ac/llvm: add vindex into ac_build_buffer_store_dword

for future work

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14266>

diff --git a/src/amd/llvm/ac_llvm_build.c b/src/amd/llvm/ac_llvm_build.c
index 8154104..fb2519e 100644 (file)
--- a/src/amd/llvm/ac_llvm_build.c
+++ b/src/amd/llvm/ac_llvm_build.c
@@ -51,6 +51,12 @@ struct ac_llvm_flow {
    LLVMBasicBlockRef loop_entry_block;
 };
 
+static void ac_build_tbuffer_store(struct ac_llvm_context *ctx, LLVMValueRef rsrc,
+                                   LLVMValueRef vdata, LLVMValueRef vindex, LLVMValueRef voffset,
+                                   LLVMValueRef soffset, LLVMValueRef immoffset,
+                                   unsigned num_channels, unsigned dfmt, unsigned nfmt,
+                                   unsigned cache_policy, bool structurized);
+
 /* Initialize module-independent parts of the context.
  *
  * The caller is responsible for initializing ctx::module and ctx::builder.
@@ -1155,8 +1161,8 @@ void ac_build_buffer_store_format(struct ac_llvm_context *ctx, LLVMValueRef rsrc
  * or v4i32 (num_channels=3,4).
  */
 void ac_build_buffer_store_dword(struct ac_llvm_context *ctx, LLVMValueRef rsrc, LLVMValueRef vdata,
-                                 unsigned num_channels, LLVMValueRef voffset, LLVMValueRef soffset,
-                                 unsigned inst_offset, unsigned cache_policy)
+                                 unsigned num_channels, LLVMValueRef vindex, LLVMValueRef voffset,
+                                 LLVMValueRef soffset, unsigned inst_offset, unsigned cache_policy)
 {
    /* Split 3 channel stores. */
    if (num_channels == 3 && !ac_has_vec3_support(ctx->chip_class, false)) {
@@ -1167,8 +1173,8 @@ void ac_build_buffer_store_dword(struct ac_llvm_context *ctx, LLVMValueRef rsrc,
       }
       v01 = ac_build_gather_values(ctx, v, 2);
 
-      ac_build_buffer_store_dword(ctx, rsrc, v01, 2, voffset, soffset, inst_offset, cache_policy);
-      ac_build_buffer_store_dword(ctx, rsrc, v[2], 1, voffset, soffset, inst_offset + 8,
+      ac_build_buffer_store_dword(ctx, rsrc, v01, 2, vindex, voffset, soffset, inst_offset, cache_policy);
+      ac_build_buffer_store_dword(ctx, rsrc, v[2], 1, vindex, voffset, soffset, inst_offset + 8,
                                   cache_policy);
       return;
    }
@@ -1183,7 +1189,7 @@ void ac_build_buffer_store_dword(struct ac_llvm_context *ctx, LLVMValueRef rsrc,
       if (inst_offset)
          offset = LLVMBuildAdd(ctx->builder, offset, LLVMConstInt(ctx->i32, inst_offset, 0), "");
 
-      ac_build_buffer_store_common(ctx, rsrc, ac_to_float(ctx, vdata), ctx->i32_0, voffset, offset,
+      ac_build_buffer_store_common(ctx, rsrc, ac_to_float(ctx, vdata), vindex, voffset, offset,
                                    cache_policy, false, false);
       return;
    }
@@ -1195,8 +1201,8 @@ void ac_build_buffer_store_dword(struct ac_llvm_context *ctx, LLVMValueRef rsrc,
    unsigned nfmt = V_008F0C_BUF_NUM_FORMAT_UINT;
    LLVMValueRef immoffset = LLVMConstInt(ctx->i32, inst_offset, 0);
 
