st/nine: Recompile optimized shaders based on b/i consts

Boolean and Integer constants are used in d3d9 for flow control.

Boolean are used for if/then/else and Integer constants
for loops.
The compilers can generate better code if these values are known
at compilation.
I haven't met so far a game that would change the values of these
constants frequently (and when they do, they set to the values used
for the previous draw call, and thus the changes get filtered out).

Thus it makes sense to inline these constants and recompile the shaders.
The commit sets a bound to the number of variants for a given shader
to avoid too many shaders to be generated.

One drawback is it means more shader compilations. It would probably
make sense to compile these shaders asynchronously or let the user
control the behaviour with an env var, but this is not done here.

The games I tested hit very few shader variants, and the performance
impact was negligible, but it could help for games with uber shaders.

Signed-off-by: Axel Davy <[email protected]>

9 files changed, 211 insertions, 5 deletions

diff --git a/src/gallium/state_trackers/nine/device9.c b/src/gallium/state_trackers/nine/device9.c
index 2cc3a9465fa..78ca58d3c27 100644
--- a/src/gallium/state_trackers/nine/device9.c
+++ b/src/gallium/state_trackers/nine/device9.c
@@ -491,6 +491,10 @@ NineDevice9_ctor( struct NineDevice9 *This,
     This->driver_caps.ps_integer = pScreen->get_shader_param(pScreen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_INTEGERS);
     This->driver_caps.offset_units_unscaled = GET_PCAP(POLYGON_OFFSET_UNITS_UNSCALED);
 
+    /* Code would be needed when integers are not available to correctly
+     * handle the conversion of integer constants */
+    This->context.inline_constants = This->driver_caps.vs_integer && This->driver_caps.ps_integer;
+
     nine_ff_init(This); /* initialize fixed function code */
 
     NineDevice9_SetDefaultState(This, FALSE);
diff --git a/src/gallium/state_trackers/nine/nine_shader.c b/src/gallium/state_trackers/nine/nine_shader.c
index 57bf5efd5dc..3035b76d816 100644
--- a/src/gallium/state_trackers/nine/nine_shader.c
+++ b/src/gallium/state_trackers/nine/nine_shader.c
@@ -3486,6 +3486,9 @@ tx_ctor(struct shader_translator *tx, struct nine_shader_info *info)
     info->position_t = FALSE;
     info->point_size = FALSE;
 
+    memset(info->int_slots_used, 0, sizeof(info->int_slots_used));
+    memset(info->bool_slots_used, 0, sizeof(info->bool_slots_used));
+
     tx->info->const_float_slots = 0;
     tx->info->const_int_slots = 0;
     tx->info->const_bool_slots = 0;
@@ -3696,6 +3699,32 @@ nine_translate_shader(struct NineDevice9 *device, struct nine_shader_info *info,
     if (tx->mul_zero_wins)
        ureg_property(tx->ureg, TGSI_PROPERTY_MUL_ZERO_WINS, 1);
 
+    /* Add additional definition of constants */
+    if (info->add_constants_defs.c_combination) {
+        unsigned i;
+
+        assert(info->add_constants_defs.int_const_added);
+        assert(info->add_constants_defs.bool_const_added);
+        /* We only add constants that are used by the shader
+         * and that are not defined in the shader */
+        for (i = 0; i < NINE_MAX_CONST_I; ++i) {
+            if ((*info->add_constants_defs.int_const_added)[i]) {
+                DBG("Defining const i%i : { %i %i %i %i }\n", i,
+                    info->add_constants_defs.c_combination->const_i[i][0],
+                    info->add_constants_defs.c_combination->const_i[i][1],
+                    info->add_constants_defs.c_combination->const_i[i][2],
+                    info->add_constants_defs.c_combination->const_i[i][3]);
+                tx_set_lconsti(tx, i, info->add_constants_defs.c_combination->const_i[i]);
+            }
+        }
+        for (i = 0; i < NINE_MAX_CONST_B; ++i) {
+            if ((*info->add_constants_defs.bool_const_added)[i]) {
+                DBG("Defining const b%i : %i\n", i, (int)(info->add_constants_defs.c_combination->const_b[i] != 0));
+                tx_set_lconstb(tx, i, info->add_constants_defs.c_combination->const_b[i]);
+            }
+        }
+    }
+
     while (!sm1_parse_eof(tx) && !tx->failure)
         sm1_parse_instruction(tx);
     tx->parse++; /* for byte_size */
diff --git a/src/gallium/state_trackers/nine/nine_shader.h b/src/gallium/state_trackers/nine/nine_shader.h
index 8d98e9e597f..8b5be52de09 100644
--- a/src/gallium/state_trackers/nine/nine_shader.h
+++ b/src/gallium/state_trackers/nine/nine_shader.h
@@ -27,6 +27,7 @@
 #include "d3d9caps.h"
 #include "nine_defines.h"
 #include "nine_helpers.h"
+#include "nine_state.h"
 #include "pipe/p_state.h" /* PIPE_MAX_ATTRIBS */
 #include "util/u_memory.h"
 
