radeonsi: add radeon_set_uconfig_reg_seq_perfctr

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8002>

diff --git a/src/gallium/drivers/radeonsi/si_build_pm4.h b/src/gallium/drivers/radeonsi/si_build_pm4.h
index b1fd406..0294c45 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_build_pm4.h
+++ b/src/gallium/drivers/radeonsi/si_build_pm4.h
@@ -102,18 +102,24 @@ static inline void radeon_set_sh_reg(struct radeon_cmdbuf *cs, unsigned reg, uns
    radeon_emit(cs, value);
 }
 
-static inline void radeon_set_uconfig_reg_seq(struct radeon_cmdbuf *cs, unsigned reg, unsigned num)
+static inline void radeon_set_uconfig_reg_seq(struct radeon_cmdbuf *cs, unsigned reg, unsigned num, bool perfctr)
 {
    SI_CHECK_SHADOWED_REGS(reg, num);
    assert(reg >= CIK_UCONFIG_REG_OFFSET && reg < CIK_UCONFIG_REG_END);
    assert(cs->current.cdw + 2 + num <= cs->current.max_dw);
-   radeon_emit(cs, PKT3(PKT3_SET_UCONFIG_REG, num, 0));
+   radeon_emit(cs, PKT3(PKT3_SET_UCONFIG_REG, num, perfctr));
    radeon_emit(cs, (reg - CIK_UCONFIG_REG_OFFSET) >> 2);
 }
 
 static inline void radeon_set_uconfig_reg(struct radeon_cmdbuf *cs, unsigned reg, unsigned value)
 {
-   radeon_set_uconfig_reg_seq(cs, reg, 1);
+   radeon_set_uconfig_reg_seq(cs, reg, 1, false);
+   radeon_emit(cs, value);
+}
+
+static inline void radeon_set_uconfig_reg_perfctr(struct radeon_cmdbuf *cs, unsigned reg, unsigned value)
+{
+   radeon_set_uconfig_reg_seq(cs, reg, 1, true);
    radeon_emit(cs, value);
 }
 

diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
index 775d924..a021301 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_compute.c
+++ b/src/gallium/drivers/radeonsi/si_compute.c
@@ -382,7 +382,7 @@ void si_emit_initial_compute_regs(struct si_context *sctx, struct radeon_cmdbuf
       }
 
       /* Set the pointer to border colors. */
-      radeon_set_uconfig_reg_seq(cs, R_030E00_TA_CS_BC_BASE_ADDR, 2);
+      radeon_set_uconfig_reg_seq(cs, R_030E00_TA_CS_BC_BASE_ADDR, 2, false);
       radeon_emit(cs, bc_va >> 8);                    /* R_030E00_TA_CS_BC_BASE_ADDR */
       radeon_emit(cs, S_030E04_ADDRESS(bc_va >> 40)); /* R_030E04_TA_CS_BC_BASE_ADDR_HI */
    }

diff --git a/src/gallium/drivers/radeonsi/si_compute_prim_discard.c b/src/gallium/drivers/radeonsi/si_compute_prim_discard.c
index 159bd85..10e4668 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_compute_prim_discard.c
+++ b/src/gallium/drivers/radeonsi/si_compute_prim_discard.c
@@ -1226,7 +1226,7 @@ void si_dispatch_prim_discard_cs_and_draw(struct si_context *sctx,
 
       /* Disable ordered alloc for OA resources. */
       for (unsigned i = 0; i < 2; i++) {
-         radeon_set_uconfig_reg_seq(cs, R_031074_GDS_OA_CNTL, 3);
+         radeon_set_uconfig_reg_seq(cs, R_031074_GDS_OA_CNTL, 3, false);
          radeon_emit(cs, S_031074_INDEX(i));
          radeon_emit(cs, 0);
          radeon_emit(cs, S_03107C_ENABLE(0));

diff --git a/src/gallium/drivers/radeonsi/si_perfcounter.c b/src/gallium/drivers/radeonsi/si_perfcounter.c
index 1f2605e..f28a51f 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_perfcounter.c
+++ b/src/gallium/drivers/radeonsi/si_perfcounter.c
@@ -730,7 +730,7 @@ static void si_pc_emit_shaders(struct si_context *sctx, unsigned shaders)
 {
    struct radeon_cmdbuf *cs = &sctx->gfx_cs;
 
