeInContext(ctx->ac.context, instr->dest.ssa.bit_size), num_elems);
results[idx] = LLVMBuildBitCast(ctx->ac.builder, results[idx], resTy, "");
}
}
assume(results[0]);
LLVMValueRef ret = results[0];
if (num_bytes > 16 || num_components == 3) {
LLVMValueRef masks[] = {
LLVMConstInt(ctx->ac.i32, 0, false), LLVMConstInt(ctx->ac.i32, 1, false),
LLVMConstInt(ctx->ac.i32, 2, false), LLVMConstInt(ctx->ac.i32, 3, false),
};
if (num_bytes > 16 && num_components == 3) {
/* we end up with a v2i64 and i64 but shuffle fails on that */
results[1] = ac_build_expand(&ctx->ac, results[1], 1, 2);
ret = ac_build_intrinsic(&ctx->ac, load_name, data_type, params, 5, 0);
}
LLVMValueRef swizzle = LLVMConstVector(masks, num_components);
ret = LLVMBuildShuffleVector(ctx->ac.builder, results[0],
results[num_bytes > 16 ? 1 : 0], swizzle, "");
LLVMTypeRef byte_vec = LLVMVectorType(ctx->ac.i8, ac_get_type_size(LLVMTypeOf(ret)));
ret = LLVMBuildBitCast(ctx->ac.builder, ret, byte_vec, "");
ret = ac_trim_vector(&ctx->ac, ret, load_bytes);
LLVMTypeRef ret_type = LLVMVectorType(def_elem_type, num_elems);
ret = LLVMBuildBitCast(ctx->ac.builder, ret, ret_type, "");
for (unsigned j = 0; j < num_elems; j++) {
results[i + j] = LLVMBuildExtractElement(ctx->ac.builder, ret, LLVMConstInt(ctx->ac.i32, j, false), "");
}
i += num_elems;
}
return LLVMBuildBitCast(ctx->ac.builder, ret,
get_def_type(ctx, &instr->dest.ssa), "");
return ac_build_gather_values(&ctx->ac, results, num_components);
}
static LLVMValueRef visit_load_ubo_buffer(struct ac_nir_context *ctx,