**********************************************************/
#include "util/u_inlines.h"
+#include "util/u_memory.h"
#include "pipe/p_defines.h"
#include "svga_screen.h"
#include "svga_cmd.h"
#include "svga_tgsi.h"
#include "svga_debug.h"
+#include "svga_resource_buffer.h"
#include "svga_hw_reg.h"
/*
- * Don't try to send more than 4k of successive constants.
+ * Don't try to send more than 4kb of successive constants.
*/
-#define MAX_CONST_REG_COUNT 256 /* 4k */
+#define MAX_CONST_REG_COUNT 256 /**< number of float[4] constants */
+/**
+ * Extra space for svga-specific VS/PS constants (such as texcoord
+ * scale factors, vertex transformation scale/translation).
+ */
+#define MAX_EXTRA_CONSTS 32
+
+/** Guest-backed surface constant buffers must be this size */
+#define GB_CONSTBUF_SIZE (SVGA3D_CONSTREG_MAX)
-/***********************************************************************
- * Hardware update
+/**
+ * Convert from PIPE_SHADER_* to SVGA3D_SHADERTYPE_*
*/
+static unsigned
+svga_shader_type(unsigned shader)
+{
+ switch (shader) {
+ case PIPE_SHADER_VERTEX:
+ return SVGA3D_SHADERTYPE_VS;
+ case PIPE_SHADER_FRAGMENT:
+ return SVGA3D_SHADERTYPE_PS;
+ default:
+ assert(!"Unexpected shader type");
+ return SVGA3D_SHADERTYPE_VS;
+ }
+}
+
-/* Convert from PIPE_SHADER_* to SVGA3D_SHADERTYPE_*
+/**
+ * Emit any extra fragment shader constants into the buffer pointed
+ * to by 'dest'.
+ * In particular, these would be the scaling factors needed for handling
+ * unnormalized texture coordinates for texture rectangles.
+ * \return number of float[4] constants put into the dest buffer
*/
-static int svga_shader_type( int shader )
+static unsigned
+svga_get_extra_fs_constants(struct svga_context *svga, float *dest)
{
- assert(PIPE_SHADER_VERTEX + 1 == SVGA3D_SHADERTYPE_VS);
- assert(PIPE_SHADER_FRAGMENT + 1 == SVGA3D_SHADERTYPE_PS);
- assert(shader <= PIPE_SHADER_FRAGMENT);
- return shader + 1;
+ const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
+ const struct svga_fs_compile_key *key = &variant->key.fkey;
+ unsigned count = 0;
+
+ /* SVGA_NEW_VS_VARIANT
+ */
+ if (key->num_unnormalized_coords) {
+ unsigned i;
+
+ for (i = 0; i < key->num_textures; i++) {
+ if (key->tex[i].unnormalized) {
+ struct pipe_resource *tex = svga->curr.sampler_views[i]->texture;
+
+ /* debug/sanity check */
+ assert(key->tex[i].width_height_idx == count);
+
+ *dest++ = 1.0 / (float)tex->width0;
+ *dest++ = 1.0 / (float)tex->height0;
+ *dest++ = 1.0;
+ *dest++ = 1.0;
+
+ count++;
+ }
+ }
+ }
+
+ assert(count <= MAX_EXTRA_CONSTS);
+
+ return count;
}
-/*
+
+/**
+ * Emit any extra vertex shader constants into the buffer pointed
+ * to by 'dest'.
+ * In particular, these would be the scale and bias factors computed
+ * from the framebuffer size which are used to copy with differences in
+ * GL vs D3D coordinate spaces. See svga_tgsi_insn.c for more info.
+ * \return number of float[4] constants put into the dest buffer
+ */
+static unsigned
+svga_get_extra_vs_constants(struct svga_context *svga, float *dest)
+{
+ const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
+ const struct svga_vs_compile_key *key = &variant->key.vkey;
+ unsigned count = 0;
+
+ /* SVGA_NEW_VS_VARIANT
+ */
+ if (key->need_prescale) {
+ memcpy(dest, svga->state.hw_clear.prescale.scale, 4 * sizeof(float));
+ dest += 4;
+
+ memcpy(dest, svga->state.hw_clear.prescale.translate, 4 * sizeof(float));
+ dest += 4;
+
+ count = 2;
+ }
+
+ assert(count <= MAX_EXTRA_CONSTS);
+
+ return count;
+}
+
+
+/**
* Check and emit one shader constant register.