-   ac_build_raw_tbuffer_store(ctx, rsrc, vdata, voffset, soffset, immoffset, num_channels, dfmt,
-                              nfmt, cache_policy);
+   ac_build_tbuffer_store(ctx, rsrc, vdata, vindex, voffset, soffset, immoffset, num_channels, dfmt,
+                          nfmt, cache_policy, false);
 }
 
 static LLVMValueRef ac_build_buffer_load_common(struct ac_llvm_context *ctx, LLVMValueRef rsrc,

diff --git a/src/amd/llvm/ac_llvm_build.h b/src/amd/llvm/ac_llvm_build.h
index 1f38cbd..1fc9b55 100644 (file)
--- a/src/amd/llvm/ac_llvm_build.h
+++ b/src/amd/llvm/ac_llvm_build.h
@@ -254,8 +254,8 @@ LLVMValueRef ac_build_load_to_sgpr_uint_wraparound(struct ac_llvm_context *ctx,
                                                    LLVMValueRef base_ptr, LLVMValueRef index);
 
 void ac_build_buffer_store_dword(struct ac_llvm_context *ctx, LLVMValueRef rsrc, LLVMValueRef vdata,
-                                 unsigned num_channels, LLVMValueRef voffset, LLVMValueRef soffset,
-                                 unsigned inst_offset, unsigned cache_policy);
+                                 unsigned num_channels, LLVMValueRef vindex, LLVMValueRef voffset,
+                                 LLVMValueRef soffset, unsigned inst_offset, unsigned cache_policy);
 
 void ac_build_buffer_store_format(struct ac_llvm_context *ctx, LLVMValueRef rsrc, LLVMValueRef data,
                                   LLVMValueRef vindex, LLVMValueRef voffset, unsigned cache_policy);

diff --git a/src/amd/llvm/ac_nir_to_llvm.c b/src/amd/llvm/ac_nir_to_llvm.c
index 697bab0..ee44d55 100644 (file)
--- a/src/amd/llvm/ac_nir_to_llvm.c
+++ b/src/amd/llvm/ac_nir_to_llvm.c
@@ -1866,8 +1866,8 @@ static void visit_store_ssbo(struct ac_nir_context *ctx, nir_intrinsic_instr *in
          }
          data = LLVMBuildBitCast(ctx->ac.builder, data, data_type, "");
 
-         ac_build_buffer_store_dword(&ctx->ac, rsrc, data, num_channels, offset, ctx->ac.i32_0, 0,
-                                     cache_policy);
+         ac_build_buffer_store_dword(&ctx->ac, rsrc, data, num_channels, NULL, offset,
+                                     ctx->ac.i32_0, 0, cache_policy);
       }
    }
 
@@ -4210,7 +4210,7 @@ static void visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
          cache_policy |= ac_slc;
 
       ac_build_buffer_store_dword(&ctx->ac, descriptor, store_data, num_components,
-                                  addr_voffset, addr_soffset, const_offset,
+                                  NULL, addr_voffset, addr_soffset, const_offset,
                                   cache_policy);
       break;
    }

diff --git a/src/amd/vulkan/radv_nir_to_llvm.c b/src/amd/vulkan/radv_nir_to_llvm.c
index 21acf41..edcdfa2 100644 (file)
--- a/src/amd/vulkan/radv_nir_to_llvm.c
+++ b/src/amd/vulkan/radv_nir_to_llvm.c
@@ -338,7 +338,7 @@ visit_emit_vertex_with_counter(struct ac_shader_abi *abi, unsigned stream, LLVMV
          out_val = ac_to_integer(&ctx->ac, out_val);
          out_val = LLVMBuildZExtOrBitCast(ctx->ac.builder, out_val, ctx->ac.i32, "");
 
-         ac_build_buffer_store_dword(&ctx->ac, ctx->gsvs_ring[stream], out_val, 1, voffset,
+         ac_build_buffer_store_dword(&ctx->ac, ctx->gsvs_ring[stream], out_val, 1, NULL, voffset,
                                      ac_get_arg(&ctx->ac, ctx->args->ac.gs2vs_offset), 0,
                                      ac_glc | ac_slc | ac_swizzled);
       }
@@ -1103,8 +1103,8 @@ radv_emit_stream_output(struct radv_shader_context *ctx, LLVMValueRef const *so_
       break;
    }
 
-   ac_build_buffer_store_dword(&ctx->ac, so_buffers[buf], vdata, num_comps, so_write_offsets[buf],
-                               ctx->ac.i32_0, offset, ac_glc | ac_slc);
+   ac_build_buffer_store_dword(&ctx->ac, so_buffers[buf], vdata, num_comps, NULL,
+                               so_write_offsets[buf], ctx->ac.i32_0, offset, ac_glc | ac_slc);
 }
 
 static void

diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm_gs.c b/src/gallium/drivers/radeonsi/si_shader_llvm_gs.c
index f3cc145..be679bc 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_llvm_gs.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm_gs.c
@@ -176,7 +176,7 @@ void si_llvm_emit_es_epilogue(struct ac_shader_abi *abi)
             continue;
          }
 
