diff --git a/src/graphics/lib/compute/hotsort/BUILD.gn b/src/graphics/lib/compute/hotsort/BUILD.gn
index 307947ea383e90260f2c9a1934f69f12e1bcd6e9..f6fe8d87068dc5631d8f39ddfbcc01ebb2d24c05 100644
--- a/src/graphics/lib/compute/hotsort/BUILD.gn
+++ b/src/graphics/lib/compute/hotsort/BUILD.gn
@@ -4,6 +4,13 @@
group("hotsort") {
deps = [
- "hotsort_gen($host_toolchain)",
+ "platforms/vk",
+ ]
+}
+
+group("tests") {
+ testonly = true
+ public_deps = [
+ "platforms/vk/tests",
]
}
diff --git a/src/graphics/lib/compute/hotsort/hotsort_gen/gen.h b/src/graphics/lib/compute/hotsort/hotsort_gen/gen.h
index 814dd008e22a50e5a88fc8844f3107dc67f5e11a..6a4a44f92ad4c017f5e0b11a2e317ebef3cbd005 100644
--- a/src/graphics/lib/compute/hotsort/hotsort_gen/gen.h
+++ b/src/graphics/lib/compute/hotsort/hotsort_gen/gen.h
@@ -70,6 +70,17 @@ struct hsg_config
struct {
uint32_t dwords; // number of dwords in a key -- .type will be extended to support key-vals
} type;
+
+ struct { // This provides GLSL-specific support for specifying the set and binding
+ struct { // locations for the input and output buffers. Note that this will be
+ uint32_t set; // removed once supported Vulkan targets support GLSL_EXT_buffer_reference.
+ uint32_t binding;
+ } in;
+ struct {
+ uint32_t set;
+ uint32_t binding;
+ } out;
+ } glsl;
};
//
diff --git a/src/graphics/lib/compute/hotsort/hotsort_gen/main.c b/src/graphics/lib/compute/hotsort/hotsort_gen/main.c
index 73c3ad1e8978e3a05e9a384b57098a7715221d15..9a39238920778ceb35d9260679084189e7315f0d 100644
--- a/src/graphics/lib/compute/hotsort/hotsort_gen/main.c
+++ b/src/graphics/lib/compute/hotsort/hotsort_gen/main.c
@@ -6,12 +6,13 @@
//
//
-#include <stdlib.h>
-#include <stdbool.h>
-#include <string.h>
-#include <getopt.h>
#include <ctype.h>
+#include <getopt.h>
#include <inttypes.h>
+#include <limits.h>
+#include <stdbool.h>
+#include <stdlib.h>
+#include <string.h>
#include <unistd.h>
//
@@ -143,14 +144,14 @@ struct hsg_config hsg_config =
.lo = 1,
.hi = 1
},
- .half = {
+ .half = {
.warps = 1,
.lo = 1,
.hi = 1
},
},
- .block = {
+ .block = {
.warps_min = 1,
.warps_max = UINT32_MAX,
.warps_mod = 2,
@@ -162,7 +163,7 @@ struct hsg_config hsg_config =
.smem_bc = UINT32_MAX
},
- .warp = {
+ .warp = {
.lanes = 32,
.lanes_log2 = 5,
},
@@ -172,8 +173,19 @@ struct hsg_config hsg_config =
.xtra = 0
},
- .type = {
+ .type = {
.dwords = 2
+ },
+
+ .glsl = {
+ .in = {
+ .set = 0,
+ .binding = 1
+ },
+ .out = {
+ .set = 0,
+ .binding = 0
+ }
}
};
@@ -1464,7 +1476,7 @@ main(int argc, char * argv[])
struct hsg_target target;
- while ((opt = getopt(argc,argv,"hvzo:a:g:G:s:S:w:b:B:m:M:k:r:x:t:f:F:c:C:p:P:D:")) != EOF)
+ while ((opt = getopt(argc,argv,"hvzo:a:g:G:s:S:w:b:B:m:M:k:r:x:t:f:F:c:C:p:P:D:L:")) != EOF)
{
switch (opt)
{
@@ -1613,6 +1625,33 @@ main(int argc, char * argv[])
hsg_optarg(optarg),
HSG_CONFIG_DEFINE_LEN_SIZE-1);
break;
+
+ case 'L':
+ {
+ // expects 4 concatenated and comma-separated unsigned integers
+ char * str = optarg;
+
+ if ((hsg_config.glsl.in.set = (uint32_t)strtoul(str,&str,10)) == (uint32_t)ULONG_MAX)
+ return EXIT_FAILURE;
+
+ if (*str++ != ',')
+ return EXIT_FAILURE;
+
+ if ((hsg_config.glsl.in.binding = (uint32_t)strtoul(str,&str,10)) == (uint32_t)ULONG_MAX)
+ return EXIT_FAILURE;
+
+ if (*str++ != ',')
+ return EXIT_FAILURE;
+
+ if ((hsg_config.glsl.out.set = (uint32_t)strtoul(str,&str,10)) == (uint32_t)ULONG_MAX)
+ return EXIT_FAILURE;
+
+ if (*str++ != ',')
+ return EXIT_FAILURE;
+
+ if ((hsg_config.glsl.out.binding = (uint32_t)strtoul(str,&str,10)) == (uint32_t)ULONG_MAX)
+ return EXIT_FAILURE;
+ }
}
}
diff --git a/src/graphics/lib/compute/hotsort/hotsort_gen/target_glsl.c b/src/graphics/lib/compute/hotsort/hotsort_gen/target_glsl.c
index eb58d89fb863a51b1bc1ed0e41e4fd27732475ec..3bc61a382d15185633ce2f81ae17cf99cb0b043f 100644
--- a/src/graphics/lib/compute/hotsort/hotsort_gen/target_glsl.c
+++ b/src/graphics/lib/compute/hotsort/hotsort_gen/target_glsl.c
@@ -93,8 +93,8 @@ hsg_macros(struct hsg_config const * const config, FILE * file)
"// target-specific config \n"
"#include \"hs_config.h\" \n"
" \n"
- "// arch/target-specific macros \n"
- "#include \"hs_glsl_macros.h\" \n"
+ "// vendor<arch<target-specific macros \n"
+ "#include \"hs_glsl_macros_vendor.h\" \n"
" \n"
"// \n"
"// \n"
@@ -187,7 +187,7 @@ hsg_target_header_and_module(struct hsg_config const * const config)
" } \n"
"}; \n"
" \n"
- "#include \"hs_target_modules_init.inl\" \n"
+ "#include \"hs_target_modules_dump.inl\" \n"
" \n"
"// \n"
"// \n"
@@ -288,9 +288,14 @@ hsg_target_glsl(struct hsg_target * const target,
"#define HS_HM_BLOCK_HEIGHT %u \n"
"#define HS_HM_SCALE_MIN %u \n"
"#define HS_HM_SCALE_MAX %u \n"
- "#define HS_EMPTY \n\n",
- config->warp.lanes_log2, // FIXME -- this matters for SIMD
- config->warp.lanes_log2,
+ "#define HS_EMPTY \n"
+ "#define HS_KV_IN_SET %u \n"
+ "#define HS_KV_IN_BINDING %u \n"
+ "#define HS_KV_OUT_SET %u \n"
+ "#define HS_KV_OUT_BINDING %u \n"
+ "#define HS_IS_IN_PLACE ((HS_KV_IN_SET == HS_KV_OUT_SET) && (HS_KV_IN_BINDING == HS_KV_OUT_BINDING))\n\n",
+ config->warp.lanes_log2, // NOTE: the number of slab threads might
+ config->warp.lanes_log2, // NOTE: not always equal slab width
config->thread.regs,
config->thread.regs,
config->type.dwords,
@@ -302,7 +307,11 @@ hsg_target_glsl(struct hsg_target * const target,
config->merge.flip.hi,
config->merge.half.warps,
config->merge.half.lo,
- config->merge.half.hi);
+ config->merge.half.hi,
+ config->glsl.in.set,
+ config->glsl.in.binding,
+ config->glsl.out.set,
+ config->glsl.out.binding);
fprintf(target->state->header,
"#define HS_SLAB_ROWS() \\\n");
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..417567f1278a1bdea1d377e8a28040038abe998c
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/BUILD.gn
@@ -0,0 +1,22 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+source_set("vk") {
+ public = [
+ "hotsort_vk.h",
+ ]
+ sources = [
+ "hotsort_vk.h",
+ "hotsort_vk.c",
+ "hotsort_vk_target.h",
+ ]
+ include_dirs = [
+ "//src/graphics/lib/compute"
+ ]
+ deps = [
+ "//src/graphics/lib/compute/common",
+ "//src/graphics/lib/compute/common/vk",
+ "//third_party/vulkan_loader_and_validation_layers:vulkan",
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/vk/hs_vk.c b/src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk.c
similarity index 80%
rename from src/graphics/lib/compute/hotsort/vk/hs_vk.c
rename to src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk.c
index 7f2be4fd212a8c4a5a4a0c0e7ff9abde0681050d..7da1f17fe576b158542f311dcc1b210a91aa7370 100644
--- a/src/graphics/lib/compute/hotsort/vk/hs_vk.c
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk.c
@@ -12,12 +12,12 @@
#include "common/vk/vk_assert.h"
#include "common/vk/vk_barrier.h"
-#if defined(HS_VK_SHADER_INFO_AMD_STATISTICS) || defined(HS_VK_SHADER_INFO_AMD_DISASSEMBLY)
+#if defined(HOTSORT_VK_SHADER_INFO_AMD_STATISTICS) || defined(HOTSORT_VK_SHADER_INFO_AMD_DISASSEMBLY)
#include "common/vk/vk_shader_info_amd.h"
#endif
-#include "hs_vk.h"
-#include "hs_vk_target.h"
+#include "hotsort_vk.h"
+#include "hotsort_vk_target.h"
//
// We want concurrent kernel execution to occur in a few places.
@@ -60,7 +60,7 @@
// |/ YES
// + |
// | v
-// | END_WITH_EVENTS(bs_full,bs_frac)
+// | END_WITH(bs_full,bs_frac)
// |
// |
// WAITFOR(pad_out,bs_full,bs_frac) >>> first iteration of loop <<<
@@ -95,32 +95,30 @@
// YES
// |
// v
-// END_WITH_EVENTS(bc)
+// END_WITH(bc)
//
-struct hs_vk
+struct hotsort_vk
{
- struct hs_vk_target_config config;
+ struct hotsort_vk_target_config config;
- bool is_in_place;
+ uint32_t slab_keys;
+ uint32_t key_val_size;
+ uint32_t bs_slabs_log2_ru;
+ uint32_t bc_slabs_log2_max;
- uint32_t slab_keys;
- uint32_t key_val_size;
- uint32_t bs_slabs_log2_ru;
- uint32_t bc_slabs_log2_max;
-
- VkPipelineLayout pl;
+ VkPipelineLayout pl;
struct {
- uint32_t count;
- VkPipeline * bs;
- VkPipeline * bc;
- VkPipeline * fm[3];
- VkPipeline * hm[3];
- VkPipeline * fill_in;
- VkPipeline * fill_out;
- VkPipeline * transpose;
- VkPipeline all[];
+ uint32_t count;
+ VkPipeline * bs;
+ VkPipeline * bc;
+ VkPipeline * fm[3];
+ VkPipeline * hm[3];
+ VkPipeline * fill_in;
+ VkPipeline * fill_out;
+ VkPipeline * transpose;
+ VkPipeline all[];
} pipelines;
};
@@ -128,13 +126,12 @@ struct hs_vk
//
//
-struct hs_vk *
-hs_vk_create(VkDevice device,
- VkAllocationCallbacks const * allocator,
- VkPipelineCache pc,
- VkPipelineLayout pl,
- struct hs_vk_ds_locations const * const locations,
- struct hs_vk_target const * const target)
+struct hotsort_vk *
+hotsort_vk_create(VkDevice device,
+ VkAllocationCallbacks const * allocator,
+ VkPipelineCache pipeline_cache,
+ VkPipelineLayout pipeline_layout,
+ struct hotsort_vk_target const * const target)
{
//
// we reference these values a lot
@@ -177,14 +174,9 @@ hs_vk_create(VkDevice device,
uint32_t const count_all = count_bs + count_bc_fm_hm_fills_transpose;
//
- // allocate hs_vk
+ // allocate hotsort_vk
//
- struct hs_vk * const hs = malloc(sizeof(*hs) + sizeof(VkPipeline*) * count_all);
-
- // in-place sort?
- hs->is_in_place =
- (locations->in.set == locations->out.set) &&
- (locations->in.binding == locations->out.binding);
+ struct hotsort_vk * const hs = malloc(sizeof(*hs) + sizeof(VkPipeline*) * count_all);
// copy the config from the target -- we need these values later
hs->config = target->config;
@@ -196,7 +188,7 @@ hs_vk_create(VkDevice device,
hs->bc_slabs_log2_max = bc_slabs_log2_max;
// save pipeline layout for vkCmdPushConstants()
- hs->pl = pl;
+ hs->pl = pipeline_layout;
// save kernel count
hs->pipelines.count = count_all;
@@ -217,7 +209,7 @@ hs_vk_create(VkDevice device,
.pName = "main",
.pSpecializationInfo = NULL
},
- .layout = pl,
+ .layout = pipeline_layout,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0
};
@@ -226,32 +218,29 @@ hs_vk_create(VkDevice device,
// Create a shader module, use it to create a pipeline... and
// dispose of the shader module.
//
- // The BS compute shaders have the same layout
- // The non-BS compute shaders have the same layout
+ // BS shaders have layout: (vout,vin)
+ // FILL_IN shaders have layout: (----,vin)
+ // FILL_OUT shaders have layout: (vout)
+ // otherwise shaders have layout: (vout)
//
VkShaderModuleCreateInfo smci = {
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = NULL,
.flags = 0,
.codeSize = 0,
- .pCode = (uint32_t const *)target->modules // FIXME -- unfortunate typecast -- fix during generation
+ .pCode = NULL
};
- //
- // BS shaders have layout: (vout,vin)
- // non-BS shaders have layout: (vout)
- //
+ uint32_t const * modules = target->modules;
+
for (uint32_t ii=0; ii<count_all; ii++)
{
- // convert bytes to words
- uint32_t const * const module = smci.pCode + smci.codeSize / sizeof(*module);
+ uint32_t const module_dwords = *modules++;
- //
- // FIXME -- patch descriptor set locations
- //
+ smci.codeSize = module_dwords * sizeof(*modules);
+ smci.pCode = modules;
- smci.codeSize = NTOHL_MACRO(module[0]);
- smci.pCode = module + 1;
+ modules += module_dwords;
vk(CreateShaderModule(device,
&smci,
@@ -259,7 +248,7 @@ hs_vk_create(VkDevice device,
&cpci.stage.module));
vk(CreateComputePipelines(device,
- pc,
+ pipeline_cache,
1,
&cpci,
allocator,
@@ -322,10 +311,10 @@ hs_vk_create(VkDevice device,
//
// optionally dump pipeline stats
//
-#ifdef HS_VK_SHADER_INFO_AMD_STATISTICS
+#ifdef HOTSORT_VK_SHADER_INFO_AMD_STATISTICS
vk_shader_info_amd_statistics(device,hs->pipelines.all,NULL,hs->pipelines.count);
#endif
-#ifdef HS_VK_SHADER_INFO_AMD_DISASSEMBLY
+#ifdef HOTSORT_VK_SHADER_INFO_AMD_DISASSEMBLY
vk_shader_info_amd_disassembly(device,hs->pipelines.all,NULL,hs->pipelines.count);
#endif
@@ -340,9 +329,9 @@ hs_vk_create(VkDevice device,
//
void
-hs_vk_release(VkDevice device,
- VkAllocationCallbacks const * const allocator,
- struct hs_vk * const hs)
+hotsort_vk_release(VkDevice device,
+ VkAllocationCallbacks const * const allocator,
+ struct hotsort_vk * const hs)
{
for (uint32_t ii=0; ii<hs->pipelines.count; ii++)
{
@@ -359,10 +348,10 @@ hs_vk_release(VkDevice device,
//
void
-hs_vk_pad(struct hs_vk const * const hs,
- uint32_t const count,
- uint32_t * const padded_in,
- uint32_t * const padded_out)
+hotsort_vk_pad(struct hotsort_vk const * const hs,
+ uint32_t const count,
+ uint32_t * const padded_in,
+ uint32_t * const padded_out)
{
//
// round up the count to slabs
@@ -411,9 +400,9 @@ hs_vk_pad(struct hs_vk const * const hs,
static
void
-hs_cmd_transpose(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- uint32_t const bx_ru)
+hs_cmd_transpose(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ uint32_t const bx_ru)
{
vkCmdBindPipeline(cb,
VK_PIPELINE_BIND_POINT_COMPUTE,
@@ -428,10 +417,10 @@ hs_cmd_transpose(VkCommandBuffer cb,
static
void
-hs_cmd_fill_in(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- uint32_t const from_slab,
- uint32_t const to_slab)
+hs_cmd_fill_in(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ uint32_t const from_slab,
+ uint32_t const to_slab)
{
vkCmdBindPipeline(cb,
VK_PIPELINE_BIND_POINT_COMPUTE,
@@ -444,10 +433,10 @@ hs_cmd_fill_in(VkCommandBuffer cb,
static
void
-hs_cmd_fill_out(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- uint32_t const from_slab,
- uint32_t const to_slab)
+hs_cmd_fill_out(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ uint32_t const from_slab,
+ uint32_t const to_slab)
{
vkCmdBindPipeline(cb,
VK_PIPELINE_BIND_POINT_COMPUTE,
@@ -465,7 +454,7 @@ hs_cmd_fill_out(VkCommandBuffer cb,
static
void
hs_cmd_bc(VkCommandBuffer cb,
- struct hs_vk const * const hs,
+ struct hotsort_vk const * const hs,
uint32_t const down_slabs,
uint32_t const clean_slabs_log2)
{
@@ -486,10 +475,10 @@ hs_cmd_bc(VkCommandBuffer cb,
static
uint32_t
-hs_cmd_hm(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- uint32_t const down_slabs,
- uint32_t const clean_slabs_log2)
+hs_cmd_hm(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ uint32_t const down_slabs,
+ uint32_t const clean_slabs_log2)
{
// how many scaled half-merge spans are there?
