radeonsi: disallow adding BOs into si_pm4_state except 1 shader BO per state

The si_shader pointer is already there, so use it and remove the array
of BOs.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5603>

diff --git a/src/gallium/drivers/radeonsi/si_cp_dma.c b/src/gallium/drivers/radeonsi/si_cp_dma.c
index 710dd332816355866b38d73fc1734b85a93460d0..56d240c0a20ade8b56002c8d890c471ef1e3987e 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_cp_dma.c
+++ b/src/gallium/drivers/radeonsi/si_cp_dma.c
@@ -409,8 +409,7 @@ void cik_prefetch_TC_L2_async(struct si_context *sctx, struct pipe_resource *buf
 
 static void cik_prefetch_shader_async(struct si_context *sctx, struct si_pm4_state *state)
 {
-   struct pipe_resource *bo = &state->bo[0]->b.b;
-   assert(state->nbo == 1);
+   struct pipe_resource *bo = &state->shader->bo->b.b;
 
    cik_prefetch_TC_L2_async(sctx, bo, 0, bo->width0);
 }

diff --git a/src/gallium/drivers/radeonsi/si_pm4.c b/src/gallium/drivers/radeonsi/si_pm4.c
index a409b6efdd60685053ff81f64f6ed6239b2f46a1..a5d93071eccadb5747f39a9dbcfa5e3ff524dd5a 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_pm4.c
+++ b/src/gallium/drivers/radeonsi/si_pm4.c
@@ -83,22 +83,8 @@ void si_pm4_set_reg(struct si_pm4_state *state, unsigned reg, uint32_t val)
    si_pm4_cmd_end(state, false);
 }
 
-void si_pm4_add_bo(struct si_pm4_state *state, struct si_resource *bo, enum radeon_bo_usage usage,
-                   enum radeon_bo_priority priority)
-{
-   unsigned idx = state->nbo++;
-   assert(idx < SI_PM4_MAX_BO);
-
-   si_resource_reference(&state->bo[idx], bo);
-   state->bo_usage[idx] = usage;
-   state->bo_priority[idx] = priority;
-}
-
 void si_pm4_clear_state(struct si_pm4_state *state)
 {
-   for (int i = 0; i < state->nbo; ++i)
-      si_resource_reference(&state->bo[i], NULL);
-   state->nbo = 0;
    state->ndw = 0;
 }
 
@@ -119,9 +105,9 @@ void si_pm4_emit(struct si_context *sctx, struct si_pm4_state *state)
 {
    struct radeon_cmdbuf *cs = sctx->gfx_cs;
 
-   for (int i = 0; i < state->nbo; ++i) {
-      radeon_add_to_buffer_list(sctx, sctx->gfx_cs, state->bo[i], state->bo_usage[i],
-                                state->bo_priority[i]);
+   if (state->shader) {
+      radeon_add_to_buffer_list(sctx, sctx->gfx_cs, state->shader->bo,
+                                RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
    }
 
    radeon_emit_array(cs, state->pm4, state->ndw);

diff --git a/src/gallium/drivers/radeonsi/si_pm4.h b/src/gallium/drivers/radeonsi/si_pm4.h
index 34dc9be847261aef15501ffcc1e463cb7c24dd2e..37ade60b25ae7c036fd1bce90a17dca4e26b05a1 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_pm4.h
+++ b/src/gallium/drivers/radeonsi/si_pm4.h
@@ -28,7 +28,6 @@
 #include "radeon/radeon_winsys.h"
 
 #define SI_PM4_MAX_DW 176
-#define SI_PM4_MAX_BO 3
 
 // forward defines
 struct si_context;
@@ -50,12 +49,6 @@ struct si_pm4_state {
    unsigned ndw;
    uint32_t pm4[SI_PM4_MAX_DW];
 
-   /* BO's referenced by this state */
-   unsigned nbo;
-   struct si_resource *bo[SI_PM4_MAX_BO];
-   enum radeon_bo_usage bo_usage[SI_PM4_MAX_BO];
-   enum radeon_bo_priority bo_priority[SI_PM4_MAX_BO];
-
    /* For shader states only */
    struct si_shader *shader;
    struct si_atom atom;
@@ -66,8 +59,6 @@ void si_pm4_cmd_add(struct si_pm4_state *state, uint32_t dw);
 void si_pm4_cmd_end(struct si_pm4_state *state, bool predicate);
 
 void si_pm4_set_reg(struct si_pm4_state *state, unsigned reg, uint32_t val);
-void si_pm4_add_bo(struct si_pm4_state *state, struct si_resource *bo, enum radeon_bo_usage usage,
-                   enum radeon_bo_priority priority);
 
 void si_pm4_clear_state(struct si_pm4_state *state);
 void si_pm4_free_state(struct si_context *sctx, struct si_pm4_state *state, unsigned idx);

diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index cee13524edc9c390f37b1fa8694991660518eb60..61af8ce2893b494397946499d539d86a5505bd18 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -479,8 +479,6 @@ static void si_shader_ls(struct si_screen *sscreen, struct si_shader *shader)
       return;
 
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
-
    si_pm4_set_reg(pm4, R_00B520_SPI_SHADER_PGM_LO_LS, va >> 8);
    si_pm4_set_reg(pm4, R_00B524_SPI_SHADER_PGM_HI_LS, S_00B524_MEM_BASE(va >> 40));
 
@@ -503,7 +501,6 @@ static void si_shader_hs(struct si_screen *sscreen, struct si_shader *shader)
       return;
 
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
 
    if (sscreen->info.chip_class >= GFX9) {
       if (sscreen->info.chip_class >= GFX10) {
@@ -589,7 +586,6 @@ static void si_shader_es(struct si_screen *sscreen, struct si_shader *shader)
 
    pm4->atom.emit = si_emit_shader_es;
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
 
    if (shader->selector->type == PIPE_SHADER_VERTEX) {
       vgpr_comp_cnt = si_get_vs_vgpr_comp_cnt(sscreen, shader, false);
@@ -818,7 +814,6 @@ static void si_shader_gs(struct si_screen *sscreen, struct si_shader *shader)
       S_028B90_CNT(MIN2(gs_num_invocations, 127)) | S_028B90_ENABLE(gs_num_invocations > 0);
 
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
 
    if (sscreen->info.chip_class >= GFX9) {
       unsigned input_prim = sel->info.properties[TGSI_PROPERTY_GS_INPUT_PRIM];
@@ -1088,7 +1083,6 @@ static void gfx10_shader_ngg(struct si_screen *sscreen, struct si_shader *shader
    }
 
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
 
    if (es_type == PIPE_SHADER_VERTEX) {
       es_vgpr_comp_cnt = si_get_vs_vgpr_comp_cnt(sscreen, shader, false);
@@ -1388,7 +1382,6 @@ static void si_shader_vs(struct si_screen *sscreen, struct si_shader *shader,
    }
 
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
 
    if (gs) {
       vgpr_comp_cnt = 0; /* only VertexID is needed for GS-COPY. */
@@ -1641,7 +1634,6 @@ static void si_shader_ps(struct si_screen *sscreen, struct si_shader *shader)
    shader->ctx_reg.ps.cb_shader_mask = cb_shader_mask;
 
    va = shader->bo->gpu_address;
-   si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
    si_pm4_set_reg(pm4, R_00B020_SPI_SHADER_PGM_LO_PS, va >> 8);
    si_pm4_set_reg(pm4, R_00B024_SPI_SHADER_PGM_HI_PS, S_00B024_MEM_BASE(va >> 40));