@@ -39,6 +40,8 @@ struct nine_lconstf /* NOTE: both pointers should be FREE'd by the user */
     float *data;
 };
 
+struct nine_shader_constant_combination;
+
 struct nine_shader_info
 {
     unsigned type; /* in, PIPE_SHADER_x */
@@ -72,6 +75,9 @@ struct nine_shader_info
     unsigned const_b_base; /* in vec4 (16 byte) units */
     unsigned const_used_size;
 
+    boolean int_slots_used[NINE_MAX_CONST_I];
+    boolean bool_slots_used[NINE_MAX_CONST_B];
+
     unsigned const_float_slots;
     unsigned const_int_slots;
     unsigned const_bool_slots;
@@ -79,6 +85,12 @@ struct nine_shader_info
     struct nine_lconstf lconstf; /* out, NOTE: members to be free'd by user */
     uint8_t bumpenvmat_needed;
 
+    struct {
+        struct nine_shader_constant_combination* c_combination;
+        boolean (*int_const_added)[NINE_MAX_CONST_I];
+        boolean (*bool_const_added)[NINE_MAX_CONST_B];
+    } add_constants_defs;
+
     boolean swvp_on;
 
     boolean process_vertices;
@@ -103,12 +115,16 @@ nine_info_mark_const_f_used(struct nine_shader_info *info, int idx)
 static inline void
 nine_info_mark_const_i_used(struct nine_shader_info *info, int idx)
 {
+    if (!info->swvp_on)
+        info->int_slots_used[idx] = TRUE;
     if (info->const_int_slots < (idx + 1))
         info->const_int_slots = idx + 1;
 }
 static inline void
 nine_info_mark_const_b_used(struct nine_shader_info *info, int idx)
 {
+    if (!info->swvp_on)
+        info->bool_slots_used[idx] = TRUE;
     if (info->const_bool_slots < (idx + 1))
         info->const_bool_slots = idx + 1;
 }
@@ -224,4 +240,91 @@ nine_shader_variants_so_free(struct nine_shader_variant_so *list)
         nine_bind(&list->vdecl, NULL);
 }
 
+struct nine_shader_constant_combination
+{
+    struct nine_shader_constant_combination *next;
+    int const_i[NINE_MAX_CONST_I][4];
+    BOOL const_b[NINE_MAX_CONST_B];
+};
+
+#define NINE_MAX_CONSTANT_COMBINATION_VARIANTS 32
+
+static inline uint8_t
+nine_shader_constant_combination_key(struct nine_shader_constant_combination **list,
+                                     boolean *int_slots_used,
+                                     boolean *bool_slots_used,
+                                     int *const_i,
+                                     BOOL *const_b)
+{
+    int i;
+    uint8_t index = 0;
+    boolean match;
+    struct nine_shader_constant_combination **next_allocate = list, *current = *list;
+
+    assert(int_slots_used);
+    assert(bool_slots_used);
+    assert(const_i);
+    assert(const_b);
+
+    while (current) {
+        index++; /* start at 1. 0 is for the variant without constant replacement */
+        match = TRUE;
+        for (i = 0; i < NINE_MAX_CONST_I; ++i) {
+            if (int_slots_used[i])
+                match &= !memcmp(const_i + 4*i, current->const_i[i], sizeof(current->const_i[0]));
+        }
+        for (i = 0; i < NINE_MAX_CONST_B; ++i) {
+            if (bool_slots_used[i])
+                match &= const_b[i] == current->const_b[i];
+        }
+        if (match)
+            return index;
+        next_allocate = &current->next;
+        current = current->next;
+    }
+
+    if (index < NINE_MAX_CONSTANT_COMBINATION_VARIANTS) {
+        *next_allocate = MALLOC_STRUCT(nine_shader_constant_combination);
+        current = *next_allocate;
+        index++;
+        current->next = NULL;
+        memcpy(current->const_i, const_i, sizeof(current->const_i));
+        memcpy(current->const_b, const_b, sizeof(current->const_b));
+        return index;
+    }
+
+    return 0; /* Too many variants, revert to no replacement */
+}
+
+static inline struct nine_shader_constant_combination *
+nine_shader_constant_combination_get(struct nine_shader_constant_combination *list, uint8_t index)
+{
+    if (index == 0)
+        return NULL;
+    while (index) {
+        assert(list != NULL);
+        index--;
+        if (index == 0)
+            return list;
+        list = list->next;
+    }
+    assert(FALSE);
+    return NULL;
+}
+
+static inline void
+nine_shader_constant_combination_free(struct nine_shader_constant_combination *list)
+{
+    if (!list)
+        return;
+
+    while (list->next) {
+        struct nine_shader_constant_combination *ptr = list->next;
+        list->next = ptr->next;
+        FREE(ptr);
+    }
+
+    FREE(list);
+}
+
 #endif /* _NINE_SHADER_H_ */
diff --git a/src/gallium/state_trackers/nine/nine_state.c b/src/gallium/state_trackers/nine/nine_state.c
index 21d51c9715d..475925ffda9 100644
--- a/src/gallium/state_trackers/nine/nine_state.c
+++ b/src/gallium/state_trackers/nine/nine_state.c
@@ -1595,7 +1595,7 @@ CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader_constant_i,
     }
 
