[Mesa-dev] [PATCH] gallivm/radeonsi: allow to pass two swizzles into fetches.
Dave Airlie
airlied at gmail.com
Mon Aug 27 21:16:00 UTC 2018
From: Dave Airlie <airlied at redhat.com>
This hijacks the top 16-bits of swizzle, to pass in the swizzle
for the second channel.
This fixes handling .yx swizzles of 64-bit values.
This should fixup radeonsi and llvmpipe.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107524
---
src/gallium/auxiliary/gallivm/lp_bld_tgsi.c | 9 ++
.../auxiliary/gallivm/lp_bld_tgsi_soa.c | 86 ++++++++++++-------
src/gallium/drivers/radeonsi/si_shader.c | 7 +-
.../drivers/radeonsi/si_shader_tgsi_setup.c | 18 ++--
4 files changed, 79 insertions(+), 41 deletions(-)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c
index 64d2cd703be..2c3be8fb127 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.c
@@ -353,6 +353,15 @@ lp_build_emit_fetch_src(
assert(0 && "invalid swizzle in emit_fetch()");
return bld_base->base.undef;
}
+ if (tgsi_type_is_64bit(stype)) {
+ unsigned swizzle2;
+ swizzle2 = tgsi_util_get_full_src_register_swizzle(reg, chan_index + 1);
+ if (swizzle2 > 3) {
+ assert(0 && "invalid swizzle in emit_fetch()");
+ return bld_base->base.undef;
+ }
+ swizzle |= (swizzle2 << 16);
+ }
}
assert(reg->Register.Index <= bld_base->info->file_max[reg->Register.File]);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
index 83d7dbea9a2..79ece639e35 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c
@@ -1190,7 +1190,7 @@ emit_fetch_constant(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld_base->base.gallivm;
@@ -1200,6 +1200,7 @@ emit_fetch_constant(
LLVMValueRef consts_ptr;
LLVMValueRef num_consts;
LLVMValueRef res;
+ unsigned swizzle = swizzle_in & 0xffff;
/* XXX: Handle fetching xyzw components as a vector */
assert(swizzle != ~0u);
@@ -1241,7 +1242,7 @@ emit_fetch_constant(
if (tgsi_type_is_64bit(stype)) {
LLVMValueRef swizzle_vec2;
- swizzle_vec2 = lp_build_const_int_vec(gallivm, uint_bld->type, swizzle + 1);
+ swizzle_vec2 = lp_build_const_int_vec(gallivm, uint_bld->type, swizzle_in >> 16);
index_vec2 = lp_build_shl_imm(uint_bld, indirect_index, 2);
index_vec2 = lp_build_add(uint_bld, index_vec2, swizzle_vec2);
}
@@ -1256,21 +1257,42 @@ emit_fetch_constant(
scalar_ptr = LLVMBuildGEP(builder, consts_ptr,
&index, 1, "");
- if (stype == TGSI_TYPE_DOUBLE) {
- LLVMTypeRef dptr_type = LLVMPointerType(LLVMDoubleTypeInContext(gallivm->context), 0);
- scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, dptr_type, "");
- bld_broad = &bld_base->dbl_bld;
- } else if (stype == TGSI_TYPE_UNSIGNED64) {
- LLVMTypeRef u64ptr_type = LLVMPointerType(LLVMInt64TypeInContext(gallivm->context), 0);
- scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, u64ptr_type, "");
- bld_broad = &bld_base->uint64_bld;
- } else if (stype == TGSI_TYPE_SIGNED64) {
- LLVMTypeRef i64ptr_type = LLVMPointerType(LLVMInt64TypeInContext(gallivm->context), 0);
- scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, i64ptr_type, "");
- bld_broad = &bld_base->int64_bld;
+
+ if (tgsi_type_is_64bit(stype) && ((swizzle_in >> 16) != swizzle + 1)) {
+
+ LLVMValueRef scalar2, scalar2_ptr;
+ LLVMValueRef shuffles[2];
+ index = lp_build_const_int32(gallivm, reg->Register.Index * 4 + (swizzle_in >> 16));
+
+ scalar2_ptr = LLVMBuildGEP(builder, consts_ptr,
+ &index, 1, "");
+
+ scalar = LLVMBuildLoad(builder, scalar_ptr, "");
+ scalar2 = LLVMBuildLoad(builder, scalar2_ptr, "");