+ * \param shader PIPE_SHADER_FRAGMENT or PIPE_SHADER_VERTEX
+ * \param i which float[4] constant to change
+ * \param value the new float[4] value
*/
-static int emit_const( struct svga_context *svga,
- int unit,
- int i,
- const float *value )
+static enum pipe_error
+emit_const(struct svga_context *svga, unsigned shader, unsigned i,
+ const float *value)
{
- int ret = PIPE_OK;
+ enum pipe_error ret = PIPE_OK;
- assert(i < CB_MAX);
+ assert(shader < PIPE_SHADER_TYPES);
+ assert(i < SVGA3D_CONSTREG_MAX);
- if (memcmp(svga->state.hw_draw.cb[unit][i], value, 4 * sizeof(float)) != 0) {
+ if (memcmp(svga->state.hw_draw.cb[shader][i], value,
+ 4 * sizeof(float)) != 0) {
if (SVGA_DEBUG & DEBUG_CONSTS)
- debug_printf("%s %s %d: %f %f %f %f\n",
+ debug_printf("%s %s %u: %f %f %f %f\n",
__FUNCTION__,
- unit == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
+ shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
i,
value[0],
value[1],
value[2],
value[3]);
- ret = SVGA3D_SetShaderConst( svga->swc,
+ ret = SVGA3D_SetShaderConst( svga->swc,
i,
- svga_shader_type(unit),
+ svga_shader_type(shader),
SVGA3D_CONST_TYPE_FLOAT,
value );
- if (ret)
+ if (ret != PIPE_OK)
return ret;
- memcpy(svga->state.hw_draw.cb[unit][i], value, 4 * sizeof(float));
+ memcpy(svga->state.hw_draw.cb[shader][i], value, 4 * sizeof(float));
}
-
+
return ret;
}
+
/*
* Check and emit a range of shader constant registers, trying to coalesce
* successive shader constant updates in a single command in order to save
* space on the command buffer. This is a HWv8 feature.
*/
-static enum pipe_error emit_const_range( struct svga_context *svga,
- unsigned unit,
- unsigned offset,
- unsigned count,
- const float (*values)[4] )
+static enum pipe_error
+emit_const_range(struct svga_context *svga,
+ unsigned shader,
+ unsigned offset,
+ unsigned count,
+ const float (*values)[4])
{
unsigned i, j;
enum pipe_error ret;
#ifdef DEBUG
- if (offset + count > CB_MAX) {
- debug_printf("svga: too many constants (offset + count = %u)\n",
- offset + count);
+ if (offset + count > SVGA3D_CONSTREG_MAX) {
+ debug_printf("svga: too many constants (offset %u + count %u = %u (max = %u))\n",
+ offset, count, offset + count, SVGA3D_CONSTREG_MAX);
}
#endif
- if (offset > CB_MAX) {
+ if (offset > SVGA3D_CONSTREG_MAX) {
/* This isn't OK, but if we propagate an error all the way up we'll
* just get into more trouble.
* XXX note that offset is always zero at this time so this is moot.
return PIPE_OK;
}
- if (offset + count > CB_MAX) {
+ if (offset + count > SVGA3D_CONSTREG_MAX) {
/* Just drop the extra constants for now.
* Ideally we should not have allowed the app to create a shader
* that exceeds our constant buffer size but there's no way to
* express that in gallium at this time.
*/
- count = CB_MAX - offset;
+ count = SVGA3D_CONSTREG_MAX - offset;
}
i = 0;
while (i < count) {
- if (memcmp(svga->state.hw_draw.cb[unit][offset + i],
+ if (memcmp(svga->state.hw_draw.cb[shader][offset + i],
values[i],
4 * sizeof(float)) != 0) {
-
- /*
- * Found one dirty constant
+ /* Found one dirty constant
*/
-
if (SVGA_DEBUG & DEBUG_CONSTS)
debug_printf("%s %s %d: %f %f %f %f\n",
__FUNCTION__,
- unit == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
+ shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
offset + i,
values[i][0],
values[i][1],
values[i][2],
values[i][3]);
- /*
- * Look for more consecutive dirty constants.
+ /* Look for more consecutive dirty constants.
*/
-
j = i + 1;
while (j < count &&
j < i + MAX_CONST_REG_COUNT &&
- memcmp(svga->state.hw_draw.cb[unit][offset + j],
+ memcmp(svga->state.hw_draw.cb[shader][offset + j],
values[j],
4 * sizeof(float)) != 0) {
if (SVGA_DEBUG & DEBUG_CONSTS)
debug_printf("%s %s %d: %f %f %f %f\n",
__FUNCTION__,
- unit == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
+ shader == PIPE_SHADER_VERTEX ? "VERT" : "FRAG",
offset + j,
values[j][0],
values[j][1],
assert(j >= i + 1);
- /*
- * Send them all together.
+ /* Send them all together.