     context->changed.vs_const_i = TRUE;
-    context->changed.group |= NINE_STATE_VS_CONST;
+    context->changed.group |= NINE_STATE_VS_CONST | NINE_STATE_VS_PARAMS_MISC;
 }
 
 CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader_constant_b,
@@ -1614,7 +1614,7 @@ CSMT_ITEM_NO_WAIT(nine_context_set_vertex_shader_constant_b,
         context->vs_const_b[StartRegister + i] = pConstantData[i] ? bool_true : 0;
 
     context->changed.vs_const_b = TRUE;
-    context->changed.group |= NINE_STATE_VS_CONST;
+    context->changed.group |= NINE_STATE_VS_CONST | NINE_STATE_VS_PARAMS_MISC;
 }
 
 CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader,
@@ -1669,7 +1669,7 @@ CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_i_transformed,
            Vector4iCount * sizeof(context->ps_const_i[0]));
 
     context->changed.ps_const_i = TRUE;
-    context->changed.group |= NINE_STATE_PS_CONST;
+    context->changed.group |= NINE_STATE_PS_CONST | NINE_STATE_PS_PARAMS_MISC;
 }
 
 CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_i,
@@ -1694,7 +1694,7 @@ CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_i,
         }
     }
     context->changed.ps_const_i = TRUE;
-    context->changed.group |= NINE_STATE_PS_CONST;
+    context->changed.group |= NINE_STATE_PS_CONST | NINE_STATE_PS_PARAMS_MISC;
 }
 
 CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_b,
@@ -1713,7 +1713,7 @@ CSMT_ITEM_NO_WAIT(nine_context_set_pixel_shader_constant_b,
         context->ps_const_b[StartRegister + i] = pConstantData[i] ? bool_true : 0;
 
     context->changed.ps_const_b = TRUE;
-    context->changed.group |= NINE_STATE_PS_CONST;
+    context->changed.group |= NINE_STATE_PS_CONST | NINE_STATE_PS_PARAMS_MISC;
 }
 
 /* XXX: use resource, as resource might change */
diff --git a/src/gallium/state_trackers/nine/nine_state.h b/src/gallium/state_trackers/nine/nine_state.h
index d8f7230e5b3..376dc561697 100644
--- a/src/gallium/state_trackers/nine/nine_state.h
+++ b/src/gallium/state_trackers/nine/nine_state.h
@@ -300,6 +300,8 @@ struct nine_context {
     int dummy_vbo_bound_at; /* -1 = not bound , >= 0 = bound index */
     boolean vbo_bound_done;
 
+    boolean inline_constants;
+
     struct nine_ff_state ff;
 