+ shuffles[0] = lp_build_const_int32(gallivm, 0);
+ shuffles[1] = lp_build_const_int32(gallivm, 1);
+
+ res = LLVMGetUndef(LLVMVectorType(LLVMFloatTypeInContext(gallivm->context), bld_base->base.type.length * 2));
+ res = LLVMBuildInsertElement(builder, res, scalar, shuffles[0], "");
+ res = LLVMBuildInsertElement(builder, res, scalar2, shuffles[1], "");
+ } else {
+ if (stype == TGSI_TYPE_DOUBLE) {
+ LLVMTypeRef dptr_type = LLVMPointerType(LLVMDoubleTypeInContext(gallivm->context), 0);
+ scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, dptr_type, "");
+ bld_broad = &bld_base->dbl_bld;
+ } else if (stype == TGSI_TYPE_UNSIGNED64) {
+ LLVMTypeRef u64ptr_type = LLVMPointerType(LLVMInt64TypeInContext(gallivm->context), 0);
+ scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, u64ptr_type, "");
+ bld_broad = &bld_base->uint64_bld;
+ } else if (stype == TGSI_TYPE_SIGNED64) {
+ LLVMTypeRef i64ptr_type = LLVMPointerType(LLVMInt64TypeInContext(gallivm->context), 0);
+ scalar_ptr = LLVMBuildBitCast(builder, scalar_ptr, i64ptr_type, "");
+ bld_broad = &bld_base->int64_bld;
+ }
+ scalar = LLVMBuildLoad(builder, scalar_ptr, "");
+ res = lp_build_broadcast_scalar(bld_broad, scalar);
}
- scalar = LLVMBuildLoad(builder, scalar_ptr, "");
- res = lp_build_broadcast_scalar(bld_broad, scalar);
+
}
if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || stype == TGSI_TYPE_DOUBLE || stype == TGSI_TYPE_SIGNED64 || stype == TGSI_TYPE_UNSIGNED64) {
@@ -1319,12 +1341,13 @@ emit_fetch_immediate(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef res = NULL;
+ unsigned swizzle = swizzle_in & 0xffff;
if (bld->use_immediates_array || reg->Register.Indirect) {
LLVMValueRef imms_array;
@@ -1355,7 +1378,7 @@ emit_fetch_immediate(
if (tgsi_type_is_64bit(stype))
index_vec2 = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
- swizzle + 1,
+ swizzle_in >> 16,
FALSE);
/* Gather values from the immediate register array */
res = build_gather(bld_base, imms_array, index_vec, NULL, index_vec2);
@@ -1371,7 +1394,7 @@ emit_fetch_immediate(
LLVMValueRef imms_ptr2;
LLVMValueRef res2;
gep[1] = lp_build_const_int32(gallivm,
- reg->Register.Index * 4 + swizzle + 1);
+ reg->Register.Index * 4 + (swizzle_in >> 16));
imms_ptr2 = LLVMBuildGEP(builder,
bld->imms_array, gep, 2, "");
res2 = LLVMBuildLoad(builder, imms_ptr2, "");
@@ -1382,7 +1405,7 @@ emit_fetch_immediate(
else {
res = bld->immediates[reg->Register.Index][swizzle];
if (tgsi_type_is_64bit(stype))
- res = emit_fetch_64bit(bld_base, stype, res, bld->immediates[reg->Register.Index][swizzle + 1]);
+ res = emit_fetch_64bit(bld_base, stype, res, bld->immediates[reg->Register.Index][swizzle_in >> 16]);
}
if (stype == TGSI_TYPE_SIGNED || stype == TGSI_TYPE_UNSIGNED || tgsi_type_is_64bit(stype)) {
@@ -1397,12 +1420,13 @@ emit_fetch_input(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef res;
+ unsigned swizzle = swizzle_in & 0xffff;
if (reg->Register.Indirect) {
LLVMValueRef indirect_index;
@@ -1423,7 +1447,7 @@ emit_fetch_input(
if (tgsi_type_is_64bit(stype)) {
index_vec2 = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
- swizzle + 1,
+ swizzle_in >> 16,
TRUE);
}
/* cast inputs_array pointer to float* */
@@ -1446,7 +1470,7 @@ emit_fetch_input(
LLVMValueRef res2;
lindex1 = lp_build_const_int32(gallivm,
- reg->Register.Index * 4 + swizzle + 1);
+ reg->Register.Index * 4 + (swizzle_in >> 16));
input_ptr2 = LLVMBuildGEP(builder,