uint32_t const frac_ru = (1 << clean_slabs_log2) - 1;
@@ -518,11 +507,11 @@ hs_cmd_hm(VkCommandBuffer cb,
static
uint32_t
-hs_cmd_fm(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- uint32_t const bx_ru,
- uint32_t * const down_slabs,
- uint32_t const up_scale_log2)
+hs_cmd_fm(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ uint32_t const bx_ru,
+ uint32_t * const down_slabs,
+ uint32_t const up_scale_log2)
{
//
// FIXME OPTIMIZATION: in previous HotSort launchers it's sometimes
@@ -623,9 +612,9 @@ hs_cmd_fm(VkCommandBuffer cb,
static
void
-hs_cmd_bs(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- uint32_t const padded_in)
+hs_cmd_bs(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ uint32_t const padded_in)
{
uint32_t const slabs_in = padded_in / hs->slab_keys;
uint32_t const full_bs = slabs_in / hs->config.block.slabs;
@@ -660,20 +649,20 @@ hs_cmd_bs(VkCommandBuffer cb,
//
void
-hs_vk_sort(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- struct hs_vk_ds_offsets const * const offsets,
- uint32_t const count,
- uint32_t const padded_in,
- uint32_t const padded_out,
- bool const linearize)
+hotsort_vk_sort(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ struct hotsort_vk_ds_offsets const * const offsets,
+ uint32_t const count,
+ uint32_t const padded_in,
+ uint32_t const padded_out,
+ bool const linearize)
{
//
// append the push constants
//
size_t const kv_size = (hs->config.dwords.key + hs->config.dwords.val) * sizeof(uint32_t);
- struct hs_vk_push const push =
+ struct hotsort_vk_push const push =
{
.kv_offset_in = (uint32_t)(offsets->in / kv_size),
.kv_offset_out = (uint32_t)(offsets->out / kv_size),
@@ -682,9 +671,9 @@ hs_vk_sort(VkCommandBuffer cb,
vkCmdPushConstants(cb,
hs->pl,
- HS_VK_PUSH_CONSTANT_RANGE_STAGE_FLAGS,
- HS_VK_PUSH_CONSTANT_RANGE_OFFSET,
- HS_VK_PUSH_CONSTANT_RANGE_SIZE,
+ HOTSORT_VK_PUSH_CONSTANT_RANGE_STAGE_FLAGS,
+ HOTSORT_VK_PUSH_CONSTANT_RANGE_OFFSET,
+ HOTSORT_VK_PUSH_CONSTANT_RANGE_SIZE,
&push);
//
@@ -694,9 +683,10 @@ hs_vk_sort(VkCommandBuffer cb,
// - pre-sort fill needs to happen before bs()
// - pre-merge fill needs to happen before fm()
//
- uint32_t const padded_pre_sort = hs->is_in_place ? padded_out : padded_in;
+ bool const is_in_place = hs->config.is_in_place && (offsets->in == offsets->out);
+ uint32_t const padded_pre_sort = is_in_place ? padded_out : padded_in;
bool const is_pre_sort_reqd = padded_pre_sort > count;
- bool const is_pre_merge_reqd = !hs->is_in_place && (padded_out > padded_in);
+ bool const is_pre_merge_reqd = !is_in_place && (padded_out > padded_in);
//
// pre-sort fill?
diff --git a/src/graphics/lib/compute/hotsort/vk/hs_vk.h b/src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk.h
similarity index 57%
rename from src/graphics/lib/compute/hotsort/vk/hs_vk.h
rename to src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk.h
index 2d52b778dcdab02ae58a386103ca0a60822d6152..81c1955d0281524f09b0d8b647c0bdceaf6fdb22 100644
--- a/src/graphics/lib/compute/hotsort/vk/hs_vk.h
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk.h
@@ -22,7 +22,7 @@
//
//
-#include "hs_vk_target.h"
+#include "hotsort_vk_target.h"
//
// HotSort/VK relies on pipeline layout compatibility:
@@ -48,49 +48,29 @@
// The locations of the input and output buffers are declared at
// HotSort instance creation.
//
-// The buffer offsets can vary with each invocation of hs_vk_sort().
+// The buffer offsets can vary with each invocation of hotsort_vk_sort().
//
//
// HotSort push constants are expected at offset 0
//
-struct hs_vk_push
+struct hotsort_vk_push
{
uint32_t kv_offset_in;
uint32_t kv_offset_out;
uint32_t kv_count;
};
-#define HS_VK_PUSH_CONSTANT_RANGE_STAGE_FLAGS VK_SHADER_STAGE_COMPUTE_BIT
-#define HS_VK_PUSH_CONSTANT_RANGE_OFFSET 0
-#define HS_VK_PUSH_CONSTANT_RANGE_SIZE sizeof(struct hs_vk_push)
-
-//
-// Declare the descriptor set layout binding locations for the
-// input and output key-value storage buffers.
-//
-// If the input and output locations are identical then the sort will
-// be performed "in place".
-//
-
-struct hs_vk_ds_locations
-{
- struct {
- uint32_t set;
- uint32_t binding;
- } in;
- struct {
- uint32_t set;
- uint32_t binding;
- } out;
-};
+#define HOTSORT_VK_PUSH_CONSTANT_RANGE_STAGE_FLAGS VK_SHADER_STAGE_COMPUTE_BIT
+#define HOTSORT_VK_PUSH_CONSTANT_RANGE_OFFSET 0
+#define HOTSORT_VK_PUSH_CONSTANT_RANGE_SIZE sizeof(struct hotsort_vk_push)
//
// Declare the offsets of the key-value arrays before sorting.
//
-struct hs_vk_ds_offsets
+struct hotsort_vk_ds_offsets
{
VkDeviceSize in;
VkDeviceSize out;
@@ -101,13 +81,12 @@ struct hs_vk_ds_offsets
// buffers at specific descriptor set locations.
//
-struct hs_vk *
-hs_vk_create(VkDevice device,
- VkAllocationCallbacks const * allocator,
- VkPipelineCache pipeline_cache,
- VkPipelineLayout pipeline_layout,
- struct hs_vk_ds_locations const * const locations,
- struct hs_vk_target const * const target);
+struct hotsort_vk *
+hotsort_vk_create(VkDevice device,
+ VkAllocationCallbacks const * allocator,
+ VkPipelineCache pipeline_cache,
+ VkPipelineLayout pipeline_layout,
+ struct hotsort_vk_target const * const target);
//
// Resources will be disposed of with the same device and allocator
@@ -115,9 +94,9 @@ hs_vk_create(VkDevice device,
//
void
-hs_vk_release(VkDevice device,
- VkAllocationCallbacks const * const allocator,
- struct hs_vk * const hs);
+hotsort_vk_release(VkDevice device,
+ VkAllocationCallbacks const * const allocator,
+ struct hotsort_vk * const hs);
//
// Explicitly reveal what padding of maximum valued key-vals will be
@@ -138,10 +117,10 @@ hs_vk_release(VkDevice device,
//
void
-hs_vk_pad(struct hs_vk const * const hs,
- uint32_t const count,
- uint32_t * const padded_in,
- uint32_t * const padded_out);
+hotsort_vk_pad(struct hotsort_vk const * const hs,
+ uint32_t const count,
+ uint32_t * const padded_in,
+ uint32_t * const padded_out);
//
// Append commands to the command buffer that, when enqueued, will:
@@ -161,13 +140,13 @@ hs_vk_pad(struct hs_vk const * const hs,
//
void
-hs_vk_sort(VkCommandBuffer cb,
- struct hs_vk const * const hs,
- struct hs_vk_ds_offsets const * const offsets,
- uint32_t const count,
- uint32_t const padded_in,
- uint32_t const padded_out,
- bool const linearize);
+hotsort_vk_sort(VkCommandBuffer cb,
+ struct hotsort_vk const * const hs,
+ struct hotsort_vk_ds_offsets const * const offsets,
+ uint32_t const count,
+ uint32_t const padded_in,
+ uint32_t const padded_out,
+ bool const linearize);
//
//
diff --git a/src/graphics/lib/compute/hotsort/vk/hs_vk_target.h b/src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk_target.h
similarity index 83%
rename from src/graphics/lib/compute/hotsort/vk/hs_vk_target.h
rename to src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk_target.h
index 5a2ae5b64af6bc4092b1eba754f70b3104a55128..38457d79bdc272d98479d1a3cc1ab639fe36d3a3 100644
--- a/src/graphics/lib/compute/hotsort/vk/hs_vk_target.h
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/hotsort_vk_target.h
@@ -10,15 +10,15 @@
#include <stdint.h>
-#include "common/macros.h"
-
//
// This structure packages target-specific HotSort parameters and
// SPIR-V modules.
//
-struct hs_vk_target_config
+struct hotsort_vk_target_config
{
+ uint8_t is_in_place;
+
struct {
uint8_t threads_log2;
uint8_t width_log2;
@@ -56,10 +56,10 @@ struct hs_vk_target_config
// one SPIR-V module then reevaluate this encoding.
//
-struct hs_vk_target
+struct hotsort_vk_target
{
- struct hs_vk_target_config config;
- ALIGN_MACRO(4) uint8_t modules[]; // modules[] must start on 32-bit boundary
+ struct hotsort_vk_target_config config;
+ uint32_t modules[];
};
//
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/targets/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/targets/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..48ecdf80ca11d3006c1282da29ed95cbddd01534
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/BUILD.gn
@@ -0,0 +1,15 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+#
+# post-process SPIR-V modules and convert to C literals
+#
+
+if (current_toolchain == host_toolchain) {
+ executable("hotsort_modules_to_literals") {
+ sources = [
+ "hotsort_modules_to_literals.c"
+ ]
+ }
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_comp_names.py b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_comp_names.py
new file mode 100755
index 0000000000000000000000000000000000000000..be380cc5e07d0f8df7d8b3b08f012bf1fc3030b7
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_comp_names.py
@@ -0,0 +1,124 @@
+#!/usr/bin/python
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+#
+# The arguments passed to the 'hotsort_gen' code generator determine
+# the compute shaders that are emitted.
+#
+# GN requires the list of compute shaders that *will* be generated by
+# 'hotsort_gen'.
+#
+# This script accepts the exact same getopt() string as hotsort_gen
+# but only uses the relevant arguments to determine which shader names
+# will be produced by 'hotsort_gen'.
+#
+
+import getopt, sys
+
+#
+# Note that the inputs to these two functions are the number of
+# warps/waves/subgroups supported by a GPU block.
+#
+# Currently, the largest GPU only supports 2^6 warps per block.
+#
+
+def msb(n):
+ """Given an integer >= 0, return the most significant bit position."""
+ assert n < 2 ** 32
+ c = 0
+ while n > 0:
+ n >>= 1
+ c += 1
+ return c
+
+def pow2_ru(n):
+ """Given an integer >= 1, return the next power of 2 >= to n."""
+ assert n <= 2 ** 31
+ n -= 1
+ n |= n >> 1
+ n |= n >> 2
+ n |= n >> 4
+ n |= n >> 8
+ n |= n >> 16
+ n += 1
+ return n
+
+#
+# Duplicate the shader file generation logic of 'hotsort_gen' for a
+# glsl platform target.
+#
+
+bs_hi = 1
+fm_lo = 1
+fm_hi = 1
+hm_lo = 1
+hm_hi = 1
+
+try:
+ # option list must always match hotsort_gen/main.c
+ opts, args = getopt.getopt(sys.argv[1:],'hvzo:a:g:G:s:S:w:b:B:m:M:k:r:x:t:f:F:c:C:p:P:D:L:')
+except getopt.GetoptError as err:
+ print str(err)
+ sys.exit(2)
+
+for o, v in opts:
+ if o == '-b':
+ bs_hi = int(v)
+ elif o == '-f':
+ fm_lo = int(v)
+ elif o == '-F':
+ fm_hi = int(v)
+ elif o == '-c':
+ hm_lo = int(v)
+ elif o == '-C':
+ hm_hi = int(v)
+
+#
+# BS
+#
+
+bs_lo_log = 0
+
+while (1<<bs_lo_log) <= bs_hi:
+ print 'hs_bs_%d.comp' % bs_lo_log
+ bs_lo_log += 1
+
+#
+# BC
+#
+
+bc_lo_log = 0
+
+while bc_lo_log <= msb(bs_hi) - 1:
+ print 'hs_bc_%d.comp' % bc_lo_log
+ bc_lo_log += 1
+
+#
+# FM
+#
+
+for fm_scale in range(fm_lo,fm_hi+1):
+ span_left = (bs_hi << fm_scale) / 2
+ span_left_ru = pow2_ru(span_left)
+ span_right = 1
+ while (span_right <= span_left_ru):
+ fm_log = msb(pow2_ru(min(span_left,span_right))) - 1
+ print 'hs_fm_%d_%d.comp' % (fm_scale, fm_log)
+ span_right *= 2
+
+#
+# HM
+#
+
+for hm_scale in range(hm_lo,hm_hi+1):
+ print 'hs_hm_%d.comp' % hm_scale
+
+#
+# EXTRAS
+#
+
+print 'hs_fill_in.comp'
+print 'hs_fill_out.comp'
+print 'hs_transpose.comp'
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_modules_to_literals.c b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_modules_to_literals.c
new file mode 100644
index 0000000000000000000000000000000000000000..8d4dfc57051f4fb5be63ffb2bb3cf5d69c18f380
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_modules_to_literals.c
@@ -0,0 +1,146 @@
+// Copyright 2019 The Fuchsia Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include <inttypes.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+//
+// hotsort_modules_to_literals:
+//
+// Write a concatenated array of post-processed SPIR-V modules to a
+// file as an array of hex-encoded 32-bit C literals.
+//
+// Arguments:
+//
+// <output file> <spir-v module:1> ... <spir-v module M>
+//
+// Operation:
+//
+// 1. For each SPIR-V module:
+// 1. Open and find length
+// 2. Reallocate incrementing by length plus dword
+// 3. Store length
+// 4. Store module
+// 5. Close module
+// 2. Open and write out literals to output file
+// 3. Close output file
+//
+// SPIR-V modules are encoded with this layout:
+//
+// DWORD 0 - N : number of dwords in SPIR-V module
+// DWORD 1 - module[0]
+// DWORD N+1 - module[N-1]
+//
+
+#define HS_LITERALS_PER_LINE 6
+
+//
+//
+//
+
+int
+main(int argc, char const * argv[])
+{
+ // This tool will typically be passed ~20 files but if there isn't
+ // at least one file then fail.
