amd/common: Move ac_export_mrt_z to ac_llvm_build.

The aim of this commit is to keep ac_shader_util LLVM-free,
since we would like to use it in ACO later.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This commit is contained in:
Timur Kristóf
2019-09-25 14:05:19 +02:00
committed by Connor Abbott
parent 06ea3325c3
commit d8b46f8964
4 changed files with 76 additions and 75 deletions

View File

@@ -34,6 +34,7 @@
#include <stdio.h> #include <stdio.h>
#include "ac_llvm_util.h" #include "ac_llvm_util.h"
#include "ac_shader_util.h"
#include "ac_exp_param.h" #include "ac_exp_param.h"
#include "util/bitscan.h" #include "util/bitscan.h"
#include "util/macros.h" #include "util/macros.h"
@@ -4432,3 +4433,73 @@ LLVMValueRef ac_build_call(struct ac_llvm_context *ctx, LLVMValueRef func,
LLVMSetInstructionCallConv(ret, LLVMGetFunctionCallConv(func)); LLVMSetInstructionCallConv(ret, LLVMGetFunctionCallConv(func));
return ret; return ret;
} }
void
ac_export_mrt_z(struct ac_llvm_context *ctx, LLVMValueRef depth,
LLVMValueRef stencil, LLVMValueRef samplemask,
struct ac_export_args *args)
{
unsigned mask = 0;
unsigned format = ac_get_spi_shader_z_format(depth != NULL,
stencil != NULL,
samplemask != NULL);
assert(depth || stencil || samplemask);
memset(args, 0, sizeof(*args));
args->valid_mask = 1; /* whether the EXEC mask is valid */
args->done = 1; /* DONE bit */
/* Specify the target we are exporting */
args->target = V_008DFC_SQ_EXP_MRTZ;
args->compr = 0; /* COMP flag */
args->out[0] = LLVMGetUndef(ctx->f32); /* R, depth */
args->out[1] = LLVMGetUndef(ctx->f32); /* G, stencil test val[0:7], stencil op val[8:15] */
args->out[2] = LLVMGetUndef(ctx->f32); /* B, sample mask */
args->out[3] = LLVMGetUndef(ctx->f32); /* A, alpha to mask */
if (format == V_028710_SPI_SHADER_UINT16_ABGR) {
assert(!depth);
args->compr = 1; /* COMPR flag */
if (stencil) {
/* Stencil should be in X[23:16]. */
stencil = ac_to_integer(ctx, stencil);
stencil = LLVMBuildShl(ctx->builder, stencil,
LLVMConstInt(ctx->i32, 16, 0), "");
args->out[0] = ac_to_float(ctx, stencil);
mask |= 0x3;
}
if (samplemask) {
/* SampleMask should be in Y[15:0]. */
args->out[1] = samplemask;
mask |= 0xc;
}
} else {
if (depth) {
args->out[0] = depth;
mask |= 0x1;
}
if (stencil) {
args->out[1] = stencil;
mask |= 0x2;
}
if (samplemask) {
args->out[2] = samplemask;
mask |= 0x4;
}
}
/* GFX6 (except OLAND and HAINAN) has a bug that it only looks
* at the X writemask component. */
if (ctx->chip_class == GFX6 &&
ctx->family != CHIP_OLAND &&
ctx->family != CHIP_HAINAN)
mask |= 0x1;
/* Specify which components to enable */
args->enabled_channels = mask;
}

View File

@@ -740,6 +740,11 @@ LLVMValueRef ac_build_atomic_cmp_xchg(struct ac_llvm_context *ctx, LLVMValueRef
LLVMValueRef cmp, LLVMValueRef val, LLVMValueRef cmp, LLVMValueRef val,
const char *sync_scope); const char *sync_scope);
void
ac_export_mrt_z(struct ac_llvm_context *ctx, LLVMValueRef depth,
LLVMValueRef stencil, LLVMValueRef samplemask,
struct ac_export_args *args);
#ifdef __cplusplus #ifdef __cplusplus
} }
#endif #endif

