Re: [Mesa-dev] [PATCH 1/4] glsl_to_tgsi: declare all SSBOs and atomics when indirect indexing is used

2017-05-18 Thread Nicolai Hähnle 

For the series:

Reviewed-by: Nicolai Hähnle 

On 17.05.2017 21:14, Marek Olšák wrote:

From: Marek Olšák 

Only the first array element was declared, so tgsi_shader_info::
shader_buffers_declared didn't match what the shader was using.
---
 src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 25 +
 1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp 
b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index 54bc70f..e4d139e 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -413,21 +413,20 @@ public:
unsigned num_inputs;
unsigned num_input_arrays;
struct inout_decl outputs[4 * PIPE_MAX_SHADER_OUTPUTS];
unsigned num_outputs;
unsigned num_output_arrays;

int num_address_regs;
uint32_t samplers_used;
glsl_base_type sampler_types[PIPE_MAX_SAMPLERS];
int sampler_targets[PIPE_MAX_SAMPLERS];   /**< One of TGSI_TEXTURE_* */
-   int buffers_used;
int images_used;
int image_targets[PIPE_MAX_SHADER_IMAGES];
unsigned image_formats[PIPE_MAX_SHADER_IMAGES];
bool indirect_addr_consts;
int wpos_transform_const;

int glsl_version;
bool native_integers;
bool have_sqrt;
bool have_fma;
@@ -4536,21 +4535,20 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
array_sizes = NULL;
max_num_arrays = 0;
next_array = 0;
num_inputs = 0;
num_outputs = 0;
num_input_arrays = 0;
num_output_arrays = 0;
num_immediates = 0;
num_address_regs = 0;
samplers_used = 0;
-   buffers_used = 0;
images_used = 0;
indirect_addr_consts = false;
wpos_transform_const = -1;
glsl_version = 0;
native_integers = false;
mem_ctx = ralloc_context(NULL);
ctx = NULL;
prog = NULL;
shader_program = NULL;
shader = NULL;
@@ -4574,21 +4572,20 @@ extern "C" void 
free_glsl_to_tgsi_visitor(glsl_to_tgsi_visitor *v)


 /**
  * Count resources used by the given gpu program (number of texture
  * samplers, etc).
  */
 static void
 count_resources(glsl_to_tgsi_visitor *v, gl_program *prog)
 {
v->samplers_used = 0;
-   v->buffers_used = 0;
v->images_used = 0;

foreach_in_list(glsl_to_tgsi_instruction, inst, >instructions) {
   if (inst->info->is_tex) {
  for (int i = 0; i < inst->sampler_array_size; i++) {
 unsigned idx = inst->sampler_base + i;
 v->samplers_used |= 1u << idx;

 debug_assert(idx < (int)ARRAY_SIZE(v->sampler_types));
 v->sampler_types[idx] = inst->tex_type;
@@ -4601,21 +4598,21 @@ count_resources(glsl_to_tgsi_visitor *v, gl_program 
*prog)
  }
   }

   if (inst->tex_target == TEXTURE_EXTERNAL_INDEX)
  prog->ExternalSamplersUsed |= 1 << inst->resource.index;

   if (inst->resource.file != PROGRAM_UNDEFINED && (
 is_resource_instruction(inst->op) ||
 inst->op == TGSI_OPCODE_STORE)) {
  if (inst->resource.file == PROGRAM_BUFFER) {
-v->buffers_used |= 1 << inst->resource.index;
+/* nothing to do */
  } else if (inst->resource.file == PROGRAM_MEMORY) {
 v->use_shared_memory = true;
  } else {
 assert(inst->resource.file == PROGRAM_IMAGE);
 for (int i = 0; i < inst->sampler_array_size; i++) {
unsigned idx = inst->sampler_base + i;
v->images_used |= 1 << idx;
v->image_targets[idx] =
   st_translate_texture_target(inst->tex_target, false);
v->image_formats[idx] = inst->image_format;
@@ -6586,30 +6583,34 @@ st_translate_program(
 break;
  default:
 unreachable("not reached");
  }

  ureg_DECL_sampler_view( ureg, i, program->sampler_targets[i],
  type, type, type, type );
   }
}