+ if (argc < 3) {
+ return EXIT_FAILURE;
+ }
+
+ //
+ // layout buffer is reallocated for each module
+ //
+ uint32_t * layout = NULL;
+ size_t layout_size = 0;
+ uint32_t layout_next = 0;
+
+ //
+ // load and process all modules
+ //
+ uint32_t const module_count = argc - 2;
+
+ for (uint32_t ii=0; ii<module_count; ii++)
+ {
+ FILE * module = fopen(argv[2+ii],"rb");
+
+ if (module == NULL) {
+ return EXIT_FAILURE;
+ }
+
+ if (fseek(module,0L,SEEK_END) != 0) {
+ return EXIT_FAILURE;
+ }
+
+ long const module_bytes = ftell(module);
+
+ if (module_bytes == EOF) {
+ return EXIT_FAILURE;
+ }
+
+ rewind(module);
+
+ // "length + module size"
+ layout_size += sizeof(uint32_t) + module_bytes;
+
+ layout = realloc(layout,layout_size);
+
+ if (layout == NULL) {
+ return EXIT_FAILURE;
+ }
+
+ // store dwords
+ uint32_t const module_dwords = (uint32_t)(module_bytes / sizeof(uint32_t));
+
+ layout[layout_next++] = module_dwords;
+
+ // load module
+ if (fread(layout+layout_next,1,module_bytes,module) != (size_t)module_bytes) {
+ return EXIT_FAILURE;
+ }
+
+ // close module
+ if (fclose(module) != 0) {
+ return EXIT_FAILURE;
+ }
+
+ // move to next module
+ layout_next += module_dwords;
+ }
+
+ //
+ // store
+ //
+ FILE * file = fopen(argv[1],"wb");
+
+ if (file == NULL) {
+ return EXIT_FAILURE;
+ }
+
+ uint32_t literals = 0;
+
+ for (uint32_t ii=0; ii<layout_next; ii++)
+ {
+ fprintf(file,"0x%08X",layout[ii]);
+
+ if ((++literals % HS_LITERALS_PER_LINE) != 0) {
+ fprintf(file,", ");
+ } else {
+ fprintf(file,",\n");
+ }
+ }
+
+ fprintf(file,"\n");
+
+ if (ferror(file) || fclose(file) != 0) {
+ return EXIT_FAILURE;
+ }
+
+ return EXIT_SUCCESS;
+}
+
+//
+//
+//
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni
new file mode 100644
index 0000000000000000000000000000000000000000..653a7a4bd9d6a422273db84e334ad4902b310899
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni
@@ -0,0 +1,249 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//build/compiled_action.gni")
+
+#
+# Generates a HotSort target
+#
+template("hotsort_target") {
+ #
+ # Expects:
+ #
+ # $hotsort_target_vendor_dir:
+ #
+ # The directory of the vendor and arch-specific includes.
+ #
+ # $hotsort_target_name:
+ #
+ # A name that approximately conveys the configuration of the
+ # generated HotSort algorithm.
+ #
+ # $hotsort_target_args:
+ #
+ # The arguments passed to the 'hotsort_gen' HotSort algorithm code
+ # generator.
+ #
+ # $hotsort_target_dump:
+ #
+ # If defined the target is group that references a target binary
+ # produced by the host toolchain. Otherwise, a source set is
+ # produced on the current toolchain.
+ #
+ # The hotsort target name and args are passed to 'hotsort_gen' as
+ # follows:
+ #
+ # 'hotsort_gen -D $hotsort_target_name $hotsort_target_args'
+ #
+ # Validation of the args will be performed by 'hotsort_gen'.
+ #
+ # Note that $hotsort_target_name is split from the args list as it
+ # serves a dual role of providing the name of files as well as being
+ # a symbol in the GLSL source and C include files.
+ #
+ assert(defined(invoker.hotsort_target_vendor_dir),"vendor directory must be defined for hotsort_target")
+ assert(defined(invoker.hotsort_target_name), "name must be defined for hotsort_target")
+ assert(defined(invoker.hotsort_target_args), "args must be defined for hotsort_target")
+
+ #
+ # Either dump a binary or produce a source set
+ #
+ _hs_target_dump = false
+
+ if (defined(invoker.hotsort_target_dump)) {
+ _hs_target_dump = invoker.hotsort_target_dump
+ }
+
+ #
+ # define sources, includes and deps
+ #
+ _hs_target_gen_includes_public = [
+ "$target_gen_dir/hs_target.h"
+ ]
+
+ _hs_target_gen_includes = [
+ "$target_gen_dir/hs_config.h",
+ ] + _hs_target_gen_includes_public
+
+ _hs_target_includes = [
+ invoker.hotsort_target_vendor_dir,
+ "//src/graphics/lib/compute/hotsort/platforms/vk/targets"
+ ] + _hs_target_gen_includes
+
+ _hs_target_gen_sources = [
+ "$target_gen_dir/" + invoker.hotsort_target_name + ".c",
+ ]
+
+ _hs_target_gen_inlines = [
+ "$target_gen_dir/hs_modules.inl",
+ ]
+
+ _hs_target_include_dirs = [
+ "$target_gen_dir",
+ "//src/graphics/lib/compute/hotsort/platforms/vk/targets",
+ "//src/graphics/lib/compute/hotsort/platforms/vk",
+ ]
+
+ _hs_target_sources = _hs_target_includes + _hs_target_gen_sources + _hs_target_gen_inlines
+
+ #
+ # generated compute shaders
+ #
+ _hs_comp_names = exec_script("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_comp_names.py",
+ invoker.hotsort_target_args,
+ "list lines")
+ _hs_comp_sources = process_file_template(_hs_comp_names,"$target_gen_dir/comp/{{source_file_part}}")
+
+ #
+ # generate the .comp shaders
+ #
+ # Note that hs_modules.txt should match names returned by script
+ #
+ compiled_action("gen_comp") {
+ tool = "//src/graphics/lib/compute/hotsort/hotsort_gen"
+ outputs = _hs_comp_sources + _hs_target_gen_sources + _hs_target_gen_includes
+ args = [
+ "-D" , invoker.hotsort_target_name
+ ] + invoker.hotsort_target_args
+ }
+
+ #
+ # compile the .comp shaders to SPIR-V modules
+ #
+ compiled_action_foreach("gen_spv") {
+ tool = "//third_party/shaderc/third_party/glslang:glslangValidator"
+ sources = _hs_comp_sources
+ inputs = _hs_target_includes
+ outputs = [
+ "$target_gen_dir/spv/{{source_name_part}}.spv"
+ ]
+ args = [
+ "-s", # Note: '-s' silences glslangValidator printing filename
+ # but detailed error messages are also silenced
+ "--target-env", "vulkan1.1",
+ "-I" + rebase_path(target_gen_dir,root_build_dir), # <target>
+ "-I" + rebase_path(invoker.hotsort_target_vendor_dir,root_build_dir), # <vendor>
+ "-I" + rebase_path("//src/graphics/lib/compute/hotsort/platforms/vk/targets",root_build_dir), # <default>
+ "-o",
+ rebase_path(outputs[0],root_build_dir),
+ "{{source}}"
+ ]
+ public_deps = [
+ ":gen_comp"
+ ]
+ }
+
+ #
+ # optimize the modules
+ #
+ compiled_action_foreach("gen_opt") {
+ tool = "//third_party/shaderc/third_party/spirv-tools:spirv-opt"
+ sources = get_target_outputs(":gen_spv")
+ outputs = [
+ "$target_gen_dir/opt/{{source_name_part}}.spv"
+ ]
+ args = [
+ "-O",
+ "{{source}}",
+ "-o",
+ rebase_path(outputs[0],root_build_dir),
+ ]
+ public_deps = [
+ ":gen_spv"
+ ]
+ }
+
+ #
+ # remap the optimized modules
+ #
+ compiled_action_foreach("gen_remap") {
+ tool = "//third_party/shaderc/third_party/glslang:spirv-remap"
+ sources = get_target_outputs(":gen_opt")
+ outputs = [
+ "$target_gen_dir/remap/{{source_name_part}}.spv"
+ ]
+ args = [
+ "--do-everything",
+ "--input",
+ "{{source}}",
+ "--output",
+ rebase_path(target_gen_dir,root_build_dir) + "/remap"
+ ]
+ public_deps = [
+ ":gen_opt"
+ ]
+ }
+
+ #
+ # dump the modules as uint32_t literals
+ #
+ compiled_action("gen_modules") {
+ tool = "//src/graphics/lib/compute/hotsort/platforms/vk/targets:hotsort_modules_to_literals"
+ sources = get_target_outputs(":gen_remap")
+ outputs = _hs_target_gen_inlines
+ args = rebase_path(outputs,root_build_dir) + rebase_path(sources,root_build_dir)
+ public_deps = [
+ ":gen_remap"
+ ]
+ }
+
+ #
+ # either dump a binary or return a source set
+ #
+ if (_hs_target_dump) {
+ #
+ # executable for dumping a binary image of target
+ #
+ _hs_target_dump_name = "hotsort_dump_" + invoker.hotsort_target_name
+
+ executable(_hs_target_dump_name) {
+ defines = [ "HS_DUMP" ]
+ sources = _hs_target_gen_sources
+ include_dirs = _hs_target_include_dirs
+ public_deps = [
+ ":gen_modules"
+ ]
+ }
+
+ #
+ # dump a binary image of target
+ #
+ compiled_action("gen_bin") {
+ tool = ":$_hs_target_dump_name"
+ sources = _hs_target_sources
+ outputs = [
+ "$target_gen_dir/hs_target.bin"
+ ]
+ args = rebase_path(outputs,root_build_dir)
+ public_deps = [
+ ":$_hs_target_dump_name"
+ ]
+ }
+
+ #
+ # dummy group invokes $host_toolchain
+ #
+ group(target_name) {
+ public_deps = [
+ ":gen_bin($host_toolchain)"
+ ]
+ }
+ } else {
+ #
+ # target is a source set
+ #
+ source_set(target_name) {
+ public = [ "$target_gen_dir/hs_target.h" ]
+ sources = _hs_target_sources
+ include_dirs = _hs_target_include_dirs
+ public_deps = [
+ ":gen_modules",
+ ]
+ }
+ }
+}
+
+#
+#
+#
diff --git a/src/graphics/lib/compute/hotsort/vk/hs_glsl_macros.h b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_glsl_macros.h
similarity index 88%
rename from src/graphics/lib/compute/hotsort/vk/hs_glsl_macros.h
rename to src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_glsl_macros.h
index 75d6af5dd6dbcc6a38142c75772391fd7b44d630..9d1c95f8517d29e491002ff9bcf0840a8b6f6c38 100644
--- a/src/graphics/lib/compute/hotsort/vk/hs_glsl_macros.h
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_glsl_macros.h
@@ -5,6 +5,18 @@
#ifndef HS_GLSL_MACROS_ONCE
#define HS_GLSL_MACROS_ONCE
+//
+// require necessary extensions -- move this downward as soon as we
+// target GPUs with no support for GPUs
+//
+
+#extension GL_KHR_shader_subgroup_basic : require
+#extension GL_KHR_shader_subgroup_shuffle : require
+
+#if HS_KEY_DWORDS == 2
+#extension GL_ARB_gpu_shader_int64 : require
+#endif
+
//
// Define the type based on key and val sizes
//
@@ -14,7 +26,7 @@
#define HS_KEY_TYPE uint
#define HS_KEY_VAL_MAX HS_KEY_TYPE(-1)
#endif
-#elif HS_KEY_DWORDS == 2 // FIXME -- might want to use uint2
+#elif HS_KEY_DWORDS == 2 // FIXME -- some targets will use uint2
#if HS_VAL_DWORDS == 0
#define HS_KEY_TYPE uint64_t // GL_ARB_gpu_shader_int64
#define HS_KEY_VAL_MAX HS_KEY_TYPE(-1L)
@@ -32,9 +44,9 @@
//
//
-#define HS_GLSL_WORKGROUP_SIZE(_x,_y,_z) \
- layout( local_size_x = _x, \
- local_size_y = _y, \
+#define HS_GLSL_WORKGROUP_SIZE(_x,_y,_z) \
+ layout( local_size_x = _x, \
+ local_size_y = _y, \
local_size_z = _z) in
#define HS_GLSL_BINDING(_m,_n) \
@@ -52,8 +64,13 @@
// These can be overidden
//
-#define HS_KV_IN kv_in
-#define HS_KV_OUT kv_out
+#ifndef HS_KV_IN
+#define HS_KV_IN kv_in
+#endif
+
+#ifndef HS_KV_OUT
+#define HS_KV_OUT kv_out
+#endif
//
//
@@ -69,54 +86,54 @@
// KERNEL PROTOS
//
-#define HS_BS_KERNEL_PROTO(slab_count,slab_count_ru_log2) \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS*slab_count,1,1); \
- HS_GLSL_BINDING(0,0) writeonly buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
- HS_GLSL_BINDING(0,1) readonly buffer _kv_in { HS_KEY_TYPE HS_KV_IN[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_BS_KERNEL_PROTO(slab_count,slab_count_ru_log2) \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS*slab_count,1,1); \
+ HS_GLSL_BINDING(HS_KV_OUT_SET,HS_KV_OUT_BINDING) writeonly buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
+ HS_GLSL_BINDING(HS_KV_IN_SET,HS_KV_IN_BINDING) readonly buffer _kv_in { HS_KEY_TYPE HS_KV_IN[]; }; \
+ HS_GLSL_PUSH(); \
void main()
-#define HS_BC_KERNEL_PROTO(slab_count,slab_count_log2) \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS*slab_count,1,1); \
- HS_GLSL_BINDING(0,0) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_BC_KERNEL_PROTO(slab_count,slab_count_log2) \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS*slab_count,1,1); \
+ HS_GLSL_BINDING(HS_KV_OUT_SET,HS_KV_OUT_BINDING) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
+ HS_GLSL_PUSH(); \
void main()
-#define HS_FM_KERNEL_PROTO(s,r) \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
- HS_GLSL_BINDING(0,0) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_FM_KERNEL_PROTO(s,r) \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
+ HS_GLSL_BINDING(HS_KV_OUT_SET,HS_KV_OUT_BINDING) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
+ HS_GLSL_PUSH(); \
void main()
-#define HS_HM_KERNEL_PROTO(s) \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
- HS_GLSL_BINDING(0,0) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_HM_KERNEL_PROTO(s) \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
+ HS_GLSL_BINDING(HS_KV_OUT_SET,HS_KV_OUT_BINDING) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
+ HS_GLSL_PUSH(); \
void main()
-#define HS_FILL_IN_KERNEL_PROTO() \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
- HS_GLSL_BINDING(0,1) writeonly buffer _kv_in { HS_KEY_TYPE HS_KV_IN[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_FILL_IN_KERNEL_PROTO() \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
+ HS_GLSL_BINDING(HS_KV_IN_SET,HS_KV_IN_BINDING) writeonly buffer _kv_in { HS_KEY_TYPE HS_KV_IN[]; }; \
+ HS_GLSL_PUSH(); \
void main()
-#define HS_FILL_OUT_KERNEL_PROTO() \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
- HS_GLSL_BINDING(0,0) writeonly buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_FILL_OUT_KERNEL_PROTO() \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
+ HS_GLSL_BINDING(HS_KV_OUT_SET,HS_KV_OUT_BINDING) writeonly buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
+ HS_GLSL_PUSH(); \
void main()
-#define HS_TRANSPOSE_KERNEL_PROTO() \
- HS_GLSL_SUBGROUP_SIZE() \
- HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
- HS_GLSL_BINDING(0,0) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
- HS_GLSL_PUSH(); \
+#define HS_TRANSPOSE_KERNEL_PROTO() \
+ HS_GLSL_SUBGROUP_SIZE() \
+ HS_GLSL_WORKGROUP_SIZE(HS_SLAB_THREADS,1,1); \
+ HS_GLSL_BINDING(HS_KV_OUT_SET,HS_KV_OUT_BINDING) buffer _kv_out { HS_KEY_TYPE HS_KV_OUT[]; }; \
+ HS_GLSL_PUSH(); \
void main()
//
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_target_config_init.inl b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_target_config_init.inl
new file mode 100644
index 0000000000000000000000000000000000000000..c6ae1d866f2ef4431d77a01c0e9d93e2f185045b
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_target_config_init.inl
@@ -0,0 +1,35 @@
+// Copyright 2019 The Fuchsia Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+.is_in_place = HS_IS_IN_PLACE,
+
+.slab = {
+ .threads_log2 = HS_SLAB_THREADS_LOG2,
+ .width_log2 = HS_SLAB_WIDTH_LOG2,
+ .height = HS_SLAB_HEIGHT
+ },
+
+.dwords = {
+ .key = HS_KEY_DWORDS,
+ .val = HS_VAL_DWORDS
+ },
+
+.block = {
+ .slabs = HS_BS_SLABS