-   radeon_set_uconfig_reg_seq(cs, R_036780_SQ_PERFCOUNTER_CTRL, 2);
+   radeon_set_uconfig_reg_seq(cs, R_036780_SQ_PERFCOUNTER_CTRL, 2, false);
    radeon_emit(cs, shaders & 0x7f);
    radeon_emit(cs, 0xffffffff);
 }
@@ -755,7 +755,7 @@ static void si_pc_emit_select(struct si_context *sctx, struct si_pc_block *block
       dw = count + regs->num_prelude;
       if (count >= regs->num_multi)
          dw += regs->num_multi;
-      radeon_set_uconfig_reg_seq(cs, regs->select0, dw);
+      radeon_set_uconfig_reg_seq(cs, regs->select0, dw, false);
       for (idx = 0; idx < regs->num_prelude; ++idx)
          radeon_emit(cs, 0);
       for (idx = 0; idx < MIN2(count, regs->num_multi); ++idx)
@@ -763,7 +763,7 @@ static void si_pc_emit_select(struct si_context *sctx, struct si_pc_block *block
 
       if (count < regs->num_multi) {
          unsigned select1 = regs->select0 + 4 * regs->num_multi;
-         radeon_set_uconfig_reg_seq(cs, select1, count);
+         radeon_set_uconfig_reg_seq(cs, select1, count, false);
       }
 
       for (idx = 0; idx < MIN2(count, regs->num_multi); ++idx)
@@ -778,7 +778,7 @@ static void si_pc_emit_select(struct si_context *sctx, struct si_pc_block *block
 
       assert(!(regs->layout & SI_PC_REG_REVERSE));
 
-      radeon_set_uconfig_reg_seq(cs, regs->select0, count + regs->num_prelude);
+      radeon_set_uconfig_reg_seq(cs, regs->select0, count + regs->num_prelude, false);
       for (idx = 0; idx < regs->num_prelude; ++idx)
          radeon_emit(cs, 0);
       for (idx = 0; idx < count; ++idx)
@@ -786,7 +786,7 @@ static void si_pc_emit_select(struct si_context *sctx, struct si_pc_block *block
 
       select1 = regs->select0 + 4 * regs->num_counters;
       select1_count = MIN2(count, regs->num_multi);
-      radeon_set_uconfig_reg_seq(cs, select1, select1_count);
+      radeon_set_uconfig_reg_seq(cs, select1, select1_count, false);
       for (idx = 0; idx < select1_count; ++idx)
          radeon_emit(cs, 0);
    } else if (layout_multi == SI_PC_MULTI_CUSTOM) {
@@ -804,7 +804,7 @@ static void si_pc_emit_select(struct si_context *sctx, struct si_pc_block *block
       reg_count += regs->num_prelude;
 
       if (!(regs->layout & SI_PC_REG_REVERSE)) {
-         radeon_set_uconfig_reg_seq(cs, reg_base, reg_count);
+         radeon_set_uconfig_reg_seq(cs, reg_base, reg_count, false);
 
          for (idx = 0; idx < regs->num_prelude; ++idx)
             radeon_emit(cs, 0);
@@ -815,7 +815,7 @@ static void si_pc_emit_select(struct si_context *sctx, struct si_pc_block *block
          }
       } else {
          reg_base -= (reg_count - 1) * 4;
-         radeon_set_uconfig_reg_seq(cs, reg_base, reg_count);
+         radeon_set_uconfig_reg_seq(cs, reg_base, reg_count, false);
 
          for (idx = count; idx > 0; --idx) {
             if (idx <= regs->num_multi)