bld->inputs_array, &lindex1, 1, "");
res2 = LLVMBuildLoad(builder, input_ptr2, "");
@@ -1456,7 +1480,7 @@ emit_fetch_input(
else {
res = bld->inputs[reg->Register.Index][swizzle];
if (tgsi_type_is_64bit(stype))
- res = emit_fetch_64bit(bld_base, stype, res, bld->inputs[reg->Register.Index][swizzle + 1]);
+ res = emit_fetch_64bit(bld_base, stype, res, bld->inputs[reg->Register.Index][swizzle_in >> 16]);
}
}
@@ -1476,7 +1500,7 @@ emit_fetch_gs_input(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
@@ -1484,6 +1508,7 @@ emit_fetch_gs_input(
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef attrib_index = NULL;
LLVMValueRef vertex_index = NULL;
+ unsigned swizzle = swizzle_in & 0xffff;
LLVMValueRef swizzle_index = lp_build_const_int32(gallivm, swizzle);
LLVMValueRef res;
@@ -1525,7 +1550,7 @@ emit_fetch_gs_input(
assert(res);
if (tgsi_type_is_64bit(stype)) {
- LLVMValueRef swizzle_index = lp_build_const_int32(gallivm, swizzle + 1);
+ LLVMValueRef swizzle_index = lp_build_const_int32(gallivm, swizzle_in >> 16);
LLVMValueRef res2;
res2 = bld->gs_iface->fetch_input(bld->gs_iface, bld_base,
reg->Dimension.Indirect,
@@ -1549,12 +1574,13 @@ emit_fetch_temporary(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
LLVMBuilderRef builder = gallivm->builder;
LLVMValueRef res;
+ unsigned swizzle = swizzle_in & 0xffff;
if (reg->Register.Indirect) {
LLVMValueRef indirect_index;
@@ -1574,7 +1600,7 @@ emit_fetch_temporary(
if (tgsi_type_is_64bit(stype)) {
index_vec2 = get_soa_array_offsets(&bld_base->uint_bld,
indirect_index,
- swizzle + 1,
+ swizzle_in >> 16,
TRUE);
}
@@ -1593,7 +1619,7 @@ emit_fetch_temporary(
if (tgsi_type_is_64bit(stype)) {
LLVMValueRef temp_ptr2, res2;
- temp_ptr2 = lp_get_temp_ptr_soa(bld, reg->Register.Index, swizzle + 1);
+ temp_ptr2 = lp_get_temp_ptr_soa(bld, reg->Register.Index, swizzle_in >> 16);
res2 = LLVMBuildLoad(builder, temp_ptr2, "");
res = emit_fetch_64bit(bld_base, stype, res, res2);
}
@@ -1616,7 +1642,7 @@ emit_fetch_system_value(
struct lp_build_tgsi_context * bld_base,
const struct tgsi_full_src_register * reg,
enum tgsi_opcode_type stype,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct lp_build_tgsi_soa_context * bld = lp_soa_context(bld_base);
struct gallivm_state *gallivm = bld->bld_base.base.gallivm;
diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index cfd99b61601..c7530c5ffbd 100644
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -2393,16 +2393,17 @@ static LLVMValueRef fetch_constant(
struct lp_build_tgsi_context *bld_base,
const struct tgsi_full_src_register *reg,
enum tgsi_opcode_type type,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
struct si_shader_selector *sel = ctx->shader->selector;
const struct tgsi_ind_register *ireg = ®->Indirect;
unsigned buf, idx;
+ unsigned swizzle = swizzle_in & 0xffff;
LLVMValueRef addr, bufp;
- if (swizzle == LP_CHAN_ALL) {
+ if (swizzle_in == LP_CHAN_ALL) {
unsigned chan;
LLVMValueRef values[4];
for (chan = 0; chan < TGSI_NUM_CHANNELS; ++chan)
@@ -2416,7 +2417,7 @@ static LLVMValueRef fetch_constant(
LLVMValueRef lo, hi;
lo = fetch_constant(bld_base, reg, TGSI_TYPE_UNSIGNED, swizzle);
- hi = fetch_constant(bld_base, reg, TGSI_TYPE_UNSIGNED, swizzle + 1);
+ hi = fetch_constant(bld_base, reg, TGSI_TYPE_UNSIGNED, (swizzle_in >> 16));
return si_llvm_emit_fetch_64bit(bld_base, tgsi2llvmtype(bld_base, type),
lo, hi);
}
diff --git a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