     /* software vertex processing */
diff --git a/src/gallium/state_trackers/nine/pixelshader9.c b/src/gallium/state_trackers/nine/pixelshader9.c
index 5d79019a1bc..8db4b6e5c37 100644
--- a/src/gallium/state_trackers/nine/pixelshader9.c
+++ b/src/gallium/state_trackers/nine/pixelshader9.c
@@ -60,6 +60,9 @@ NinePixelShader9_ctor( struct NinePixelShader9 *This,
     info.sampler_ps1xtypes = 0x0;
     info.fog_enable = 0;
     info.projected = 0;
+    info.add_constants_defs.c_combination = NULL;
+    info.add_constants_defs.int_const_added = NULL;
+    info.add_constants_defs.bool_const_added = NULL;
     info.process_vertices = false;
 
     pipe = nine_context_get_pipe_acquire(device);
@@ -82,6 +85,15 @@ NinePixelShader9_ctor( struct NinePixelShader9 *This,
     This->rt_mask = info.rt_mask;
     This->const_used_size = info.const_used_size;
     This->bumpenvmat_needed = info.bumpenvmat_needed;
+
+    memcpy(This->int_slots_used, info.int_slots_used, sizeof(This->int_slots_used));
+    memcpy(This->bool_slots_used, info.bool_slots_used, sizeof(This->bool_slots_used));
+
+    This->const_int_slots = info.const_int_slots;
+    This->const_bool_slots = info.const_bool_slots;
+
+    This->c_combinations = NULL;
+
     /* no constant relative addressing for ps */
     assert(info.lconstf.data == NULL);
     assert(info.lconstf.ranges == NULL);
@@ -115,6 +127,8 @@ NinePixelShader9_dtor( struct NinePixelShader9 *This )
     }
     nine_shader_variants_free(&This->variant);
 
+    nine_shader_constant_combination_free(This->c_combinations);
+
     FREE((void *)This->byte_code.tokens); /* const_cast */
 
     NineUnknown_dtor(&This->base);
@@ -169,6 +183,10 @@ NinePixelShader9_GetVariant( struct NinePixelShader9 *This )
         info.fog_mode = device->context.rs[D3DRS_FOGTABLEMODE];
         info.force_color_in_centroid = key >> 34 & 1;
         info.projected = (key >> 48) & 0xffff;
+        info.add_constants_defs.c_combination =
+            nine_shader_constant_combination_get(This->c_combinations, (key >> 40) & 0xff);
+        info.add_constants_defs.int_const_added = &This->int_slots_used;
+        info.add_constants_defs.bool_const_added = &This->bool_slots_used;
         info.process_vertices = false;
 
         hr = nine_translate_shader(This->base.device, &info, pipe);
diff --git a/src/gallium/state_trackers/nine/pixelshader9.h b/src/gallium/state_trackers/nine/pixelshader9.h
index bcbadd71057..b616d9d7dee 100644
--- a/src/gallium/state_trackers/nine/pixelshader9.h
+++ b/src/gallium/state_trackers/nine/pixelshader9.h
@@ -49,6 +49,14 @@ struct NinePixelShader9
     uint16_t sampler_mask;
     uint8_t rt_mask;
 
+    boolean int_slots_used[NINE_MAX_CONST_I];
+    boolean bool_slots_used[NINE_MAX_CONST_B];
+
+    unsigned const_int_slots;
+    unsigned const_bool_slots;
+
+    struct nine_shader_constant_combination *c_combinations;
+
     uint64_t ff_key[6];
     void *ff_cso;
 
@@ -99,6 +107,13 @@ NinePixelShader9_UpdateKey( struct NinePixelShader9 *ps,
     if (context->rt[0]->base.info.nr_samples)
         key |= ((uint64_t)1) << 34;
 
+    if ((ps->const_int_slots > 0 || ps->const_bool_slots > 0) && context->inline_constants)
+        key |= ((uint64_t)nine_shader_constant_combination_key(&ps->c_combinations,
+                                                               ps->int_slots_used,
+                                                               ps->bool_slots_used,
+                                                               (void *)context->ps_const_i,
+                                                               context->ps_const_b)) << 40;
+
     if (unlikely(ps->byte_code.version < 0x14)) {
         projected = nine_ff_get_projected_key(context);
         key |= ((uint64_t) projected) << 48;
diff --git a/src/gallium/state_trackers/nine/vertexshader9.c b/src/gallium/state_trackers/nine/vertexshader9.c
index f104a9ad134..cffe8501677 100644
--- a/src/gallium/state_trackers/nine/vertexshader9.c
+++ b/src/gallium/state_trackers/nine/vertexshader9.c
@@ -66,6 +66,9 @@ NineVertexShader9_ctor( struct NineVertexShader9 *This,
     info.fog_enable = 0;
     info.point_size_min = 0;
     info.point_size_max = 0;
+    info.add_constants_defs.c_combination = NULL;
+    info.add_constants_defs.int_const_added = NULL;
+    info.add_constants_defs.bool_const_added = NULL;
     info.swvp_on = !!(device->params.BehaviorFlags & D3DCREATE_SOFTWARE_VERTEXPROCESSING);
     info.process_vertices = false;
 