-   for (i = 0; i < frag_const->MaxAtomicBuffers; i++) {
-  if (program->buffers_used & (1 << i)) {
- t->buffers[i] = ureg_DECL_buffer(ureg, i, true);
+   /* Declare atomic and shader storage buffers. */
+   {
+  struct gl_program *prog = program->prog;
+
+  for (i = 0; i < prog->info.num_abos; i++) {
+ unsigned index = prog->sh.AtomicBuffers[i]->Binding;
+ assert(index < frag_const->MaxAtomicBuffers);
+ t->buffers[index] = ureg_DECL_buffer(ureg, index, true);
   }
-   }

-   for (; i < frag_const->MaxAtomicBuffers + 
frag_const->MaxShaderStorageBlocks;
-i++) {
-  if (program->buffers_used & (1 << i)) {
- t->buffers[i] = ureg_DECL_buffer(ureg, i, false);
+  assert(prog->info.num_ssbos <= frag_const->MaxShaderStorageBlocks);
+  for (i = 0; i < prog->info.num_ssbos; i++) {
+ unsigned index = frag_const->MaxAtomicBuffers + i;
+ t->buffers[index] = ureg_DECL_buffer(ureg, index, false);
   }
}

if

Re: [Mesa-dev] [PATCH 1/4] glsl_to_tgsi: declare all SSBOs and atomics when indirect indexing is used

2017-05-17 Thread Samuel Pitoiset 



On 05/17/2017 09:14 PM, Marek Olšák wrote:

From: Marek Olšák 

Only the first array element was declared, so tgsi_shader_info::
shader_buffers_declared didn't match what the shader was using.
---
  src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 25 +
  1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp 
b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index 54bc70f..e4d139e 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -413,21 +413,20 @@ public:
 unsigned num_inputs;
 unsigned num_input_arrays;
 struct inout_decl outputs[4 * PIPE_MAX_SHADER_OUTPUTS];
 unsigned num_outputs;
 unsigned num_output_arrays;
  
 int num_address_regs;

 uint32_t samplers_used;
 glsl_base_type sampler_types[PIPE_MAX_SAMPLERS];
 int sampler_targets[PIPE_MAX_SAMPLERS];   /**< One of TGSI_TEXTURE_* */
-   int buffers_used;
 int images_used;
 int image_targets[PIPE_MAX_SHADER_IMAGES];
 unsigned image_formats[PIPE_MAX_SHADER_IMAGES];
 bool indirect_addr_consts;
 int wpos_transform_const;
  
 int glsl_version;

 bool native_integers;
 bool have_sqrt;
 bool have_fma;
@@ -4536,21 +4535,20 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
 array_sizes = NULL;
 max_num_arrays = 0;
 next_array = 0;
 num_inputs = 0;
 num_outputs = 0;
 num_input_arrays = 0;
 num_output_arrays = 0;
 num_immediates = 0;
 num_address_regs = 0;
 samplers_used = 0;
-   buffers_used = 0;
 images_used = 0;
 indirect_addr_consts = false;
 wpos_transform_const = -1;
 glsl_version = 0;
 native_integers = false;
 mem_ctx = ralloc_context(NULL);
 ctx = NULL;
 prog = NULL;
 shader_program = NULL;
 shader = NULL;
@@ -4574,21 +4572,20 @@ extern "C" void 
free_glsl_to_tgsi_visitor(glsl_to_tgsi_visitor *v)
  
  
  /**

   * Count resources used by the given gpu program (number of texture
   * samplers, etc).
   */
  static void
  count_resources(glsl_to_tgsi_visitor *v, gl_program *prog)
  {
 v->samplers_used = 0;
-   v->buffers_used = 0;
 v->images_used = 0;
  
 foreach_in_list(glsl_to_tgsi_instruction, inst, >instructions) {

if (inst->info->is_tex) {
   for (int i = 0; i < inst->sampler_array_size; i++) {
  unsigned idx = inst->sampler_base + i;
  v->samplers_used |= 1u << idx;
  
  debug_assert(idx < (int)ARRAY_SIZE(v->sampler_types));