+ },
+
+.merge = {
+ .fm = {
+ .scale_min = HS_FM_SCALE_MIN,
+ .scale_max = HS_FM_SCALE_MAX
+ },
+ .hm = {
+ .scale_min = HS_HM_SCALE_MIN,
+ .scale_max = HS_HM_SCALE_MAX,
+ }
+ }
+
+//
+//
+//
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_target_modules_dump.inl b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_target_modules_dump.inl
new file mode 100644
index 0000000000000000000000000000000000000000..ad3b1227a26740f829c97ccbebc06b6aeef32a04
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/hs_target_modules_dump.inl
@@ -0,0 +1,58 @@
+// Copyright 2019 The Fuchsia Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifdef HS_DUMP
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <inttypes.h>
+
+//
+// Dump 'struct hotsort_vk_target HS_TARGET_NAME' to a file
+//
+
+int
+main(int argc, char const * argv[])
+{
+ FILE * file = fopen(argv[1],"wb");
+
+ if (file == NULL) {
+ return EXIT_FAILURE;
+ }
+
+ size_t const size_config = sizeof(HS_TARGET_NAME.config);
+
+ if (fwrite(&HS_TARGET_NAME.config,1,size_config,file) != size_config) {
+ return EXIT_FAILURE;
+ }
+
+ uint32_t const * modules = HS_TARGET_NAME.modules;
+ uint32_t dwords = modules[0];
+
+ while (dwords > 0)
+ {
+ dwords += 1;
+
+ size_t const size_module = dwords * sizeof(*modules);
+
+ if (fwrite(modules,1,size_module,file) != size_module) {
+ return EXIT_FAILURE;
+ }
+
+ modules += dwords;
+ dwords = modules[0];
+ }
+
+ if (fclose(file) != 0) {
+ return EXIT_FAILURE;
+ }
+
+ return EXIT_SUCCESS;
+}
+
+#endif
+
+//
+//
+//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/hs_glsl_macros.h b/src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/amd/hs_glsl_macros_vendor.h
similarity index 96%
rename from src/graphics/lib/compute/hotsort/vk/targets/amd/hs_glsl_macros.h
rename to src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/amd/hs_glsl_macros_vendor.h
index 8b6129fc116afb0550d2edbe41cc9c7b6650ed55..e6a9bfb4113999dae70f146522fcc1b8c9a395f8 100644
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/hs_glsl_macros.h
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/amd/hs_glsl_macros_vendor.h
@@ -9,7 +9,7 @@
//
//
-#include "../../hs_glsl_macros.h"
+#include "hs_glsl_macros.h"
//
// OVERRIDE SUBGROUP LANE ID
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/hs_glsl_macros.h b/src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/intel/hs_glsl_macros_vendor.h
similarity index 97%
rename from src/graphics/lib/compute/hotsort/vk/targets/intel/hs_glsl_macros.h
rename to src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/intel/hs_glsl_macros_vendor.h
index 59b74b96752de31c5de3a39299b3f1dacc63752f..ffc30b44b766fb20212632daf1661bd37a5f4cdf 100644
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/hs_glsl_macros.h
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/intel/hs_glsl_macros_vendor.h
@@ -9,7 +9,7 @@
//
//
-#include "../../hs_glsl_macros.h"
+#include "hs_glsl_macros.h"
//
// Waiting for Intel's to provide an equivalent to their OpenCL
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/hs_glsl_macros.h b/src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/nvidia/hs_glsl_macros_vendor.h
similarity index 96%
rename from src/graphics/lib/compute/hotsort/vk/targets/nvidia/hs_glsl_macros.h
rename to src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/nvidia/hs_glsl_macros_vendor.h
index 2872d9381ad67cb9a482a6ca25e5052aa84f2680..91d6b257adbd1f0586ced6f86c04bb4bbfd268e2 100644
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/hs_glsl_macros.h
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/nvidia/hs_glsl_macros_vendor.h
@@ -9,7 +9,7 @@
//
//
-#include "../../hs_glsl_macros.h"
+#include "hs_glsl_macros.h"
//
// OVERRIDE SUBGROUP LANE ID
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..8e604d8330b08467d0cee6d01aa31dc9f59ecfe0
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/BUILD.gn
@@ -0,0 +1,9 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+group("tests") {
+ public_deps = [
+ "hotsort_vk_bench"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..5d15a50c20a0ce1556d145c45454570a5e20f57c
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/BUILD.gn
@@ -0,0 +1,75 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//build/package.gni")
+import("//third_party/vulkan_loader_and_validation_layers/layers/layers.gni")
+
+#
+# Generate and build several HotSort targets
+#
+
+group("targets") {
+ public_deps = [
+ "targets/amd/gcn3/u32:generate",
+ "targets/amd/gcn3/u64:generate",
+ "targets/intel/gen8/u32:generate",
+ "targets/intel/gen8/u64:generate",
+ "targets/nvidia/sm35/u32:generate",
+ "targets/nvidia/sm35/u64:generate",
+ ]
+}
+
+#
+# hotsort_vk_bench: benchmark HotSort
+#
+
+package_name = "hotsort_vk_bench"
+
+executable("bin") {
+ output_name = package_name
+ sources = [
+ "main.c",
+ "sort.cpp"
+ ]
+ include_dirs = [
+ "$target_gen_dir",
+ "//src/graphics/lib/compute",
+ "//src/graphics/lib/compute/hotsort/platforms/vk"
+ ]
+ deps = [
+ ":targets",
+ "//src/graphics/lib/compute/common",
+ "//src/graphics/lib/compute/common/vk",
+ "//src/graphics/lib/compute/hotsort/platforms/vk",
+ "//third_party/vulkan_loader_and_validation_layers:vulkan"
+ ]
+}
+
+#
+#
+#
+
+package(package_name) {
+ deps = [
+ ":bin",
+ "//third_party/vulkan_loader_and_validation_layers/layers"
+ ]
+
+ binary = package_name
+
+ meta = [
+ {
+ path = rebase_path("meta/$package_name.cmx")
+ dest = "$package_name.cmx"
+ },
+ ]
+
+ public_deps = vulkan_validation_layers.public_deps
+ loadable_modules = vulkan_validation_layers.loadable_modules
+ resources = vulkan_validation_layers.resources
+}
+
+#
+#
+#
diff --git a/src/graphics/lib/compute/hotsort/vk/bench/main.c b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/main.c
similarity index 92%
rename from src/graphics/lib/compute/hotsort/vk/bench/main.c
rename to src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/main.c
index e0dcf3b1c0fbc0ba4c67336fca12160ebd2051fd..eee52a6862ae9c6bdf20d6f85f472675efa005b5 100644
--- a/src/graphics/lib/compute/hotsort/vk/bench/main.c
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/main.c
@@ -24,20 +24,20 @@
//
//
-#include "hs_vk.h"
+#include "hotsort_vk.h"
//
// Compile-time images of HotSort targets
//
-#include "hotsort/vk/targets/intel/gen8/u32/hs_target.h"
-#include "hotsort/vk/targets/intel/gen8/u64/hs_target.h"
+#include "targets/intel/gen8/u32/hs_target.h"
+#include "targets/intel/gen8/u64/hs_target.h"
-#include "hotsort/vk/targets/nvidia/sm_35/u32/hs_target.h"
-#include "hotsort/vk/targets/nvidia/sm_35/u64/hs_target.h"
+#include "targets/nvidia/sm35/u32/hs_target.h"
+#include "targets/nvidia/sm35/u64/hs_target.h"
-#include "hotsort/vk/targets/amd/gcn3/u32/hs_target.h"
-#include "hotsort/vk/targets/amd/gcn3/u64/hs_target.h"
+#include "targets/amd/gcn3/u32/hs_target.h"
+#include "targets/amd/gcn3/u64/hs_target.h"
//
//
@@ -73,7 +73,7 @@ hs_transpose_slabs_u32(uint32_t const hs_words,
{
uint32_t const slab_keys = hs_width * hs_height;
size_t const slab_size = sizeof(uint32_t) * hs_words * slab_keys;
- uint32_t * const slab = ALLOCA_MACRO(slab_size);
+ uint32_t * const slab = malloc(slab_size);
uint32_t slab_count = count / slab_keys;
while (slab_count-- > 0)
@@ -86,6 +86,8 @@ hs_transpose_slabs_u32(uint32_t const hs_words,
vout_h += slab_keys;
}
+
+ free(slab);
}
static
@@ -98,7 +100,7 @@ hs_transpose_slabs_u64(uint32_t const hs_words,
{
uint32_t const slab_keys = hs_width * hs_height;
size_t const slab_size = sizeof(uint32_t) * hs_words * slab_keys;
- uint64_t * const slab = ALLOCA_MACRO(slab_size);
+ uint64_t * const slab = malloc(slab_size);
uint32_t slab_count = count / slab_keys;
while (slab_count-- > 0)
@@ -111,6 +113,8 @@ hs_transpose_slabs_u64(uint32_t const hs_words,
vout_h += slab_keys;
}
+
+ free(slab);
}
static
@@ -165,7 +169,7 @@ vk_debug_report_cb(VkDebugReportFlagsEXT flags,
if (is_error)
{
- fprintf(stderr,"%s %s %s\n",
+ fprintf(stderr,"%-43s - %-12s - %s\n",
flag_str,
pLayerPrefix,
pMessage);
@@ -219,6 +223,8 @@ hs_fill_rand(uint32_t * vin_h, uint32_t const count, uint32_t const words)
//
//
+#if !defined(NDEBUG) && defined(HS_VK_DUMP_SLABS)
+
static
void
hs_debug_u32(uint32_t const hs_width,
@@ -259,16 +265,18 @@ hs_debug_u64(uint32_t const hs_width,
}
}
+#endif
+
//
//
//
bool
-is_matching_device(VkPhysicalDeviceProperties const * const phy_device_props,
- struct hs_vk_target const * * const hs_target,
- uint32_t const vendor_id,
- uint32_t const device_id,
- uint32_t const key_val_words)
+is_matching_device(VkPhysicalDeviceProperties const * const phy_device_props,
+ struct hotsort_vk_target const * * const hs_target,
+ uint32_t const vendor_id,
+ uint32_t const device_id,
+ uint32_t const key_val_words)
{
if ((phy_device_props->vendorID != vendor_id) || (phy_device_props->deviceID != device_id))
return false;
@@ -305,9 +313,9 @@ is_matching_device(VkPhysicalDeviceProperties const * const phy_device_props,
// AMD GCN
//
if (key_val_words == 1)
- *hs_target = &hs_amd_gcn_u32;
+ *hs_target = &hs_amd_gcn3_u32;
else
- *hs_target = &hs_amd_gcn_u64;
+ *hs_target = &hs_amd_gcn3_u64;
}
else
{
@@ -353,13 +361,13 @@ main(int argc, char const * argv[])
// create a Vulkan instances
//
VkApplicationInfo const app_info = {
- .sType = VK_STRUCTURE_TYPE_APPLICATION_INFO,
- .pNext = NULL,
- .pApplicationName = "HotSort Bench",
- .applicationVersion = 0,
- .pEngineName = "HotSort",
- .engineVersion = 0,
- .apiVersion = VK_API_VERSION_1_1
+ .sType = VK_STRUCTURE_TYPE_APPLICATION_INFO,
+ .pNext = NULL,
+ .pApplicationName = "HotSort Bench",
+ .applicationVersion = 0,
+ .pEngineName = "HotSort",
+ .engineVersion = 0,
+ .apiVersion = VK_API_VERSION_1_1
};
char const * const instance_enabled_layers[] = {
@@ -447,7 +455,7 @@ main(int argc, char const * argv[])
VkPhysicalDevice phy_device = VK_NULL_HANDLE;
VkPhysicalDeviceProperties phy_device_props;
- struct hs_vk_target const * hs_target;
+ struct hotsort_vk_target const * hs_target;
for (uint32_t ii=0; ii<phy_device_count; ii++)
{
@@ -532,7 +540,7 @@ main(int argc, char const * argv[])
// clumsily enable AMD GCN shader info extension
//
char const * const device_enabled_extensions[] = {
-#if defined( HS_VK_VERBOSE_STATISTICS_AMD ) || defined( HS_VK_VERBOSE_DISASSEMBLY_AMD )
+#if defined( HOTSORT_VK_VERBOSE_STATISTICS_AMD ) || defined( HOTSORT_VK_VERBOSE_DISASSEMBLY_AMD )
VK_AMD_SHADER_INFO_EXTENSION_NAME
#else
NULL
@@ -541,7 +549,7 @@ main(int argc, char const * argv[])
uint32_t device_enabled_extension_count = 0;
-#if defined( HS_VK_VERBOSE_STATISTICS_AMD ) || defined( HS_VK_VERBOSE_DISASSEMBLY_AMD )
+#if defined( HOTSORT_VK_VERBOSE_STATISTICS_AMD ) || defined( HOTSORT_VK_VERBOSE_DISASSEMBLY_AMD )
if (phy_device_props.vendorID == 0x1002)
device_enabled_extension_count = 1;
#endif
@@ -658,9 +666,9 @@ main(int argc, char const * argv[])
.pushConstantRangeCount = 1,
.pPushConstantRanges = (VkPushConstantRange[]){
{
- .stageFlags = HS_VK_PUSH_CONSTANT_RANGE_STAGE_FLAGS,
- .offset = HS_VK_PUSH_CONSTANT_RANGE_OFFSET,
- .size = HS_VK_PUSH_CONSTANT_RANGE_SIZE
+ .stageFlags = HOTSORT_VK_PUSH_CONSTANT_RANGE_STAGE_FLAGS,
+ .offset = HOTSORT_VK_PUSH_CONSTANT_RANGE_OFFSET,
+ .size = HOTSORT_VK_PUSH_CONSTANT_RANGE_SIZE
}
}
};
@@ -687,27 +695,11 @@ main(int argc, char const * argv[])
vk(AllocateDescriptorSets(device,&dsai,&ds));
- //
- // create HotSort instance
- //
- struct hs_vk_ds_locations const ds_locations =
- {
- .in = {
- .set = 0,
- .binding = 1
- },
- .out = {
- .set = 0,
- .binding = 0
- }
- };
-
- struct hs_vk * const hs = hs_vk_create(device,
- NULL,
- pc,
- pl,
- &ds_locations,
- hs_target);
+ struct hotsort_vk * const hs = hotsort_vk_create(device,
+ NULL,
+ pc,
+ pl,
+ hs_target);
//
// create a command pool for this thread
//
@@ -749,7 +741,7 @@ main(int argc, char const * argv[])
//
uint32_t buffer_in_count, buffer_out_count;
- hs_vk_pad(hs,count_hi,&buffer_in_count,&buffer_out_count);
+ hotsort_vk_pad(hs,count_hi,&buffer_in_count,&buffer_out_count);
size_t const buffer_out_size = buffer_out_count * key_val_words * sizeof(uint32_t);
@@ -770,10 +762,10 @@ main(int argc, char const * argv[])
VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
VK_BUFFER_USAGE_TRANSFER_DST_BIT,
- vk(CreateBuffer(device,
- &bci,
- NULL,
- &vin));
+ vk(CreateBuffer(device,
+ &bci,
+ NULL,
+ &vin));
vk(CreateBuffer(device,
&bci,
@@ -962,7 +954,7 @@ main(int argc, char const * argv[])
//
// buffer offsets
//
- struct hs_vk_ds_offsets const offsets =
+ struct hotsort_vk_ds_offsets const offsets =
{
.in = 0UL,
.out = 0UL
@@ -992,6 +984,11 @@ main(int argc, char const * argv[])
"Avg. Mkeys/s, "
"Max. Mkeys/s\n");
+ //
+ // accumulate verifications
+ //
+ bool all_verified = true;
+
//
// test a range
//
@@ -1002,7 +999,7 @@ main(int argc, char const * argv[])
//
uint32_t count_padded_in, count_padded_out;
- hs_vk_pad(hs,count,&count_padded_in,&count_padded_out);
+ hotsort_vk_pad(hs,count,&count_padded_in,&count_padded_out);
//
// initialize vin with 'count' random keys
@@ -1057,13 +1054,13 @@ main(int argc, char const * argv[])
//
// append sorting commands
//
- hs_vk_sort(cb,
- hs,
- &offsets,
- count,
- count_padded_in,
- count_padded_out,
- linearize);
+ hotsort_vk_sort(cb,
+ hs,
+ &offsets,
+ count,
+ count_padded_in,
+ count_padded_out,
+ linearize);
//
// end timestamp
@@ -1124,7 +1121,7 @@ main(int argc, char const * argv[])
//
char const * cpu_algo = NULL;
double cpu_ns = 0.0;
- bool verified = false;
+ bool verified = true;
if (verify)
{
@@ -1178,7 +1175,7 @@ main(int argc, char const * argv[])
// verify
verified = memcmp(sorted_h,sorted_map,size_padded_in) == 0;
-#ifndef NDEBUG
+#if !defined(NDEBUG) && defined(HS_VK_DUMP_SLABS)
if (!verified)
{
if (key_val_words == 1)
@@ -1211,6 +1208,11 @@ main(int argc, char const * argv[])
vkUnmapMemory(device,mem_sorted);
}
+ //
+ // any verification failures?