View File

@@ -25,7 +25,6 @@
#include <stdlib.h> #include <stdlib.h>
#include <string.h> #include <string.h>
#include "ac_nir_to_llvm.h"
#include "ac_shader_util.h" #include "ac_shader_util.h"
#include "sid.h" #include "sid.h"
@@ -109,71 +108,3 @@ ac_vgt_gs_mode(unsigned gs_max_vert_out, enum chip_class chip_class)
S_028A40_ONCHIP(chip_class >= GFX9 ? 1 : 0); S_028A40_ONCHIP(chip_class >= GFX9 ? 1 : 0);
} }
void
ac_export_mrt_z(struct ac_llvm_context *ctx, LLVMValueRef depth,
LLVMValueRef stencil, LLVMValueRef samplemask,
struct ac_export_args *args)
{
unsigned mask = 0;
unsigned format = ac_get_spi_shader_z_format(depth != NULL,
stencil != NULL,
samplemask != NULL);
assert(depth || stencil || samplemask);
memset(args, 0, sizeof(*args));
args->valid_mask = 1; /* whether the EXEC mask is valid */
args->done = 1; /* DONE bit */
/* Specify the target we are exporting */
args->target = V_008DFC_SQ_EXP_MRTZ;
args->compr = 0; /* COMP flag */
args->out[0] = LLVMGetUndef(ctx->f32); /* R, depth */
args->out[1] = LLVMGetUndef(ctx->f32); /* G, stencil test val[0:7], stencil op val[8:15] */
args->out[2] = LLVMGetUndef(ctx->f32); /* B, sample mask */
args->out[3] = LLVMGetUndef(ctx->f32); /* A, alpha to mask */
if (format == V_028710_SPI_SHADER_UINT16_ABGR) {
assert(!depth);
args->compr = 1; /* COMPR flag */
if (stencil) {
/* Stencil should be in X[23:16]. */
stencil = ac_to_integer(ctx, stencil);
stencil = LLVMBuildShl(ctx->builder, stencil,
LLVMConstInt(ctx->i32, 16, 0), "");
args->out[0] = ac_to_float(ctx, stencil);
mask |= 0x3;
}
if (samplemask) {
/* SampleMask should be in Y[15:0]. */
args->out[1] = samplemask;
mask |= 0xc;
}
} else {
if (depth) {
args->out[0] = depth;
mask |= 0x1;
}
if (stencil) {
args->out[1] = stencil;
mask |= 0x2;
}
if (samplemask) {
args->out[2] = samplemask;
mask |= 0x4;
}
}
/* GFX6 (except OLAND and HAINAN) has a bug that it only looks
* at the X writemask component. */
if (ctx->chip_class == GFX6 &&
ctx->family != CHIP_OLAND &&
ctx->family != CHIP_HAINAN)
mask |= 0x1;
/* Specify which components to enable */
args->enabled_channels = mask;
}

View File

@@ -28,7 +28,6 @@
#include <stdint.h> #include <stdint.h>
#include "amd_family.h" #include "amd_family.h"
#include "ac_llvm_build.h"
unsigned unsigned
ac_get_spi_shader_z_format(bool writes_z, bool writes_stencil, ac_get_spi_shader_z_format(bool writes_z, bool writes_stencil,
@@ -40,9 +39,4 @@ ac_get_cb_shader_mask(unsigned spi_shader_col_format);
uint32_t uint32_t
ac_vgt_gs_mode(unsigned gs_max_vert_out, enum chip_class chip_class); ac_vgt_gs_mode(unsigned gs_max_vert_out, enum chip_class chip_class);
void
ac_export_mrt_z(struct ac_llvm_context *ctx, LLVMValueRef depth,
LLVMValueRef stencil, LLVMValueRef samplemask,
struct ac_export_args *args);
#endif #endif