  v->sampler_types[idx] = inst->tex_type;
@@ -4601,21 +4598,21 @@ count_resources(glsl_to_tgsi_visitor *v, gl_program 
*prog)
   }
}
  
if (inst->tex_target == TEXTURE_EXTERNAL_INDEX)

   prog->ExternalSamplersUsed |= 1 << inst->resource.index;
  
if (inst->resource.file != PROGRAM_UNDEFINED && (

  is_resource_instruction(inst->op) ||
  inst->op == TGSI_OPCODE_STORE)) {
   if (inst->resource.file == PROGRAM_BUFFER) {
-v->buffers_used |= 1 << inst->resource.index;
+/* nothing to do */


I would suggest to remove the whole 'if' and replace 'else' by 'else if' 
(for images).



   } else if (inst->resource.file == PROGRAM_MEMORY) {
  v->use_shared_memory = true;
   } else {
  assert(inst->resource.file == PROGRAM_IMAGE);
  for (int i = 0; i < inst->sampler_array_size; i++) {
 unsigned idx = inst->sampler_base + i;
 v->images_used |= 1 << idx;
 v->image_targets[idx] =
st_translate_texture_target(inst->tex_target, false);
 v->image_formats[idx] = inst->image_format;
@@ -6586,30 +6583,34 @@ st_translate_program(
  break;
   default:
  unreachable("not reached");
   }
  
   ureg_DECL_sampler_view( ureg, i, program->sampler_targets[i],

   type, type, type, type );
}
 }
  
-   for (i = 0; i < frag_const->MaxAtomicBuffers; i++) {

-  if (program->buffers_used & (1 << i)) {
- t->buffers[i] = ureg_DECL_buffer(ureg, i, true);
+   /* Declare atomic and shader storage buffers. */
+   {
+  struct gl_program *prog = program->prog;
+
+  for (i = 0; i < prog->info.num_abos; i++) {
+ unsigned index = prog->sh.AtomicBuffers[i]->Binding;
+ assert(index < frag_const->MaxAtomicBuffers);
+ t->buffers[index] = ureg_DECL_buffer(ureg, index, true);
}
-   }
  
-   for (; i < frag_const->MaxAtomicBuffers + frag_const->MaxShaderStorageBlocks;

-i++) {
-  if (program->buffers_used & (1 << i)) {
- t->buffers[i] = ureg_DECL_buffer(ureg, i, false);
+  assert(prog->info.num_ssbos <= frag_const->MaxShaderStorageBlocks);
+  for (i = 0; i < prog->info.num_ssbos; i++) {
+ unsigned index

[Mesa-dev] [PATCH 1/4] glsl_to_tgsi: declare all SSBOs and atomics when indirect indexing is used

2017-05-17 Thread Marek Olšák 
From: Marek Olšák 

Only the first array element was declared, so tgsi_shader_info::
shader_buffers_declared didn't match what the shader was using.
---
 src/mesa/state_tracker/st_glsl_to_tgsi.cpp | 25 +
 1 file changed, 13 insertions(+), 12 deletions(-)

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp 
b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index 54bc70f..e4d139e 100644
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -413,21 +413,20 @@ public:
unsigned num_inputs;
unsigned num_input_arrays;
struct inout_decl outputs[4 * PIPE_MAX_SHADER_OUTPUTS];
unsigned num_outputs;
unsigned num_output_arrays;
 
int num_address_regs;
uint32_t samplers_used;
glsl_base_type sampler_types[PIPE_MAX_SAMPLERS];
int sampler_targets[PIPE_MAX_SAMPLERS];   /**< One of TGSI_TEXTURE_* */
-   int buffers_used;
int images_used;
int image_targets[PIPE_MAX_SHADER_IMAGES];
unsigned image_formats[PIPE_MAX_SHADER_IMAGES];
bool indirect_addr_consts;
int wpos_transform_const;
 