+ //
+ all_verified = all_verified && verified;
+
//
// REPORT
//
@@ -1254,7 +1256,7 @@ main(int argc, char const * argv[])
vkDestroyPipelineLayout(device,pl,NULL);
// release shared HotSort state
- hs_vk_release(device,NULL,hs);
+ hotsort_vk_release(device,NULL,hs);
// destroy the vin/vout buffers (before device memory)
vkDestroyBuffer(device,vin, NULL);
@@ -1287,7 +1289,7 @@ main(int argc, char const * argv[])
vkDestroyInstance(instance,NULL);
- return EXIT_SUCCESS;
+ return all_verified ? EXIT_SUCCESS : EXIT_FAILURE;
}
//
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/meta/hotsort_vk_bench.cmx b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/meta/hotsort_vk_bench.cmx
new file mode 100644
index 0000000000000000000000000000000000000000..1c51dfe9310ab622e5b0a47e22531a541315c1fd
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/meta/hotsort_vk_bench.cmx
@@ -0,0 +1,15 @@
+{
+ "program": {
+ "binary": "bin/app"
+ },
+ "sandbox": {
+ "features": [
+ "vulkan"
+ ],
+ "services": [
+ "fuchsia.sysmem.Allocator",
+ "fuchsia.vulkan.loader.Loader",
+ "fuchsia.tracelink.Registry"
+ ]
+ }
+}
diff --git a/src/graphics/lib/compute/hotsort/vk/bench/sort.cpp b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/sort.cpp
similarity index 91%
rename from src/graphics/lib/compute/hotsort/vk/bench/sort.cpp
rename to src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/sort.cpp
index fb438d390930051d4d5046e81e508d6926a8fd98..70b1b55865f49feb1a648b9220df5c2305527f51 100644
--- a/src/graphics/lib/compute/hotsort/vk/bench/sort.cpp
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/sort.cpp
@@ -30,7 +30,12 @@
//
//
-#if (__cplusplus >= 201703L) && !defined(HS_USE_STD_SORT) && !defined(HS_USE_QSORT)
+#if defined(__cpp_lib_execution) \
+ && (__cpp_lib_execution >= 201603L) \
+ && defined(__cpp_lib_parallel_algorithm) \
+ && (__cpp_lib_parallel_algorithm >= 201603L) \
+ && !defined(HS_USE_STD_SORT) \
+ && !defined(HS_USE_QSORT)
#define HS_USE_PARALLEL_SORT
#include <algorithm>
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/amd/gcn3/u32/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/amd/gcn3/u32/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..3fc827d841fe6debc5324359d69fb20441eea1a3
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/amd/gcn3/u32/BUILD.gn
@@ -0,0 +1,36 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni")
+
+#
+# configuration
+#
+# $HS_GEN -v -a "glsl" -D HS_AMD_GCN3 -t 1 -w 64 -r 16 -s 32768 -S 32768 -b 16 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
+#
+
+hotsort_target("generate")
+{
+ hotsort_target_vendor_dir = "//src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/amd"
+ hotsort_target_name = "hs_amd_gcn3_u32"
+ hotsort_target_args = [
+ # "-v",
+ "-o" , rebase_path(target_gen_dir,root_build_dir),
+ "-a" , "glsl",
+ "-t" , "1",
+ "-w" , "64",
+ "-r" , "16",
+ "-s" , "32768",
+ "-S" , "32768",
+ "-b" , "16",
+ "-m" , "1",
+ "-M" , "1",
+ "-f" , "1",
+ "-F" , "1",
+ "-c" , "1",
+ "-C" , "1",
+ "-L" , "0,1,0,0",
+ "-z"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/amd/gcn3/u64/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/amd/gcn3/u64/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..2484994f776d7016e5d6ea64cf81d59cae2360df
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/amd/gcn3/u64/BUILD.gn
@@ -0,0 +1,36 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni")
+
+#
+# configuration
+#
+# $HS_GEN -v -a "glsl" -D HS_AMD_GCN3 -t 2 -w 64 -r 8 -s 32768 -S 32768 -b 16 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
+#
+
+hotsort_target("generate")
+{
+ hotsort_target_vendor_dir = "//src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/amd"
+ hotsort_target_name = "hs_amd_gcn3_u64"
+ hotsort_target_args = [
+ # "-v",
+ "-o" , rebase_path(target_gen_dir,root_build_dir),
+ "-a" , "glsl",
+ "-t" , "2",
+ "-w" , "64",
+ "-r" , "8",
+ "-s" , "32768",
+ "-S" , "32768",
+ "-b" , "16",
+ "-m" , "1",
+ "-M" , "1",
+ "-f" , "1",
+ "-F" , "1",
+ "-c" , "1",
+ "-C" , "1",
+ "-L" , "0,1,0,0",
+ "-z"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/intel/gen8/u32/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/intel/gen8/u32/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..f3c179d57f7375ba2905db4bd19925c5ed1cb203
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/intel/gen8/u32/BUILD.gn
@@ -0,0 +1,37 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni")
+
+#
+# configuration
+#
+# $HS_GEN -v -a "glsl" -D HS_INTEL_GEN8 -t 1 -w 16 -r 8 -s 21504 -S 65536 -b 16 -B 48 -m 1 -M 1 -f 0 -F 0 -c 0 -C 0 -z
+#
+
+hotsort_target("generate")
+{
+ hotsort_target_vendor_dir = "//src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/intel"
+ hotsort_target_name = "hs_intel_gen8_u32"
+ hotsort_target_args = [
+ # "-v",
+ "-o" , rebase_path(target_gen_dir,root_build_dir),
+ "-a" , "glsl",
+ "-t" , "1",
+ "-w" , "16",
+ "-r" , "8",
+ "-s" , "21504",
+ "-S" , "65536",
+ "-b" , "16",
+ "-B" , "48",
+ "-m" , "1",
+ "-M" , "1",
+ "-f" , "0",
+ "-F" , "0",
+ "-c" , "0",
+ "-C" , "0",
+ "-L" , "0,1,0,0",
+ "-z"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/intel/gen8/u64/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/intel/gen8/u64/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..47e98f188f86b4ec4f7054b60fe58e7b0c902be3
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/intel/gen8/u64/BUILD.gn
@@ -0,0 +1,37 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni")
+
+#
+# configuration
+#
+# $HS_GEN -v -a "glsl" -D HS_INTEL_GEN8 -t 2 -w 8 -r 16 -s 21504 -S 65536 -b 16 -B 48 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
+#
+
+hotsort_target("generate")
+{
+ hotsort_target_vendor_dir = "//src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/intel"
+ hotsort_target_name = "hs_intel_gen8_u64"
+ hotsort_target_args = [
+ # "-v",
+ "-o" , rebase_path(target_gen_dir,root_build_dir),
+ "-a" , "glsl",
+ "-t" , "2",
+ "-w" , "8",
+ "-r" , "16",
+ "-s" , "21504",
+ "-S" , "65536",
+ "-b" , "16",
+ "-B" , "48",
+ "-m" , "1",
+ "-M" , "1",
+ "-f" , "1",
+ "-F" , "1",
+ "-c" , "1",
+ "-C" , "1",
+ "-L" , "0,1,0,0",
+ "-z"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/nvidia/sm35/u32/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/nvidia/sm35/u32/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..9c7bab9128febf861796dcb12815d73a906483ef
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/nvidia/sm35/u32/BUILD.gn
@@ -0,0 +1,39 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni")
+
+#
+# configuration
+#
+# OLD: $HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 1 -w 32 -r 32 -s 49152 -S 65536 -b 32 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
+# NEW: $HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 1 -w 32 -r 16 -s 32768 -S 32768 -b 16 -m 1 -M 1 -p 1 -P 1 -f 0 -F 0 -c 0 -C 0 -z
+#
+
+hotsort_target("generate")
+{
+ hotsort_target_vendor_dir = "//src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/nvidia"
+ hotsort_target_name = "hs_nvidia_sm35_u32"
+ hotsort_target_args = [
+ # "-v",
+ "-o" , rebase_path(target_gen_dir,root_build_dir),
+ "-a" , "glsl",
+ "-t" , "1",
+ "-w" , "32",
+ "-r" , "16",
+ "-s" , "32768",
+ "-S" , "32768",
+ "-b" , "16",
+ "-m" , "1",
+ "-M" , "1",
+ "-p" , "1",
+ "-P" , "1",
+ "-f" , "0",
+ "-F" , "0",
+ "-c" , "0",
+ "-C" , "0",
+ "-L" , "0,1,0,0",
+ "-z"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/nvidia/sm35/u64/BUILD.gn b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/nvidia/sm35/u64/BUILD.gn
new file mode 100644
index 0000000000000000000000000000000000000000..40d54223a8f4b2fc137c025c03bb1d3e389743a5
--- /dev/null
+++ b/src/graphics/lib/compute/hotsort/platforms/vk/tests/hotsort_vk_bench/targets/nvidia/sm35/u64/BUILD.gn
@@ -0,0 +1,39 @@
+# Copyright 2019 The Fuchsia Authors. All rights reserved.
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+import("//src/graphics/lib/compute/hotsort/platforms/vk/targets/hotsort_target.gni")
+
+#
+# configuration
+#
+# OLD: $HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 2 -w 32 -r 32 -s 49152 -S 65536 -b 16 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
+# NEW: $HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 2 -w 32 -r 8 -s 32768 -S 32768 -b 16 -m 1 -M 1 -p 1 -P 1 -f 0 -F 0 -c 0 -C 0 -z
+#
+
+hotsort_target("generate")
+{
+ hotsort_target_vendor_dir = "//src/graphics/lib/compute/hotsort/platforms/vk/targets/vendors/nvidia"
+ hotsort_target_name = "hs_nvidia_sm35_u64"
+ hotsort_target_args = [
+ # "-v",
+ "-o" , rebase_path(target_gen_dir,root_build_dir),
+ "-a" , "glsl",
+ "-t" , "2",
+ "-w" , "32",
+ "-r" , "8",
+ "-s" , "32768",
+ "-S" , "32768",
+ "-b" , "16",
+ "-m" , "1",
+ "-M" , "1",
+ "-p" , "1",
+ "-P" , "1",
+ "-f" , "0",
+ "-F" , "0",
+ "-c" , "0",
+ "-C" , "0",
+ "-L" , "0,1,0,0",
+ "-z"
+ ]
+}
diff --git a/src/graphics/lib/compute/hotsort/vk/hs_glsl_preamble.h b/src/graphics/lib/compute/hotsort/vk/hs_glsl_preamble.h
deleted file mode 100644
index fa55d76fd7309c5f7a2b1858d8cb6a2f1a0bedc1..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/hs_glsl_preamble.h
+++ /dev/null
@@ -1,43 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_PREAMBLE_ONCE
-#define HS_GLSL_PREAMBLE_ONCE
-
-//
-//
-//
-
-#define HS_EVAL(a) a
-#define HS_HASH_SYMBOL #
-#define HS_HASH() HS_EVAL(HS_HASH_SYMBOL)
-#define HS_GLSL_EXT() HS_HASH()extension // will be indented one space
-#define HS_GLSL_EXT_REQUIRE(name) HS_GLSL_EXT() name : require
-#define HS_GLSL_VERSION(ver) HS_HASH()version ver // will be indented one space
-
-//
-//
-//
-
-HS_GLSL_VERSION(460)
-HS_GLSL_EXT_REQUIRE(GL_KHR_shader_subgroup_basic)
-HS_GLSL_EXT_REQUIRE(GL_KHR_shader_subgroup_shuffle)
-
-//
-//
-//
-
-#if HS_KEY_DWORDS == 2
-HS_GLSL_EXT_REQUIRE(GL_ARB_gpu_shader_int64)
-#endif
-
-//
-//
-//
-
-#endif
-
-//
-//
-//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/gen.sh b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/gen.sh
deleted file mode 100644
index 8e5c4f957a513a328d003a2ee7e0d3ad4950a609..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/gen.sh
+++ /dev/null
@@ -1,80 +0,0 @@
-#!/bin/bash
-
-##
-## Copyright 2019 The Fuchsia Authors. All rights reserved.
-## Use of this source code is governed by a BSD-style license that can be
-## found in the LICENSE file.