*/
-
- ret = SVGA3D_SetShaderConsts(svga->swc,
- offset + i, j - i,
- svga_shader_type(unit),
- SVGA3D_CONST_TYPE_FLOAT,
- values + i);
+ if (svga_have_gb_objects(svga)) {
+ ret = SVGA3D_SetGBShaderConstsInline(svga->swc,
+ offset + i, /* start */
+ j - i, /* count */
+ svga_shader_type(shader),
+ SVGA3D_CONST_TYPE_FLOAT,
+ values + i);
+ }
+ else {
+ ret = SVGA3D_SetShaderConsts(svga->swc,
+ offset + i, j - i,
+ svga_shader_type(shader),
+ SVGA3D_CONST_TYPE_FLOAT,
+ values + i);
+ }
if (ret != PIPE_OK) {
return ret;
}
/*
* Local copy of the hardware state.
*/
-
- memcpy(svga->state.hw_draw.cb[unit][offset + i],
+ memcpy(svga->state.hw_draw.cb[shader][offset + i],
values[i],
(j - i) * 4 * sizeof(float));
return PIPE_OK;
}
-static int emit_consts( struct svga_context *svga,
- int offset,
- int unit )
+
+/**
+ * Emit all the constants in a constant buffer for a shader stage.
+ */
+static enum pipe_error
+emit_consts(struct svga_context *svga, unsigned shader)
{
struct svga_screen *ss = svga_screen(svga->pipe.screen);
struct pipe_transfer *transfer = NULL;
unsigned count;
const float (*data)[4] = NULL;
unsigned i;
- int ret = PIPE_OK;
+ enum pipe_error ret = PIPE_OK;
+ const unsigned offset = 0;
- if (svga->curr.cb[unit] == NULL)
- goto done;
+ assert(shader < PIPE_SHADER_TYPES);
- count = svga->curr.cb[unit]->width0 / (4 * sizeof(float));
+ if (svga->curr.cbufs[shader].buffer == NULL)
+ goto done;
data = (const float (*)[4])pipe_buffer_map(&svga->pipe,
- svga->curr.cb[unit],
+ svga->curr.cbufs[shader].buffer,
PIPE_TRANSFER_READ,
&transfer);
if (data == NULL) {
goto done;
}
+ /* sanity check */
+ assert(svga->curr.cbufs[shader].buffer->width0 >=
+ svga->curr.cbufs[shader].buffer_size);
+
+ /* Use/apply the constant buffer size and offsets here */
+ count = svga->curr.cbufs[shader].buffer_size / (4 * sizeof(float));
+ data += svga->curr.cbufs[shader].buffer_offset / (4 * sizeof(float));
+
if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
- ret = emit_const_range( svga, unit, offset, count, data );
+ ret = emit_const_range( svga, shader, offset, count, data );
if (ret != PIPE_OK) {
goto done;
}
} else {
for (i = 0; i < count; i++) {
- ret = emit_const( svga, unit, offset + i, data[i] );
+ ret = emit_const( svga, shader, offset + i, data[i] );
if (ret != PIPE_OK) {
goto done;
}
return ret;
}
-
-static int emit_fs_consts( struct svga_context *svga,
- unsigned dirty )
+
+
+static enum pipe_error
+emit_fs_consts(struct svga_context *svga, unsigned dirty)
{
- const struct svga_shader_result *result = svga->state.hw_draw.fs;
- const struct svga_fs_compile_key *key = &result->key.fkey;
- int ret = 0;
+ struct svga_screen *ss = svga_screen(svga->pipe.screen);
+ const struct svga_shader_variant *variant = svga->state.hw_draw.fs;
+ enum pipe_error ret = PIPE_OK;
- ret = emit_consts( svga, 0, PIPE_SHADER_FRAGMENT );
- if (ret)
- return ret;
+ /* SVGA_NEW_FS_VARIANT
+ */
+ if (variant == NULL)
+ return PIPE_OK;
- /* The internally generated fragment shader for xor blending
- * doesn't have a 'result' struct. It should be fixed to avoid
- * this special case, but work around it with a NULL check:
+ /* SVGA_NEW_FS_CONST_BUFFER
*/
- if (result != NULL &&
- key->num_unnormalized_coords)
+ ret = emit_consts( svga, PIPE_SHADER_FRAGMENT );
+ if (ret != PIPE_OK)
+ return ret;
+
+ /* emit extra shader constants */
{
- unsigned offset = result->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
- int i;
+ unsigned offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
+ float extras[MAX_EXTRA_CONSTS][4];
+ unsigned count, i;