index 20164939cb7..d48eda1b100 100644
--- a/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
+++ b/src/gallium/drivers/radeonsi/si_shader_tgsi_setup.c
@@ -445,13 +445,14 @@ get_output_ptr(struct lp_build_tgsi_context *bld_base, unsigned index,
LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
const struct tgsi_full_src_register *reg,
enum tgsi_opcode_type type,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
LLVMBuilderRef builder = ctx->ac.builder;
LLVMValueRef result = NULL, ptr, ptr2;
+ unsigned swizzle = swizzle_in & 0xffff;
- if (swizzle == ~0) {
+ if (swizzle_in == ~0) {
LLVMValueRef values[TGSI_NUM_CHANNELS];
unsigned chan;
for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
@@ -476,7 +477,7 @@ LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
ctx->imms[reg->Register.Index * TGSI_NUM_CHANNELS + swizzle],
ctx->i32_0);
result = LLVMConstInsertElement(result,
- ctx->imms[reg->Register.Index * TGSI_NUM_CHANNELS + swizzle + 1],
+ ctx->imms[reg->Register.Index * TGSI_NUM_CHANNELS + (swizzle_in >> 16)],
ctx->i32_1);
return LLVMConstBitCast(result, ctype);
} else {
@@ -503,7 +504,7 @@ LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
if (tgsi_type_is_64bit(type)) {
ptr = result;
- ptr2 = input[swizzle + 1];
+ ptr2 = input[swizzle_in >> 16];
return si_llvm_emit_fetch_64bit(bld_base, tgsi2llvmtype(bld_base, type),
ptr, ptr2);
}
@@ -515,7 +516,7 @@ LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
return LLVMGetUndef(tgsi2llvmtype(bld_base, type));
ptr = ctx->temps[reg->Register.Index * TGSI_NUM_CHANNELS + swizzle];
if (tgsi_type_is_64bit(type)) {
- ptr2 = ctx->temps[reg->Register.Index * TGSI_NUM_CHANNELS + swizzle + 1];
+ ptr2 = ctx->temps[reg->Register.Index * TGSI_NUM_CHANNELS + (swizzle_in >> 16)];
return si_llvm_emit_fetch_64bit(bld_base, tgsi2llvmtype(bld_base, type),
LLVMBuildLoad(builder, ptr, ""),
LLVMBuildLoad(builder, ptr2, ""));
@@ -526,7 +527,7 @@ LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
case TGSI_FILE_OUTPUT:
ptr = get_output_ptr(bld_base, reg->Register.Index, swizzle);
if (tgsi_type_is_64bit(type)) {
- ptr2 = get_output_ptr(bld_base, reg->Register.Index, swizzle + 1);
+ ptr2 = get_output_ptr(bld_base, reg->Register.Index, (swizzle_in >> 16));
return si_llvm_emit_fetch_64bit(bld_base, tgsi2llvmtype(bld_base, type),
LLVMBuildLoad(builder, ptr, ""),
LLVMBuildLoad(builder, ptr2, ""));
@@ -544,11 +545,12 @@ LLVMValueRef si_llvm_emit_fetch(struct lp_build_tgsi_context *bld_base,
static LLVMValueRef fetch_system_value(struct lp_build_tgsi_context *bld_base,
const struct tgsi_full_src_register *reg,
enum tgsi_opcode_type type,
- unsigned swizzle)
+ unsigned swizzle_in)
{
struct si_shader_context *ctx = si_shader_context(bld_base);
LLVMBuilderRef builder = ctx->ac.builder;
LLVMValueRef cval = ctx->system_values[reg->Register.Index];
+ unsigned swizzle = swizzle_in & 0xffff;
if (tgsi_type_is_64bit(type)) {
LLVMValueRef lo, hi;
@@ -558,7 +560,7 @@ static LLVMValueRef fetch_system_value(struct lp_build_tgsi_context *bld_base,
lo = LLVMBuildExtractElement(
builder, cval, LLVMConstInt(ctx->i32, swizzle, 0), "");
hi = LLVMBuildExtractElement(
- builder, cval, LLVMConstInt(ctx->i32, swizzle + 1, 0), "");
+ builder, cval, LLVMConstInt(ctx->i32, (swizzle_in >> 16), 0), "");
return si_llvm_emit_fetch_64bit(bld_base, tgsi2llvmtype(bld_base, type),
lo, hi);
--
2.17.1
More information about the mesa-dev
mailing list