@@ -100,6 +103,14 @@ NineVertexShader9_ctor( struct NineVertexShader9 *This,
     This->position_t = info.position_t;
     This->point_size = info.point_size;
 
+    memcpy(This->int_slots_used, info.int_slots_used, sizeof(This->int_slots_used));
+    memcpy(This->bool_slots_used, info.bool_slots_used, sizeof(This->bool_slots_used));
+
+    This->const_int_slots = info.const_int_slots;
+    This->const_bool_slots = info.const_bool_slots;
+
+    This->c_combinations = NULL;
+
     for (i = 0; i < info.num_inputs && i < ARRAY_SIZE(This->input_map); ++i)
         This->input_map[i].ndecl = info.input_map[i];
     This->num_inputs = i;
@@ -142,6 +153,8 @@ NineVertexShader9_dtor( struct NineVertexShader9 *This )
     nine_shader_variants_free(&This->variant);
     nine_shader_variants_so_free(&This->variant_so);
 
+    nine_shader_constant_combination_free(This->c_combinations);
+
     FREE((void *)This->byte_code.tokens); /* const_cast */
 
     FREE(This->lconstf.data);
@@ -195,6 +208,10 @@ NineVertexShader9_GetVariant( struct NineVertexShader9 *This )
         info.fog_enable = device->context.rs[D3DRS_FOGENABLE];
         info.point_size_min = asfloat(device->context.rs[D3DRS_POINTSIZE_MIN]);
         info.point_size_max = asfloat(device->context.rs[D3DRS_POINTSIZE_MAX]);
+        info.add_constants_defs.c_combination =
+            nine_shader_constant_combination_get(This->c_combinations, (key >> 16) & 0xff);
+        info.add_constants_defs.int_const_added = &This->int_slots_used;
+        info.add_constants_defs.bool_const_added = &This->bool_slots_used;
         info.swvp_on = device->context.swvp;
         info.process_vertices = false;
 
@@ -232,6 +249,9 @@ NineVertexShader9_GetVariantProcessVertices( struct NineVertexShader9 *This,
     info.fog_enable = false;
     info.point_size_min = 0;
     info.point_size_max = 0;
+    info.add_constants_defs.c_combination = NULL;
+    info.add_constants_defs.int_const_added = NULL;
+    info.add_constants_defs.bool_const_added = NULL;
     info.swvp_on = true;
     info.vdecl_out = vdecl_out;
     info.process_vertices = true;
diff --git a/src/gallium/state_trackers/nine/vertexshader9.h b/src/gallium/state_trackers/nine/vertexshader9.h
index 888f1de35be..766b2fd13ea 100644
--- a/src/gallium/state_trackers/nine/vertexshader9.h
+++ b/src/gallium/state_trackers/nine/vertexshader9.h
@@ -59,6 +59,14 @@ struct NineVertexShader9
 
     struct nine_lconstf lconstf;
 
+    boolean int_slots_used[NINE_MAX_CONST_I];
+    boolean bool_slots_used[NINE_MAX_CONST_B];
+
+    unsigned const_int_slots;
+    unsigned const_bool_slots;
+
+    struct nine_shader_constant_combination *c_combinations;
+
     uint64_t ff_key[3];
     void *ff_cso;
 
@@ -93,6 +101,13 @@ NineVertexShader9_UpdateKey( struct NineVertexShader9 *vs,
         key |= (uint32_t) ((!!context->rs[D3DRS_FOGENABLE]) << 8);
     key |= (uint32_t) (context->swvp << 9);
 
+    if ((vs->const_int_slots > 0 || vs->const_bool_slots > 0) && context->inline_constants && !context->swvp)
+        key |= ((uint64_t)nine_shader_constant_combination_key(&vs->c_combinations,
+                                                               vs->int_slots_used,
+                                                               vs->bool_slots_used,
+                                                               context->vs_const_i,
+                                                               context->vs_const_b)) << 16;
+
     /* We want to use a 64 bits key for performance.
      * Use compressed float16 values for the pointsize min/max in the key.
      * Shaders do not usually output psize.*/