-##
-
-##
-## exit on error
-##
-
-## set -e
-
-##
-## delete the previous images
-##
-
-rm *.comp
-rm *.spv
-rm *.xxd
-
-##
-##
-##
-
-HS_GEN=../../../../../gen/hs_gen
-
-## --- 32-bit keys ---
-
-$HS_GEN -v -a "glsl" -D HS_AMD_GCN -t 1 -w 64 -r 16 -s 32768 -S 32768 -b 16 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
-
-##
-## remove trailing whitespace from generated files
-##
-
-sed -i 's/[[:space:]]*$//' hs_config.h
-sed -i 's/[[:space:]]*$//' hs_modules.h
-
-##
-##
-##
-
-whereis glslangValidator
-
-##
-## FIXME -- convert this to a bash script
-##
-## Note that we can use xargs instead of the cmd for/do
-##
-
-for f in *.comp
-do
- dos2unix $f
- clang-format -style=Mozilla -i $f
- cpp -P -I ../.. -I ../../../.. $f > ${f%%.*}.pre.comp
- clang-format -style=Mozilla -i ${f%%.*}.pre.comp
- glslangValidator --target-env vulkan1.1 -o ${f%%.*}.spv ${f%%.*}.pre.comp
- spirv-opt -O ${f%%.*}.spv -o ${f%%.*}.spv
-## spirv-remap -v --do-everything --input %%~nf.spv --output remap
- xxd -i < ${f%%.*}.spv > ${f%%.*}.spv.xxd
- len=$(wc -c < ${f%%.*}.spv)
- echo ${f%%.*}.spv $len
- printf "%.8x" $len | xxd -r -p | xxd -i > ${f%%.*}.len.xxd
-done
-
-##
-## dump a binary
-##
-
-cc -I ../../../.. -I ../../../../../.. -D=HS_DUMP -o hs_dump *.c
-hs_dump
-
-##
-## delete temporary files
-##
-
-rm *.pre.comp
-rm *.comp
-rm *.spv
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_config.h b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_config.h
deleted file mode 100644
index 57ffbe2e55a0950889c9ebb67e7cd29630bd08c2..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_config.h
+++ /dev/null
@@ -1,127 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_ONCE
-#define HS_GLSL_ONCE
-
-#define HS_SLAB_THREADS_LOG2 6
-#define HS_SLAB_THREADS (1 << HS_SLAB_THREADS_LOG2)
-#define HS_SLAB_WIDTH_LOG2 6
-#define HS_SLAB_WIDTH (1 << HS_SLAB_WIDTH_LOG2)
-#define HS_SLAB_HEIGHT 16
-#define HS_SLAB_KEYS (HS_SLAB_WIDTH * HS_SLAB_HEIGHT)
-#define HS_REG_LAST(c) c##16
-#define HS_KEY_DWORDS 1
-#define HS_VAL_DWORDS 0
-#define HS_BS_SLABS 16
-#define HS_BS_SLABS_LOG2_RU 4
-#define HS_BC_SLABS_LOG2_MAX 4
-#define HS_FM_BLOCK_HEIGHT 1
-#define HS_FM_SCALE_MIN 1
-#define HS_FM_SCALE_MAX 1
-#define HS_HM_BLOCK_HEIGHT 1
-#define HS_HM_SCALE_MIN 1
-#define HS_HM_SCALE_MAX 1
-#define HS_EMPTY
-
-#define HS_AMD_GCN
-
-#define HS_SLAB_ROWS() \
- HS_SLAB_ROW( 1, 0 ) \
- HS_SLAB_ROW( 2, 1 ) \
- HS_SLAB_ROW( 3, 2 ) \
- HS_SLAB_ROW( 4, 3 ) \
- HS_SLAB_ROW( 5, 4 ) \
- HS_SLAB_ROW( 6, 5 ) \
- HS_SLAB_ROW( 7, 6 ) \
- HS_SLAB_ROW( 8, 7 ) \
- HS_SLAB_ROW( 9, 8 ) \
- HS_SLAB_ROW( 10, 9 ) \
- HS_SLAB_ROW( 11, 10 ) \
- HS_SLAB_ROW( 12, 11 ) \
- HS_SLAB_ROW( 13, 12 ) \
- HS_SLAB_ROW( 14, 13 ) \
- HS_SLAB_ROW( 15, 14 ) \
- HS_SLAB_ROW( 16, 15 ) \
- HS_EMPTY
-
-#define HS_TRANSPOSE_SLAB() \
- HS_TRANSPOSE_STAGE( 1 ) \
- HS_TRANSPOSE_STAGE( 2 ) \
- HS_TRANSPOSE_STAGE( 3 ) \
- HS_TRANSPOSE_STAGE( 4 ) \
- HS_TRANSPOSE_STAGE( 5 ) \
- HS_TRANSPOSE_STAGE( 6 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 2, 1 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 4, 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 6, 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 8, 7 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 10, 9 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 12, 11 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 14, 13 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 16, 15 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 3, 1 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 4, 2 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 7, 5 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 8, 6 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 11, 9 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 12, 10 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 15, 13 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 16, 14 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 5, 1 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 6, 2 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 7, 3 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 8, 4 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 13, 9 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 14, 10 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 15, 11 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 16, 12 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 9, 1 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 10, 2 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 11, 3 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 12, 4 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 13, 5 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 14, 6 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 15, 7 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 16, 8 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 2, 1 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 4, 3 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 6, 5 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 8, 7 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 10, 9 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 12, 11 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 14, 13 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 16, 15 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 3, 1 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 4, 2 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 7, 5 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 8, 6 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 11, 9 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 12, 10 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 15, 13 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 16, 14 ) \
- HS_TRANSPOSE_REMAP( x, 1, 1 ) \
- HS_TRANSPOSE_REMAP( x, 2, 5 ) \
- HS_TRANSPOSE_REMAP( x, 3, 9 ) \
- HS_TRANSPOSE_REMAP( x, 4, 13 ) \
- HS_TRANSPOSE_REMAP( x, 5, 2 ) \
- HS_TRANSPOSE_REMAP( x, 6, 6 ) \
- HS_TRANSPOSE_REMAP( x, 7, 10 ) \
- HS_TRANSPOSE_REMAP( x, 8, 14 ) \
- HS_TRANSPOSE_REMAP( x, 9, 3 ) \
- HS_TRANSPOSE_REMAP( x, 10, 7 ) \
- HS_TRANSPOSE_REMAP( x, 11, 11 ) \
- HS_TRANSPOSE_REMAP( x, 12, 15 ) \
- HS_TRANSPOSE_REMAP( x, 13, 4 ) \
- HS_TRANSPOSE_REMAP( x, 14, 8 ) \
- HS_TRANSPOSE_REMAP( x, 15, 12 ) \
- HS_TRANSPOSE_REMAP( x, 16, 16 ) \
- HS_EMPTY
-
-#endif
-
-//
-//
-//
-
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_modules.h b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_modules.h
deleted file mode 100644
index 5099a65f3358c5d050f53aac0f66df5d26be5047..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_modules.h
+++ /dev/null
@@ -1,80 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "hs_bs_0.len.xxd"
-,
-#include "hs_bs_0.spv.xxd"
-,
-#include "hs_bs_1.len.xxd"
-,
-#include "hs_bs_1.spv.xxd"
-,
-#include "hs_bs_2.len.xxd"
-,
-#include "hs_bs_2.spv.xxd"
-,
-#include "hs_bs_3.len.xxd"
-,
-#include "hs_bs_3.spv.xxd"
-,
-#include "hs_bs_4.len.xxd"
-,
-#include "hs_bs_4.spv.xxd"
-,
-#include "hs_bc_0.len.xxd"
-,
-#include "hs_bc_0.spv.xxd"
-,
-#include "hs_bc_1.len.xxd"
-,
-#include "hs_bc_1.spv.xxd"
-,
-#include "hs_bc_2.len.xxd"
-,
-#include "hs_bc_2.spv.xxd"
-,
-#include "hs_bc_3.len.xxd"
-,
-#include "hs_bc_3.spv.xxd"
-,
-#include "hs_bc_4.len.xxd"
-,
-#include "hs_bc_4.spv.xxd"
-,
-#include "hs_fm_1_0.len.xxd"
-,
-#include "hs_fm_1_0.spv.xxd"
-,
-#include "hs_fm_1_1.len.xxd"
-,
-#include "hs_fm_1_1.spv.xxd"
-,
-#include "hs_fm_1_2.len.xxd"
-,
-#include "hs_fm_1_2.spv.xxd"
-,
-#include "hs_fm_1_3.len.xxd"
-,
-#include "hs_fm_1_3.spv.xxd"
-,
-#include "hs_fm_1_4.len.xxd"
-,
-#include "hs_fm_1_4.spv.xxd"
-,
-#include "hs_hm_1.len.xxd"
-,
-#include "hs_hm_1.spv.xxd"
-,
-#include "hs_fill_in.len.xxd"
-,
-#include "hs_fill_in.spv.xxd"
-,
-#include "hs_fill_out.len.xxd"
-,
-#include "hs_fill_out.spv.xxd"
-,
-#include "hs_transpose.len.xxd"
-,
-#include "hs_transpose.spv.xxd"
-,
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_target.h b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_target.h
deleted file mode 100644
index 00b3c9877b1560b7a53211ba032a3b6c0638c7bb..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u32/hs_target.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#pragma once
-
-//
-//
-//
-
-#include "hs_vk_target.h"
-
-//
-//
-//
-
-#undef HS_TARGET_NAME
-#define HS_TARGET_NAME hs_amd_gcn_u32
-
-//
-//
-//
-
-extern struct hs_vk_target const HS_TARGET_NAME;
-
-//
-//
-//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/gen.sh b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/gen.sh
deleted file mode 100644
index c5aec69482e8676d8fdc2a395dde673e5b8e7f80..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/gen.sh
+++ /dev/null
@@ -1,80 +0,0 @@
-#!/bin/bash
-
-##
-## Copyright 2019 The Fuchsia Authors. All rights reserved.
-## Use of this source code is governed by a BSD-style license that can be
-## found in the LICENSE file.
-##
-
-##
-## exit on error
-##
-
-## set -e
-
-##
-## delete the previous images
-##
-
-rm *.comp
-rm *.spv
-rm *.xxd
-
-##
-##
-##
-
-HS_GEN=../../../../../gen/hs_gen
-
-## --- 64-bit keys
-
-$HS_GEN -v -a "glsl" -D HS_AMD_GCN -t 2 -w 64 -r 8 -s 32768 -S 32768 -b 16 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
-
-##
-## remove trailing whitespace from generated files
-##
-
-sed -i 's/[[:space:]]*$//' hs_config.h
-sed -i 's/[[:space:]]*$//' hs_modules.h
-
-##
-##
-##
-
-whereis glslangValidator
-
-##
-## FIXME -- convert this to a bash script
-##
-## Note that we can use xargs instead of the cmd for/do
-##
-
-for f in *.comp
-do
- dos2unix $f
- clang-format -style=Mozilla -i $f
- cpp -P -I ../.. -I ../../../.. $f > ${f%%.*}.pre.comp
- clang-format -style=Mozilla -i ${f%%.*}.pre.comp
- glslangValidator --target-env vulkan1.1 -o ${f%%.*}.spv ${f%%.*}.pre.comp
- spirv-opt -O ${f%%.*}.spv -o ${f%%.*}.spv
-## spirv-remap -v --do-everything --input %%~nf.spv --output remap
- xxd -i < ${f%%.*}.spv > ${f%%.*}.spv.xxd
- len=$(wc -c < ${f%%.*}.spv)
- echo ${f%%.*}.spv $len
- printf "%.8x" $len | xxd -r -p | xxd -i > ${f%%.*}.len.xxd
-done
-
-##
-## dump a binary
-##
-
-cc -I ../../../.. -I ../../../../../.. -D=HS_DUMP -o hs_dump *.c
-hs_dump
-
-##
-## delete temporary files
-##
-
-rm *.pre.comp
-rm *.comp
-rm *.spv
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_config.h b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_config.h
deleted file mode 100644
index ce5ddcaee9b7c98b825b35f80056c59ffe264c26..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_config.h
+++ /dev/null
@@ -1,87 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_ONCE
-#define HS_GLSL_ONCE
-
-#define HS_SLAB_THREADS_LOG2 6
-#define HS_SLAB_THREADS (1 << HS_SLAB_THREADS_LOG2)
-#define HS_SLAB_WIDTH_LOG2 6
-#define HS_SLAB_WIDTH (1 << HS_SLAB_WIDTH_LOG2)
-#define HS_SLAB_HEIGHT 8
-#define HS_SLAB_KEYS (HS_SLAB_WIDTH * HS_SLAB_HEIGHT)
-#define HS_REG_LAST(c) c##8
-#define HS_KEY_DWORDS 2
-#define HS_VAL_DWORDS 0
-#define HS_BS_SLABS 16
-#define HS_BS_SLABS_LOG2_RU 4
-#define HS_BC_SLABS_LOG2_MAX 4
-#define HS_FM_BLOCK_HEIGHT 1
-#define HS_FM_SCALE_MIN 1
-#define HS_FM_SCALE_MAX 1
-#define HS_HM_BLOCK_HEIGHT 1
-#define HS_HM_SCALE_MIN 1
-#define HS_HM_SCALE_MAX 1
-#define HS_EMPTY
-
-#define HS_AMD_GCN
-
-#define HS_SLAB_ROWS() \
- HS_SLAB_ROW( 1, 0 ) \
- HS_SLAB_ROW( 2, 1 ) \
- HS_SLAB_ROW( 3, 2 ) \
- HS_SLAB_ROW( 4, 3 ) \
- HS_SLAB_ROW( 5, 4 ) \
- HS_SLAB_ROW( 6, 5 ) \
- HS_SLAB_ROW( 7, 6 ) \
- HS_SLAB_ROW( 8, 7 ) \
- HS_EMPTY
-
-#define HS_TRANSPOSE_SLAB() \
- HS_TRANSPOSE_STAGE( 1 ) \
- HS_TRANSPOSE_STAGE( 2 ) \
- HS_TRANSPOSE_STAGE( 3 ) \
- HS_TRANSPOSE_STAGE( 4 ) \
- HS_TRANSPOSE_STAGE( 5 ) \
- HS_TRANSPOSE_STAGE( 6 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 2, 1 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 4, 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 6, 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 8, 7 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 3, 1 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 4, 2 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 7, 5 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 8, 6 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 5, 1 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 6, 2 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 7, 3 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 8, 4 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 2, 1 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 4, 3 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 6, 5 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 8, 7 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 3, 1 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 4, 2 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 7, 5 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 8, 6 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 5, 1 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 6, 2 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 7, 3 ) \
- HS_TRANSPOSE_BLEND( w, x, 6, 8, 4 ) \
- HS_TRANSPOSE_REMAP( x, 1, 1 ) \
- HS_TRANSPOSE_REMAP( x, 2, 2 ) \
- HS_TRANSPOSE_REMAP( x, 3, 3 ) \
- HS_TRANSPOSE_REMAP( x, 4, 4 ) \
- HS_TRANSPOSE_REMAP( x, 5, 5 ) \
- HS_TRANSPOSE_REMAP( x, 6, 6 ) \
- HS_TRANSPOSE_REMAP( x, 7, 7 ) \
- HS_TRANSPOSE_REMAP( x, 8, 8 ) \
- HS_EMPTY
-
-#endif
-
-//
-//
-//
-
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_modules.h b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_modules.h
deleted file mode 100644
index 5099a65f3358c5d050f53aac0f66df5d26be5047..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_modules.h
+++ /dev/null
@@ -1,80 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "hs_bs_0.len.xxd"
-,
-#include "hs_bs_0.spv.xxd"
-,
-#include "hs_bs_1.len.xxd"
-,
-#include "hs_bs_1.spv.xxd"
-,
-#include "hs_bs_2.len.xxd"
-,
-#include "hs_bs_2.spv.xxd"
-,
-#include "hs_bs_3.len.xxd"
-,
-#include "hs_bs_3.spv.xxd"
-,
-#include "hs_bs_4.len.xxd"
-,
-#include "hs_bs_4.spv.xxd"
-,
-#include "hs_bc_0.len.xxd"
-,
-#include "hs_bc_0.spv.xxd"
-,
-#include "hs_bc_1.len.xxd"
-,
-#include "hs_bc_1.spv.xxd"
-,
-#include "hs_bc_2.len.xxd"
-,
-#include "hs_bc_2.spv.xxd"
-,
-#include "hs_bc_3.len.xxd"
-,
-#include "hs_bc_3.spv.xxd"
-,
-#include "hs_bc_4.len.xxd"
-,
-#include "hs_bc_4.spv.xxd"
-,
-#include "hs_fm_1_0.len.xxd"
-,
-#include "hs_fm_1_0.spv.xxd"
-,
-#include "hs_fm_1_1.len.xxd"
-,
-#include "hs_fm_1_1.spv.xxd"
-,
-#include "hs_fm_1_2.len.xxd"
-,
-#include "hs_fm_1_2.spv.xxd"
-,
-#include "hs_fm_1_3.len.xxd"
-,
-#include "hs_fm_1_3.spv.xxd"
-,
-#include "hs_fm_1_4.len.xxd"
-,
-#include "hs_fm_1_4.spv.xxd"
-,
-#include "hs_hm_1.len.xxd"
-,
-#include "hs_hm_1.spv.xxd"
-,
-#include "hs_fill_in.len.xxd"
-,
-#include "hs_fill_in.spv.xxd"
-,
-#include "hs_fill_out.len.xxd"
-,
-#include "hs_fill_out.spv.xxd"
-,
-#include "hs_transpose.len.xxd"
-,
-#include "hs_transpose.spv.xxd"
-,
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_target.h b/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_target.h
deleted file mode 100644
index aa21e11fa1c37fcf838b05031b7f05e24341d906..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/amd/gcn3/u64/hs_target.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#pragma once
-
-//
-//
-//
-
-#include "hs_vk_target.h"
-
-//
-//
-//
-
-#undef HS_TARGET_NAME
-#define HS_TARGET_NAME hs_amd_gcn_u64
-
-//
-//
-//
-
-extern struct hs_vk_target const HS_TARGET_NAME;
-
-//
-//
-//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/gen.sh b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/gen.sh
deleted file mode 100644
index 647cf3c36192eb1db679f5aede1c43cd05d090ca..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/gen.sh
+++ /dev/null
@@ -1,80 +0,0 @@
-#!/bin/bash
-
-##
-## Copyright 2019 The Fuchsia Authors. All rights reserved.
-## Use of this source code is governed by a BSD-style license that can be
-## found in the LICENSE file.