- for (i = 0; i < key->num_textures; i++) {
- if (key->tex[i].unnormalized) {
- struct pipe_resource *tex = svga->curr.sampler_views[i]->texture;
- float data[4];
-
- data[0] = 1.0 / (float)tex->width0;
- data[1] = 1.0 / (float)tex->height0;
- data[2] = 1.0;
- data[3] = 1.0;
-
- ret = emit_const( svga,
- PIPE_SHADER_FRAGMENT,
- key->tex[i].width_height_idx + offset,
- data );
- if (ret)
+ count = svga_get_extra_fs_constants(svga, (float *) extras);
+
+ if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
+ ret = emit_const_range(svga, PIPE_SHADER_FRAGMENT, offset, count,
+ (const float (*) [4])extras);
+ } else {
+ for (i = 0; i < count; i++) {
+ ret = emit_const(svga, PIPE_SHADER_FRAGMENT, offset + i, extras[i]);
+ if (ret != PIPE_OK)
return ret;
}
}
-
- offset += key->num_unnormalized_coords;
}
- return 0;
+ return ret;
}
-struct svga_tracked_state svga_hw_fs_parameters =
+struct svga_tracked_state svga_hw_fs_constants =
{
"hw fs params",
(SVGA_NEW_FS_CONST_BUFFER |
- SVGA_NEW_FS_RESULT |
+ SVGA_NEW_FS_VARIANT |
SVGA_NEW_TEXTURE_BINDING),
emit_fs_consts
};
-/***********************************************************************
- */
-static int emit_vs_consts( struct svga_context *svga,
- unsigned dirty )
+
+static enum pipe_error
+emit_vs_consts(struct svga_context *svga, unsigned dirty)
{
- const struct svga_shader_result *result = svga->state.hw_draw.vs;
- const struct svga_vs_compile_key *key = &result->key.vkey;
- int ret = 0;
- unsigned offset;
+ struct svga_screen *ss = svga_screen(svga->pipe.screen);
+ const struct svga_shader_variant *variant = svga->state.hw_draw.vs;
+ enum pipe_error ret = PIPE_OK;
- /* SVGA_NEW_VS_RESULT
+ /* SVGA_NEW_VS_VARIANT
*/
- if (result == NULL)
- return 0;
+ if (variant == NULL)
+ return PIPE_OK;
- /* SVGA_NEW_VS_CONST_BUFFER
+ /* SVGA_NEW_VS_CONST_BUFFER
*/
- ret = emit_consts( svga, 0, PIPE_SHADER_VERTEX );
- if (ret)
+ ret = emit_consts( svga, PIPE_SHADER_VERTEX );
+ if (ret != PIPE_OK)
return ret;
- offset = result->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
-
- /* SVGA_NEW_VS_RESULT
- */
- if (key->need_prescale) {
- ret = emit_const( svga, PIPE_SHADER_VERTEX, offset++,
- svga->state.hw_clear.prescale.scale );
- if (ret)
- return ret;
+ /* emit extra shader constants */
+ {
+ unsigned offset = variant->shader->info.file_max[TGSI_FILE_CONSTANT] + 1;
+ float extras[MAX_EXTRA_CONSTS][4];
+ unsigned count, i;
- ret = emit_const( svga, PIPE_SHADER_VERTEX, offset++,
- svga->state.hw_clear.prescale.translate );
- if (ret)
- return ret;
- }
+ count = svga_get_extra_vs_constants(svga, (float *) extras);
+ assert(count <= Elements(extras));
- /* SVGA_NEW_ZERO_STRIDE
- */
- if (key->zero_stride_vertex_elements) {
- unsigned i, curr_zero_stride = 0;
- for (i = 0; i < PIPE_MAX_ATTRIBS; ++i) {
- if (key->zero_stride_vertex_elements & (1 << i)) {
- ret = emit_const( svga, PIPE_SHADER_VERTEX, offset++,
- svga->curr.zero_stride_constants +
- 4 * curr_zero_stride );
- if (ret)
+ if (ss->hw_version >= SVGA3D_HWVERSION_WS8_B1) {
+ ret = emit_const_range(svga, PIPE_SHADER_VERTEX, offset, count,
+ (const float (*) [4]) extras);
+ } else {
+ for (i = 0; i < count; i++) {
+ ret = emit_const(svga, PIPE_SHADER_VERTEX, offset + i, extras[i]);
+ if (ret != PIPE_OK)
return ret;
- ++curr_zero_stride;
}
}
}
- return 0;
+ return ret;
}
-struct svga_tracked_state svga_hw_vs_parameters =
+struct svga_tracked_state svga_hw_vs_constants =
{
"hw vs params",
(SVGA_NEW_PRESCALE |
SVGA_NEW_VS_CONST_BUFFER |
- SVGA_NEW_ZERO_STRIDE |
- SVGA_NEW_VS_RESULT),
+ SVGA_NEW_VS_VARIANT),
emit_vs_consts
};
-
projects / mesa.git / blobdiff