int glsl_version;
bool native_integers;
bool have_sqrt;
bool have_fma;
@@ -4536,21 +4535,20 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
array_sizes = NULL;
max_num_arrays = 0;
next_array = 0;
num_inputs = 0;
num_outputs = 0;
num_input_arrays = 0;
num_output_arrays = 0;
num_immediates = 0;
num_address_regs = 0;
samplers_used = 0;
-   buffers_used = 0;
images_used = 0;
indirect_addr_consts = false;
wpos_transform_const = -1;
glsl_version = 0;
native_integers = false;
mem_ctx = ralloc_context(NULL);
ctx = NULL;
prog = NULL;
shader_program = NULL;
shader = NULL;
@@ -4574,21 +4572,20 @@ extern "C" void 
free_glsl_to_tgsi_visitor(glsl_to_tgsi_visitor *v)
 
 
 /**
  * Count resources used by the given gpu program (number of texture
  * samplers, etc).
  */
 static void
 count_resources(glsl_to_tgsi_visitor *v, gl_program *prog)
 {
v->samplers_used = 0;
-   v->buffers_used = 0;
v->images_used = 0;
 
foreach_in_list(glsl_to_tgsi_instruction, inst, >instructions) {
   if (inst->info->is_tex) {
  for (int i = 0; i < inst->sampler_array_size; i++) {
 unsigned idx = inst->sampler_base + i;
 v->samplers_used |= 1u << idx;
 
 debug_assert(idx < (int)ARRAY_SIZE(v->sampler_types));
 v->sampler_types[idx] = inst->tex_type;
@@ -4601,21 +4598,21 @@ count_resources(glsl_to_tgsi_visitor *v, gl_program 
*prog)
  }
   }
 
   if (inst->tex_target == TEXTURE_EXTERNAL_INDEX)
  prog->ExternalSamplersUsed |= 1 << inst->resource.index;
 
   if (inst->resource.file != PROGRAM_UNDEFINED && (
 is_resource_instruction(inst->op) ||
 inst->op == TGSI_OPCODE_STORE)) {
  if (inst->resource.file == PROGRAM_BUFFER) {
-v->buffers_used |= 1 << inst->resource.index;
+/* nothing to do */
  } else if (inst->resource.file == PROGRAM_MEMORY) {
 v->use_shared_memory = true;
  } else {
 assert(inst->resource.file == PROGRAM_IMAGE);
 for (int i = 0; i < inst->sampler_array_size; i++) {
unsigned idx = inst->sampler_base + i;
v->images_used |= 1 << idx;
v->image_targets[idx] =
   st_translate_texture_target(inst->tex_target, false);
v->image_formats[idx] = inst->image_format;
@@ -6586,30 +6583,34 @@ st_translate_program(
 break;
  default:
 unreachable("not reached");
  }
 
  ureg_DECL_sampler_view( ureg, i, program->sampler_targets[i],
  type, type, type, type );
   }
}
 
-   for (i = 0; i < frag_const->MaxAtomicBuffers; i++) {
-  if (program->buffers_used & (1 << i)) {
- t->buffers[i] = ureg_DECL_buffer(ureg, i, true);
+   /* Declare atomic and shader storage buffers. */
+   {
+  struct gl_program *prog = program->prog;
+
+  for (i = 0; i < prog->info.num_abos; i++) {
+ unsigned index = prog->sh.AtomicBuffers[i]->Binding;
+ assert(index < frag_const->MaxAtomicBuffers);
+ t->buffers[index] = ureg_DECL_buffer(ureg, index, true);
   }
-   }
 
-   for (; i < frag_const->MaxAtomicBuffers + 
frag_const->MaxShaderStorageBlocks;
-i++) {
-  if (program->buffers_used & (1 << i)) {
- t->buffers[i] = ureg_DECL_buffer(ureg, i, false);
+  assert(prog->info.num_ssbos <= frag_const->MaxShaderStorageBlocks);
+  for (i = 0; i < prog->info.num_ssbos; i++) {
+ unsigned index = frag_const->MaxAtomicBuffers + i;
+ t->buffers[index] = ureg_DECL_buffer(ureg, index, false);
   }
}
 
if (program->use_shared_memory)
   t->shared_memory = ureg_DECL_memory(ureg, TGSI_MEMORY_TYPE_SHARED);
 
for