-##
-
-##
-## exit on error
-##
-
-## set -e
-
-##
-## delete the previous images
-##
-
-rm *.comp
-rm *.spv
-rm *.xxd
-
-##
-##
-##
-
-HS_GEN=../../../../../gen/hs_gen
-
-## --- 32-bit keys ---
-
-$HS_GEN -v -a "glsl" -D HS_INTEL_GEN8 -t 1 -w 16 -r 8 -s 21504 -S 65536 -b 16 -B 48 -m 1 -M 1 -f 0 -F 0 -c 0 -C 0 -z
-
-##
-## remove trailing whitespace from generated files
-##
-
-sed -i 's/[[:space:]]*$//' hs_config.h
-sed -i 's/[[:space:]]*$//' hs_modules.h
-
-##
-##
-##
-
-whereis glslangValidator
-
-##
-## FIXME -- convert this to a bash script
-##
-## Note that we can use xargs instead of the cmd for/do
-##
-
-for f in *.comp
-do
- dos2unix $f
- clang-format -style=Mozilla -i $f
- cpp -P -I ../.. -I ../../../.. $f > ${f%%.*}.pre.comp
- clang-format -style=Mozilla -i ${f%%.*}.pre.comp
- glslangValidator --target-env vulkan1.1 -o ${f%%.*}.spv ${f%%.*}.pre.comp
- spirv-opt -O ${f%%.*}.spv -o ${f%%.*}.spv
-## spirv-remap -v --do-everything --input %%~nf.spv --output remap
- xxd -i < ${f%%.*}.spv > ${f%%.*}.spv.xxd
- len=$(wc -c < ${f%%.*}.spv)
- echo ${f%%.*}.spv $len
- printf "%.8x" $len | xxd -r -p | xxd -i > ${f%%.*}.len.xxd
-done
-
-##
-## dump a binary
-##
-
-cc -I ../../../.. -I ../../../../../.. -D=HS_DUMP -o hs_dump *.c
-hs_dump
-
-##
-## delete temporary files
-##
-
-rm *.pre.comp
-rm *.comp
-rm *.spv
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_config.h b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_config.h
deleted file mode 100644
index fad716aae2879d4c64d71d2e96f54534b126f56d..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_config.h
+++ /dev/null
@@ -1,77 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_ONCE
-#define HS_GLSL_ONCE
-
-#define HS_SLAB_THREADS_LOG2 4
-#define HS_SLAB_THREADS (1 << HS_SLAB_THREADS_LOG2)
-#define HS_SLAB_WIDTH_LOG2 4
-#define HS_SLAB_WIDTH (1 << HS_SLAB_WIDTH_LOG2)
-#define HS_SLAB_HEIGHT 8
-#define HS_SLAB_KEYS (HS_SLAB_WIDTH * HS_SLAB_HEIGHT)
-#define HS_REG_LAST(c) c##8
-#define HS_KEY_DWORDS 1
-#define HS_VAL_DWORDS 0
-#define HS_BS_SLABS 16
-#define HS_BS_SLABS_LOG2_RU 4
-#define HS_BC_SLABS_LOG2_MAX 4
-#define HS_FM_BLOCK_HEIGHT 1
-#define HS_FM_SCALE_MIN 0
-#define HS_FM_SCALE_MAX 0
-#define HS_HM_BLOCK_HEIGHT 1
-#define HS_HM_SCALE_MIN 0
-#define HS_HM_SCALE_MAX 0
-#define HS_EMPTY
-
-#define HS_INTEL_GEN8
-
-#define HS_SLAB_ROWS() \
- HS_SLAB_ROW( 1, 0 ) \
- HS_SLAB_ROW( 2, 1 ) \
- HS_SLAB_ROW( 3, 2 ) \
- HS_SLAB_ROW( 4, 3 ) \
- HS_SLAB_ROW( 5, 4 ) \
- HS_SLAB_ROW( 6, 5 ) \
- HS_SLAB_ROW( 7, 6 ) \
- HS_SLAB_ROW( 8, 7 ) \
- HS_EMPTY
-
-#define HS_TRANSPOSE_SLAB() \
- HS_TRANSPOSE_STAGE( 1 ) \
- HS_TRANSPOSE_STAGE( 2 ) \
- HS_TRANSPOSE_STAGE( 3 ) \
- HS_TRANSPOSE_STAGE( 4 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 2, 1 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 4, 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 6, 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 8, 7 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 3, 1 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 4, 2 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 7, 5 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 8, 6 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 5, 1 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 6, 2 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 7, 3 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 8, 4 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 2, 1 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 4, 3 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 6, 5 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 8, 7 ) \
- HS_TRANSPOSE_REMAP( v, 1, 1 ) \
- HS_TRANSPOSE_REMAP( v, 2, 5 ) \
- HS_TRANSPOSE_REMAP( v, 3, 2 ) \
- HS_TRANSPOSE_REMAP( v, 4, 6 ) \
- HS_TRANSPOSE_REMAP( v, 5, 3 ) \
- HS_TRANSPOSE_REMAP( v, 6, 7 ) \
- HS_TRANSPOSE_REMAP( v, 7, 4 ) \
- HS_TRANSPOSE_REMAP( v, 8, 8 ) \
- HS_EMPTY
-
-#endif
-
-//
-//
-//
-
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_modules.h b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_modules.h
deleted file mode 100644
index db653be10496975d1d485d110727057c14df7af5..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_modules.h
+++ /dev/null
@@ -1,76 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "hs_bs_0.len.xxd"
-,
-#include "hs_bs_0.spv.xxd"
-,
-#include "hs_bs_1.len.xxd"
-,
-#include "hs_bs_1.spv.xxd"
-,
-#include "hs_bs_2.len.xxd"
-,
-#include "hs_bs_2.spv.xxd"
-,
-#include "hs_bs_3.len.xxd"
-,
-#include "hs_bs_3.spv.xxd"
-,
-#include "hs_bs_4.len.xxd"
-,
-#include "hs_bs_4.spv.xxd"
-,
-#include "hs_bc_0.len.xxd"
-,
-#include "hs_bc_0.spv.xxd"
-,
-#include "hs_bc_1.len.xxd"
-,
-#include "hs_bc_1.spv.xxd"
-,
-#include "hs_bc_2.len.xxd"
-,
-#include "hs_bc_2.spv.xxd"
-,
-#include "hs_bc_3.len.xxd"
-,
-#include "hs_bc_3.spv.xxd"
-,
-#include "hs_bc_4.len.xxd"
-,
-#include "hs_bc_4.spv.xxd"
-,
-#include "hs_fm_0_0.len.xxd"
-,
-#include "hs_fm_0_0.spv.xxd"
-,
-#include "hs_fm_0_1.len.xxd"
-,
-#include "hs_fm_0_1.spv.xxd"
-,
-#include "hs_fm_0_2.len.xxd"
-,
-#include "hs_fm_0_2.spv.xxd"
-,
-#include "hs_fm_0_3.len.xxd"
-,
-#include "hs_fm_0_3.spv.xxd"
-,
-#include "hs_hm_0.len.xxd"
-,
-#include "hs_hm_0.spv.xxd"
-,
-#include "hs_fill_in.len.xxd"
-,
-#include "hs_fill_in.spv.xxd"
-,
-#include "hs_fill_out.len.xxd"
-,
-#include "hs_fill_out.spv.xxd"
-,
-#include "hs_transpose.len.xxd"
-,
-#include "hs_transpose.spv.xxd"
-,
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_target.h b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_target.h
deleted file mode 100644
index fb8ddd55cdff7f7d3350b531ba440c63e056b8c2..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u32/hs_target.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#pragma once
-
-//
-//
-//
-
-#include "hs_vk_target.h"
-
-//
-//
-//
-
-#undef HS_TARGET_NAME
-#define HS_TARGET_NAME hs_intel_gen8_u32
-
-//
-//
-//
-
-extern struct hs_vk_target const HS_TARGET_NAME;
-
-//
-//
-//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/gen.sh b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/gen.sh
deleted file mode 100644
index 63d5285e3b2c8380e201cae32b9e6ea288c87f71..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/gen.sh
+++ /dev/null
@@ -1,80 +0,0 @@
-#!/bin/bash
-
-##
-## Copyright 2019 The Fuchsia Authors. All rights reserved.
-## Use of this source code is governed by a BSD-style license that can be
-## found in the LICENSE file.
-##
-
-##
-## exit on error
-##
-
-## set -e
-
-##
-## delete the previous images
-##
-
-rm *.comp
-rm *.spv
-rm *.xxd
-
-##
-##
-##
-
-HS_GEN=../../../../../gen/hs_gen
-
-## --- 64-bit keys ---
-
-$HS_GEN -v -a "glsl" -D HS_INTEL_GEN8 -t 2 -w 8 -r 16 -s 21504 -S 65536 -b 16 -B 48 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
-
-##
-## remove trailing whitespace from generated files
-##
-
-sed -i 's/[[:space:]]*$//' hs_config.h
-sed -i 's/[[:space:]]*$//' hs_modules.h
-
-##
-##
-##
-
-whereis glslangValidator
-
-##
-## FIXME -- convert this to a bash script
-##
-## Note that we can use xargs instead of the cmd for/do
-##
-
-for f in *.comp
-do
- dos2unix $f
- clang-format -style=Mozilla -i $f
- cpp -P -I ../.. -I ../../../.. $f > ${f%%.*}.pre.comp
- clang-format -style=Mozilla -i ${f%%.*}.pre.comp
- glslangValidator --target-env vulkan1.1 -o ${f%%.*}.spv ${f%%.*}.pre.comp
- spirv-opt -O ${f%%.*}.spv -o ${f%%.*}.spv
-## spirv-remap -v --do-everything --input %%~nf.spv --output remap
- xxd -i < ${f%%.*}.spv > ${f%%.*}.spv.xxd
- len=$(wc -c < ${f%%.*}.spv)
- echo ${f%%.*}.spv $len
- printf "%.8x" $len | xxd -r -p | xxd -i > ${f%%.*}.len.xxd
-done
-
-##
-## dump a binary
-##
-
-cc -I ../../../.. -I ../../../../../.. -D=HS_DUMP -o hs_dump *.c
-hs_dump
-
-##
-## delete temporary files
-##
-
-rm *.pre.comp
-rm *.comp
-rm *.spv
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_config.h b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_config.h
deleted file mode 100644
index 6fd4c85283a9d0f19a0455b5fb8789caaa31c0b4..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_config.h
+++ /dev/null
@@ -1,100 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_ONCE
-#define HS_GLSL_ONCE
-
-#define HS_SLAB_THREADS_LOG2 3
-#define HS_SLAB_THREADS (1 << HS_SLAB_THREADS_LOG2)
-#define HS_SLAB_WIDTH_LOG2 3
-#define HS_SLAB_WIDTH (1 << HS_SLAB_WIDTH_LOG2)
-#define HS_SLAB_HEIGHT 16
-#define HS_SLAB_KEYS (HS_SLAB_WIDTH * HS_SLAB_HEIGHT)
-#define HS_REG_LAST(c) c##16
-#define HS_KEY_DWORDS 2
-#define HS_VAL_DWORDS 0
-#define HS_BS_SLABS 16
-#define HS_BS_SLABS_LOG2_RU 4
-#define HS_BC_SLABS_LOG2_MAX 4
-#define HS_FM_BLOCK_HEIGHT 1
-#define HS_FM_SCALE_MIN 1
-#define HS_FM_SCALE_MAX 1
-#define HS_HM_BLOCK_HEIGHT 1
-#define HS_HM_SCALE_MIN 1
-#define HS_HM_SCALE_MAX 1
-#define HS_EMPTY
-
-#define HS_INTEL_GEN8
-
-#define HS_SLAB_ROWS() \
- HS_SLAB_ROW( 1, 0 ) \
- HS_SLAB_ROW( 2, 1 ) \
- HS_SLAB_ROW( 3, 2 ) \
- HS_SLAB_ROW( 4, 3 ) \
- HS_SLAB_ROW( 5, 4 ) \
- HS_SLAB_ROW( 6, 5 ) \
- HS_SLAB_ROW( 7, 6 ) \
- HS_SLAB_ROW( 8, 7 ) \
- HS_SLAB_ROW( 9, 8 ) \
- HS_SLAB_ROW( 10, 9 ) \
- HS_SLAB_ROW( 11, 10 ) \
- HS_SLAB_ROW( 12, 11 ) \
- HS_SLAB_ROW( 13, 12 ) \
- HS_SLAB_ROW( 14, 13 ) \
- HS_SLAB_ROW( 15, 14 ) \
- HS_SLAB_ROW( 16, 15 ) \
- HS_EMPTY
-
-#define HS_TRANSPOSE_SLAB() \
- HS_TRANSPOSE_STAGE( 1 ) \
- HS_TRANSPOSE_STAGE( 2 ) \
- HS_TRANSPOSE_STAGE( 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 2, 1 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 4, 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 6, 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 8, 7 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 10, 9 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 12, 11 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 14, 13 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 16, 15 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 3, 1 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 4, 2 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 7, 5 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 8, 6 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 11, 9 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 12, 10 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 15, 13 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 16, 14 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 5, 1 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 6, 2 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 7, 3 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 8, 4 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 13, 9 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 14, 10 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 15, 11 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 16, 12 ) \
- HS_TRANSPOSE_REMAP( u, 1, 1 ) \
- HS_TRANSPOSE_REMAP( u, 2, 3 ) \
- HS_TRANSPOSE_REMAP( u, 3, 5 ) \
- HS_TRANSPOSE_REMAP( u, 4, 7 ) \
- HS_TRANSPOSE_REMAP( u, 5, 9 ) \
- HS_TRANSPOSE_REMAP( u, 6, 11 ) \
- HS_TRANSPOSE_REMAP( u, 7, 13 ) \
- HS_TRANSPOSE_REMAP( u, 8, 15 ) \
- HS_TRANSPOSE_REMAP( u, 9, 2 ) \
- HS_TRANSPOSE_REMAP( u, 10, 4 ) \
- HS_TRANSPOSE_REMAP( u, 11, 6 ) \
- HS_TRANSPOSE_REMAP( u, 12, 8 ) \
- HS_TRANSPOSE_REMAP( u, 13, 10 ) \
- HS_TRANSPOSE_REMAP( u, 14, 12 ) \
- HS_TRANSPOSE_REMAP( u, 15, 14 ) \
- HS_TRANSPOSE_REMAP( u, 16, 16 ) \
- HS_EMPTY
-
-#endif
-
-//
-//
-//
-
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_modules.h b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_modules.h
deleted file mode 100644
index 5099a65f3358c5d050f53aac0f66df5d26be5047..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_modules.h
+++ /dev/null
@@ -1,80 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "hs_bs_0.len.xxd"
-,
-#include "hs_bs_0.spv.xxd"
-,
-#include "hs_bs_1.len.xxd"
-,
-#include "hs_bs_1.spv.xxd"
-,
-#include "hs_bs_2.len.xxd"
-,
-#include "hs_bs_2.spv.xxd"
-,
-#include "hs_bs_3.len.xxd"
-,
-#include "hs_bs_3.spv.xxd"
-,
-#include "hs_bs_4.len.xxd"
-,
-#include "hs_bs_4.spv.xxd"
-,
-#include "hs_bc_0.len.xxd"
-,
-#include "hs_bc_0.spv.xxd"
-,
-#include "hs_bc_1.len.xxd"
-,
-#include "hs_bc_1.spv.xxd"
-,
-#include "hs_bc_2.len.xxd"
-,
-#include "hs_bc_2.spv.xxd"
-,
-#include "hs_bc_3.len.xxd"
-,
-#include "hs_bc_3.spv.xxd"
-,
-#include "hs_bc_4.len.xxd"
-,
-#include "hs_bc_4.spv.xxd"
-,
-#include "hs_fm_1_0.len.xxd"
-,
-#include "hs_fm_1_0.spv.xxd"
-,
-#include "hs_fm_1_1.len.xxd"
-,
-#include "hs_fm_1_1.spv.xxd"
-,
-#include "hs_fm_1_2.len.xxd"
-,
-#include "hs_fm_1_2.spv.xxd"
-,
-#include "hs_fm_1_3.len.xxd"
-,
-#include "hs_fm_1_3.spv.xxd"
-,
-#include "hs_fm_1_4.len.xxd"
-,
-#include "hs_fm_1_4.spv.xxd"
-,
-#include "hs_hm_1.len.xxd"
-,
-#include "hs_hm_1.spv.xxd"
-,
-#include "hs_fill_in.len.xxd"
-,
-#include "hs_fill_in.spv.xxd"
-,
-#include "hs_fill_out.len.xxd"
-,
-#include "hs_fill_out.spv.xxd"
-,
-#include "hs_transpose.len.xxd"
-,
-#include "hs_transpose.spv.xxd"
-,
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_target.h b/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_target.h
deleted file mode 100644
index c4329c342c9a84bf8d78602d4b77c10682826e0a..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/intel/gen8/u64/hs_target.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#pragma once
-
-//
-//
-//
-
-#include "hs_vk_target.h"
-
-//
-//
-//
-
-#undef HS_TARGET_NAME
-#define HS_TARGET_NAME hs_intel_gen8_u64
-
-//
-//
-//
-
-extern struct hs_vk_target const HS_TARGET_NAME;
-
-//
-//
-//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/gen.sh b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/gen.sh
deleted file mode 100644
index 107d0cba6db9fed011e5dcae9789f32578e0aa0a..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/gen.sh
+++ /dev/null
@@ -1,81 +0,0 @@
-#!/bin/bash
-
-##
-## Copyright 2019 The Fuchsia Authors. All rights reserved.
-## Use of this source code is governed by a BSD-style license that can be
-## found in the LICENSE file.
