radeonsi: add s_sethalt to shaders for debugging

Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>

diff --git a/src/amd/common/ac_rtld.c b/src/amd/common/ac_rtld.c
index 92020c5..c750dbf 100644 (file)
--- a/src/amd/common/ac_rtld.c
+++ b/src/amd/common/ac_rtld.c
@@ -243,6 +243,7 @@ bool ac_rtld_open(struct ac_rtld_binary *binary,
        elf_version(EV_CURRENT);
 
        memset(binary, 0, sizeof(*binary));
+       memcpy(&binary->options, &i.options, sizeof(binary->options));
        binary->num_parts = i.num_parts;
        binary->parts = calloc(sizeof(*binary->parts), i.num_parts);
        if (!binary->parts)
@@ -290,6 +291,9 @@ bool ac_rtld_open(struct ac_rtld_binary *binary,
         * sections in the memory image, and collect and layout private LDS symbols. */
        uint32_t lds_end_align = 0;
 
+       if (binary->options.halt_at_entry)
+               pasted_text_size += 4;
+
        for (unsigned part_idx = 0; part_idx < i.num_parts; ++part_idx) {
                struct ac_rtld_part *part = &binary->parts[part_idx];
                unsigned part_lds_symbols_begin =
@@ -692,6 +696,11 @@ bool ac_rtld_upload(struct ac_rtld_upload_info *u)
                } \
        } while (false)
 
+       if (u->binary->options.halt_at_entry) {
+               /* s_sethalt 1 */
+               *(uint32_t *)u->rx_ptr = util_cpu_to_le32(0xbf8d0001);
+       }
+
        /* First pass: upload raw section data and lay out private LDS symbols. */
        for (unsigned i = 0; i < u->binary->num_parts; ++i) {
                struct ac_rtld_part *part = &u->binary->parts[i];

diff --git a/src/amd/common/ac_rtld.h b/src/amd/common/ac_rtld.h
index 01c29b5..b13270b 100644 (file)
--- a/src/amd/common/ac_rtld.h
+++ b/src/amd/common/ac_rtld.h
@@ -42,8 +42,16 @@ struct ac_rtld_symbol {
        unsigned part_idx; /* shader part in which this symbol appears */
 };
 
+struct ac_rtld_options {
+       /* Loader will insert an s_sethalt 1 instruction as the
+        * first instruction. */
+       bool halt_at_entry:1;
+};
+
 /* Lightweight wrapper around underlying ELF objects. */
 struct ac_rtld_binary {
+       struct ac_rtld_options options;
+
        /* Required buffer sizes, currently read/executable only. */
        uint64_t rx_size;
 
@@ -75,6 +83,7 @@ typedef bool (*ac_rtld_get_external_symbol_cb)(
  */
 struct ac_rtld_open_info {
        const struct radeon_info *info;
+       struct ac_rtld_options options;
 
        unsigned num_parts;
        const char * const *elf_ptrs; /* in-memory ELF objects of each part */

diff --git a/src/gallium/drivers/radeonsi/si_debug_options.h b/src/gallium/drivers/radeonsi/si_debug_options.h
index aa8d64e..d6cb315 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_debug_options.h
+++ b/src/gallium/drivers/radeonsi/si_debug_options.h
@@ -4,6 +4,7 @@ OPT_BOOL(aux_debug, false, "Generate ddebug_dumps for the auxiliary context")
 OPT_BOOL(sync_compile, false, "Always compile synchronously (will cause stalls)")
 OPT_BOOL(dump_shader_binary, false, "Dump shader binary as part of ddebug_dumps")
 OPT_BOOL(debug_disassembly, false, "Report shader disassembly as part of driver debug messages (for shader db)")
+OPT_BOOL(halt_shaders, false, "Halt shaders at the start (will hang)")
 OPT_BOOL(vs_fetch_always_opencode, false, "Always open code vertex fetches (less efficient, purely for testing)")
 OPT_BOOL(prim_restart_tri_strips_only, false, "Only enable primitive restart for triangle strips")
 

diff --git a/src/gallium/drivers/radeonsi/si_shader.c b/src/gallium/drivers/radeonsi/si_shader.c
index 1f4954c..231cb8c 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.c
+++ b/src/gallium/drivers/radeonsi/si_shader.c
@@ -5133,6 +5133,9 @@ static bool si_shader_binary_open(struct si_screen *screen,
 
        bool ok = ac_rtld_open(rtld, (struct ac_rtld_open_info){
                        .info = &screen->info,
+                       .options = {
+                               .halt_at_entry = screen->options.halt_shaders,
+                       },
                        .num_parts = num_parts,
                        .elf_ptrs = part_elfs,
                        .elf_sizes = part_sizes,