-##
-
-##
-## exit on error
-##
-
-## set -e
-
-##
-## delete the previous images
-##
-
-rm *.comp
-rm *.spv
-rm *.xxd
-
-##
-##
-##
-
-HS_GEN=../../../../../gen/hs_gen
-
-## --- 32-bit keys ---
-
-## $HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 1 -w 32 -r 32 -s 49152 -S 65536 -b 32 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
-$HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 1 -w 32 -r 16 -s 32768 -S 32768 -b 16 -m 1 -M 1 -p 1 -P 1 -f 0 -F 0 -c 0 -C 0 -z
-
-##
-## remove trailing whitespace from generated files
-##
-
-sed -i 's/[[:space:]]*$//' hs_config.h
-sed -i 's/[[:space:]]*$//' hs_modules.h
-
-##
-##
-##
-
-whereis glslangValidator
-
-##
-## FIXME -- convert this to a bash script
-##
-## Note that we can use xargs instead of the cmd for/do
-##
-
-for f in *.comp
-do
- dos2unix $f
- clang-format -style=Mozilla -i $f
- cpp -P -I ../.. -I ../../../.. $f > ${f%%.*}.pre.comp
- clang-format -style=Mozilla -i ${f%%.*}.pre.comp
- glslangValidator --target-env vulkan1.1 -o ${f%%.*}.spv ${f%%.*}.pre.comp
- spirv-opt -O ${f%%.*}.spv -o ${f%%.*}.spv
-## spirv-remap -v --do-everything --input %%~nf.spv --output remap
- xxd -i < ${f%%.*}.spv > ${f%%.*}.spv.xxd
- len=$(wc -c < ${f%%.*}.spv)
- echo ${f%%.*}.spv $len
- printf "%.8x" $len | xxd -r -p | xxd -i > ${f%%.*}.len.xxd
-done
-
-##
-## dump a binary
-##
-
-cc -I ../../../.. -I ../../../../../.. -D=HS_DUMP -o hs_dump *.c
-hs_dump
-
-##
-## delete temporary files
-##
-
-rm *.pre.comp
-rm *.comp
-rm *.spv
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_config.h b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_config.h
deleted file mode 100644
index f94a30480270083509fdc9944f46d1e95cba24bb..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_config.h
+++ /dev/null
@@ -1,118 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_ONCE
-#define HS_GLSL_ONCE
-
-#define HS_SLAB_THREADS_LOG2 5
-#define HS_SLAB_THREADS (1 << HS_SLAB_THREADS_LOG2)
-#define HS_SLAB_WIDTH_LOG2 5
-#define HS_SLAB_WIDTH (1 << HS_SLAB_WIDTH_LOG2)
-#define HS_SLAB_HEIGHT 16
-#define HS_SLAB_KEYS (HS_SLAB_WIDTH * HS_SLAB_HEIGHT)
-#define HS_REG_LAST(c) c##16
-#define HS_KEY_DWORDS 1
-#define HS_VAL_DWORDS 0
-#define HS_BS_SLABS 16
-#define HS_BS_SLABS_LOG2_RU 4
-#define HS_BC_SLABS_LOG2_MAX 4
-#define HS_FM_BLOCK_HEIGHT 1
-#define HS_FM_SCALE_MIN 0
-#define HS_FM_SCALE_MAX 0
-#define HS_HM_BLOCK_HEIGHT 1
-#define HS_HM_SCALE_MIN 0
-#define HS_HM_SCALE_MAX 0
-#define HS_EMPTY
-
-#define HS_NVIDIA_SM35
-
-#define HS_SLAB_ROWS() \
- HS_SLAB_ROW( 1, 0 ) \
- HS_SLAB_ROW( 2, 1 ) \
- HS_SLAB_ROW( 3, 2 ) \
- HS_SLAB_ROW( 4, 3 ) \
- HS_SLAB_ROW( 5, 4 ) \
- HS_SLAB_ROW( 6, 5 ) \
- HS_SLAB_ROW( 7, 6 ) \
- HS_SLAB_ROW( 8, 7 ) \
- HS_SLAB_ROW( 9, 8 ) \
- HS_SLAB_ROW( 10, 9 ) \
- HS_SLAB_ROW( 11, 10 ) \
- HS_SLAB_ROW( 12, 11 ) \
- HS_SLAB_ROW( 13, 12 ) \
- HS_SLAB_ROW( 14, 13 ) \
- HS_SLAB_ROW( 15, 14 ) \
- HS_SLAB_ROW( 16, 15 ) \
- HS_EMPTY
-
-#define HS_TRANSPOSE_SLAB() \
- HS_TRANSPOSE_STAGE( 1 ) \
- HS_TRANSPOSE_STAGE( 2 ) \
- HS_TRANSPOSE_STAGE( 3 ) \
- HS_TRANSPOSE_STAGE( 4 ) \
- HS_TRANSPOSE_STAGE( 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 2, 1 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 4, 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 6, 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 8, 7 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 10, 9 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 12, 11 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 14, 13 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 16, 15 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 3, 1 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 4, 2 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 7, 5 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 8, 6 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 11, 9 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 12, 10 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 15, 13 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 16, 14 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 5, 1 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 6, 2 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 7, 3 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 8, 4 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 13, 9 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 14, 10 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 15, 11 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 16, 12 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 9, 1 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 10, 2 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 11, 3 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 12, 4 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 13, 5 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 14, 6 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 15, 7 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 16, 8 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 2, 1 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 4, 3 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 6, 5 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 8, 7 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 10, 9 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 12, 11 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 14, 13 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 16, 15 ) \
- HS_TRANSPOSE_REMAP( w, 1, 1 ) \
- HS_TRANSPOSE_REMAP( w, 2, 9 ) \
- HS_TRANSPOSE_REMAP( w, 3, 2 ) \
- HS_TRANSPOSE_REMAP( w, 4, 10 ) \
- HS_TRANSPOSE_REMAP( w, 5, 3 ) \
- HS_TRANSPOSE_REMAP( w, 6, 11 ) \
- HS_TRANSPOSE_REMAP( w, 7, 4 ) \
- HS_TRANSPOSE_REMAP( w, 8, 12 ) \
- HS_TRANSPOSE_REMAP( w, 9, 5 ) \
- HS_TRANSPOSE_REMAP( w, 10, 13 ) \
- HS_TRANSPOSE_REMAP( w, 11, 6 ) \
- HS_TRANSPOSE_REMAP( w, 12, 14 ) \
- HS_TRANSPOSE_REMAP( w, 13, 7 ) \
- HS_TRANSPOSE_REMAP( w, 14, 15 ) \
- HS_TRANSPOSE_REMAP( w, 15, 8 ) \
- HS_TRANSPOSE_REMAP( w, 16, 16 ) \
- HS_EMPTY
-
-#endif
-
-//
-//
-//
-
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_modules.h b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_modules.h
deleted file mode 100644
index db653be10496975d1d485d110727057c14df7af5..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_modules.h
+++ /dev/null
@@ -1,76 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "hs_bs_0.len.xxd"
-,
-#include "hs_bs_0.spv.xxd"
-,
-#include "hs_bs_1.len.xxd"
-,
-#include "hs_bs_1.spv.xxd"
-,
-#include "hs_bs_2.len.xxd"
-,
-#include "hs_bs_2.spv.xxd"
-,
-#include "hs_bs_3.len.xxd"
-,
-#include "hs_bs_3.spv.xxd"
-,
-#include "hs_bs_4.len.xxd"
-,
-#include "hs_bs_4.spv.xxd"
-,
-#include "hs_bc_0.len.xxd"
-,
-#include "hs_bc_0.spv.xxd"
-,
-#include "hs_bc_1.len.xxd"
-,
-#include "hs_bc_1.spv.xxd"
-,
-#include "hs_bc_2.len.xxd"
-,
-#include "hs_bc_2.spv.xxd"
-,
-#include "hs_bc_3.len.xxd"
-,
-#include "hs_bc_3.spv.xxd"
-,
-#include "hs_bc_4.len.xxd"
-,
-#include "hs_bc_4.spv.xxd"
-,
-#include "hs_fm_0_0.len.xxd"
-,
-#include "hs_fm_0_0.spv.xxd"
-,
-#include "hs_fm_0_1.len.xxd"
-,
-#include "hs_fm_0_1.spv.xxd"
-,
-#include "hs_fm_0_2.len.xxd"
-,
-#include "hs_fm_0_2.spv.xxd"
-,
-#include "hs_fm_0_3.len.xxd"
-,
-#include "hs_fm_0_3.spv.xxd"
-,
-#include "hs_hm_0.len.xxd"
-,
-#include "hs_hm_0.spv.xxd"
-,
-#include "hs_fill_in.len.xxd"
-,
-#include "hs_fill_in.spv.xxd"
-,
-#include "hs_fill_out.len.xxd"
-,
-#include "hs_fill_out.spv.xxd"
-,
-#include "hs_transpose.len.xxd"
-,
-#include "hs_transpose.spv.xxd"
-,
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_target.h b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_target.h
deleted file mode 100644
index 0b0b0bd1abd9d9a5aa2c01bd6871fd2fb85c814c..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u32/hs_target.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#pragma once
-
-//
-//
-//
-
-#include "hs_vk_target.h"
-
-//
-//
-//
-
-#undef HS_TARGET_NAME
-#define HS_TARGET_NAME hs_nvidia_sm35_u32
-
-//
-//
-//
-
-extern struct hs_vk_target const HS_TARGET_NAME;
-
-//
-//
-//
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/gen.sh b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/gen.sh
deleted file mode 100644
index a8ac7411089c2779f9dfaec6a5ab6ac09a272e10..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/gen.sh
+++ /dev/null
@@ -1,81 +0,0 @@
-#!/bin/bash
-
-##
-## Copyright 2019 The Fuchsia Authors. All rights reserved.
-## Use of this source code is governed by a BSD-style license that can be
-## found in the LICENSE file.
-##
-
-##
-## exit on error
-##
-
-## set -e
-
-##
-## delete the previous images
-##
-
-rm *.comp
-rm *.spv
-rm *.xxd
-
-##
-##
-##
-
-HS_GEN=../../../../../gen/hs_gen
-
-## --- 64-bit keys ---
-
-## $HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 2 -w 32 -r 32 -s 49152 -S 65536 -b 16 -m 1 -M 1 -f 1 -F 1 -c 1 -C 1 -z
-$HS_GEN -v -a "glsl" -D HS_NVIDIA_SM35 -t 2 -w 32 -r 8 -s 32768 -S 32768 -b 16 -m 1 -M 1 -p 1 -P 1 -f 0 -F 0 -c 0 -C 0 -z
-
-##
-## remove trailing whitespace from generated files
-##
-
-sed -i 's/[[:space:]]*$//' hs_config.h
-sed -i 's/[[:space:]]*$//' hs_modules.h
-
-##
-##
-##
-
-whereis glslangValidator
-
-##
-## FIXME -- convert this to a bash script
-##
-## Note that we can use xargs instead of the cmd for/do
-##
-
-for f in *.comp
-do
- dos2unix $f
- clang-format -style=Mozilla -i $f
- cpp -P -I ../.. -I ../../../.. $f > ${f%%.*}.pre.comp
- clang-format -style=Mozilla -i ${f%%.*}.pre.comp
- glslangValidator --target-env vulkan1.1 -o ${f%%.*}.spv ${f%%.*}.pre.comp
- spirv-opt -O ${f%%.*}.spv -o ${f%%.*}.spv
-## spirv-remap -v --do-everything --input %%~nf.spv --output remap
- xxd -i < ${f%%.*}.spv > ${f%%.*}.spv.xxd
- len=$(wc -c < ${f%%.*}.spv)
- echo ${f%%.*}.spv $len
- printf "%.8x" $len | xxd -r -p | xxd -i > ${f%%.*}.len.xxd
-done
-
-##
-## dump a binary
-##
-
-cc -I ../../../.. -I ../../../../../.. -D=HS_DUMP -o hs_dump *.c
-hs_dump
-
-##
-## delete temporary files
-##
-
-rm *.pre.comp
-rm *.comp
-rm *.spv
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_config.h b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_config.h
deleted file mode 100644
index f711d14ea738ee359d25c0900cf71f1cf34b288b..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_config.h
+++ /dev/null
@@ -1,82 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#ifndef HS_GLSL_ONCE
-#define HS_GLSL_ONCE
-
-#define HS_SLAB_THREADS_LOG2 5
-#define HS_SLAB_THREADS (1 << HS_SLAB_THREADS_LOG2)
-#define HS_SLAB_WIDTH_LOG2 5
-#define HS_SLAB_WIDTH (1 << HS_SLAB_WIDTH_LOG2)
-#define HS_SLAB_HEIGHT 8
-#define HS_SLAB_KEYS (HS_SLAB_WIDTH * HS_SLAB_HEIGHT)
-#define HS_REG_LAST(c) c##8
-#define HS_KEY_DWORDS 2
-#define HS_VAL_DWORDS 0
-#define HS_BS_SLABS 16
-#define HS_BS_SLABS_LOG2_RU 4
-#define HS_BC_SLABS_LOG2_MAX 4
-#define HS_FM_BLOCK_HEIGHT 1
-#define HS_FM_SCALE_MIN 0
-#define HS_FM_SCALE_MAX 0
-#define HS_HM_BLOCK_HEIGHT 1
-#define HS_HM_SCALE_MIN 0
-#define HS_HM_SCALE_MAX 0
-#define HS_EMPTY
-
-#define HS_NVIDIA_SM35
-
-#define HS_SLAB_ROWS() \
- HS_SLAB_ROW( 1, 0 ) \
- HS_SLAB_ROW( 2, 1 ) \
- HS_SLAB_ROW( 3, 2 ) \
- HS_SLAB_ROW( 4, 3 ) \
- HS_SLAB_ROW( 5, 4 ) \
- HS_SLAB_ROW( 6, 5 ) \
- HS_SLAB_ROW( 7, 6 ) \
- HS_SLAB_ROW( 8, 7 ) \
- HS_EMPTY
-
-#define HS_TRANSPOSE_SLAB() \
- HS_TRANSPOSE_STAGE( 1 ) \
- HS_TRANSPOSE_STAGE( 2 ) \
- HS_TRANSPOSE_STAGE( 3 ) \
- HS_TRANSPOSE_STAGE( 4 ) \
- HS_TRANSPOSE_STAGE( 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 2, 1 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 4, 3 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 6, 5 ) \
- HS_TRANSPOSE_BLEND( r, s, 1, 8, 7 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 3, 1 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 4, 2 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 7, 5 ) \
- HS_TRANSPOSE_BLEND( s, t, 2, 8, 6 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 5, 1 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 6, 2 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 7, 3 ) \
- HS_TRANSPOSE_BLEND( t, u, 3, 8, 4 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 2, 1 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 4, 3 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 6, 5 ) \
- HS_TRANSPOSE_BLEND( u, v, 4, 8, 7 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 3, 1 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 4, 2 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 7, 5 ) \
- HS_TRANSPOSE_BLEND( v, w, 5, 8, 6 ) \
- HS_TRANSPOSE_REMAP( w, 1, 1 ) \
- HS_TRANSPOSE_REMAP( w, 2, 3 ) \
- HS_TRANSPOSE_REMAP( w, 3, 5 ) \
- HS_TRANSPOSE_REMAP( w, 4, 7 ) \
- HS_TRANSPOSE_REMAP( w, 5, 2 ) \
- HS_TRANSPOSE_REMAP( w, 6, 4 ) \
- HS_TRANSPOSE_REMAP( w, 7, 6 ) \
- HS_TRANSPOSE_REMAP( w, 8, 8 ) \
- HS_EMPTY
-
-#endif
-
-//
-//
-//
-
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_modules.h b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_modules.h
deleted file mode 100644
index db653be10496975d1d485d110727057c14df7af5..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_modules.h
+++ /dev/null
@@ -1,76 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#include "hs_bs_0.len.xxd"
-,
-#include "hs_bs_0.spv.xxd"
-,
-#include "hs_bs_1.len.xxd"
-,
-#include "hs_bs_1.spv.xxd"
-,
-#include "hs_bs_2.len.xxd"
-,
-#include "hs_bs_2.spv.xxd"
-,
-#include "hs_bs_3.len.xxd"
-,
-#include "hs_bs_3.spv.xxd"
-,
-#include "hs_bs_4.len.xxd"
-,
-#include "hs_bs_4.spv.xxd"
-,
-#include "hs_bc_0.len.xxd"
-,
-#include "hs_bc_0.spv.xxd"
-,
-#include "hs_bc_1.len.xxd"
-,
-#include "hs_bc_1.spv.xxd"
-,
-#include "hs_bc_2.len.xxd"
-,
-#include "hs_bc_2.spv.xxd"
-,
-#include "hs_bc_3.len.xxd"
-,
-#include "hs_bc_3.spv.xxd"
-,
-#include "hs_bc_4.len.xxd"
-,
-#include "hs_bc_4.spv.xxd"
-,
-#include "hs_fm_0_0.len.xxd"
-,
-#include "hs_fm_0_0.spv.xxd"
-,
-#include "hs_fm_0_1.len.xxd"
-,
-#include "hs_fm_0_1.spv.xxd"
-,
-#include "hs_fm_0_2.len.xxd"
-,
-#include "hs_fm_0_2.spv.xxd"
-,
-#include "hs_fm_0_3.len.xxd"
-,
-#include "hs_fm_0_3.spv.xxd"
-,
-#include "hs_hm_0.len.xxd"
-,
-#include "hs_hm_0.spv.xxd"
-,
-#include "hs_fill_in.len.xxd"
-,
-#include "hs_fill_in.spv.xxd"
-,
-#include "hs_fill_out.len.xxd"
-,
-#include "hs_fill_out.spv.xxd"
-,
-#include "hs_transpose.len.xxd"
-,
-#include "hs_transpose.spv.xxd"
-,
diff --git a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_target.h b/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_target.h
deleted file mode 100644
index 6cf4dd3d05318246c832e16d6d61e849422a33e4..0000000000000000000000000000000000000000
--- a/src/graphics/lib/compute/hotsort/vk/targets/nvidia/sm_35/u64/hs_target.h
+++ /dev/null
@@ -1,28 +0,0 @@
-// Copyright 2019 The Fuchsia Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style license that can be
-// found in the LICENSE file.
-
-#pragma once
-
-//
-//
-//
-
-#include "hs_vk_target.h"
-
-//
-//
-//
-
-#undef HS_TARGET_NAME
-#define HS_TARGET_NAME hs_nvidia_sm35_u64
-
-//
-//
-//
-
-extern struct hs_vk_target const HS_TARGET_NAME;
-
-//
-//
-//