-         ac_build_buffer_store_dword(&ctx->ac, ctx->esgs_ring, out_val, 1, NULL,
+         ac_build_buffer_store_dword(&ctx->ac, ctx->esgs_ring, out_val, 1, NULL, NULL,
                                      ac_get_arg(&ctx->ac, ctx->args.es2gs_offset),
                                      (4 * param + chan) * 4, ac_glc | ac_slc | ac_swizzled);
       }
@@ -277,8 +277,8 @@ static void si_llvm_emit_vertex(struct ac_shader_abi *abi, unsigned stream, LLVM
 
          out_val = ac_to_integer(&ctx->ac, out_val);
 
-         ac_build_buffer_store_dword(&ctx->ac, ctx->gsvs_ring[stream], out_val, 1, voffset, soffset,
-                                     0, ac_glc | ac_slc | ac_swizzled);
+         ac_build_buffer_store_dword(&ctx->ac, ctx->gsvs_ring[stream], out_val, 1, NULL,
+                                     voffset, soffset, 0, ac_glc | ac_slc | ac_swizzled);
       }
    }
 

diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm_tess.c b/src/gallium/drivers/radeonsi/si_shader_llvm_tess.c
index b0aa0a0..c1b4e64 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_llvm_tess.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm_tess.c
@@ -537,7 +537,7 @@ static void si_nir_store_output_tcs(struct ac_shader_abi *abi,
       values[chan] = value;
 
       if (writemask != 0xF && !is_tess_factor) {
-         ac_build_buffer_store_dword(&ctx->ac, buffer, value, 1, addr, base,
+         ac_build_buffer_store_dword(&ctx->ac, buffer, value, 1, NULL, addr, base,
                                      4 * chan, ac_glc);
       }
 
@@ -555,7 +555,7 @@ static void si_nir_store_output_tcs(struct ac_shader_abi *abi,
 
    if (writemask == 0xF && !is_tess_factor) {
       LLVMValueRef value = ac_build_gather_values(&ctx->ac, values, 4);
-      ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, addr, base, 0, ac_glc);
+      ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, NULL, addr, base, 0, ac_glc);
    }
 }
 
@@ -662,7 +662,7 @@ static void si_copy_tcs_inputs(struct si_shader_context *ctx)
 
       LLVMValueRef value = lshs_lds_load(ctx, ctx->ac.i32, ~0, lds_ptr);
 
-      ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, buffer_addr, buffer_offset, 0,
+      ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, NULL, buffer_addr, buffer_offset, 0,
                                   ac_glc);
    }
 }
@@ -775,18 +775,18 @@ static void si_write_tess_factors(struct si_shader_context *ctx, LLVMValueRef re
       ac_build_ifcc(&ctx->ac,
                     LLVMBuildICmp(ctx->ac.builder, LLVMIntEQ, rel_patch_id, ctx->ac.i32_0, ""), 6504);
       ac_build_buffer_store_dword(&ctx->ac, buffer, LLVMConstInt(ctx->ac.i32, 0x80000000, 0), 1,
-                                  ctx->ac.i32_0, tf_base, offset, ac_glc);
+                                  NULL, ctx->ac.i32_0, tf_base, offset, ac_glc);
       ac_build_endif(&ctx->ac, 6504);
       offset += 4;
    }
 
    /* Store the tessellation factors. */
-   ac_build_buffer_store_dword(&ctx->ac, buffer, vec0, MIN2(stride, 4), byteoffset, tf_base, offset,
-                               ac_glc);
+   ac_build_buffer_store_dword(&ctx->ac, buffer, vec0, MIN2(stride, 4), NULL, byteoffset,
+                               tf_base, offset, ac_glc);
    offset += 16;
    if (vec1)
-      ac_build_buffer_store_dword(&ctx->ac, buffer, vec1, stride - 4, byteoffset, tf_base, offset,
-                                  ac_glc);
+      ac_build_buffer_store_dword(&ctx->ac, buffer, vec1, stride - 4, NULL, byteoffset,
+                                  tf_base, offset, ac_glc);
 
    /* Store the tess factors into the offchip buffer if TES reads them. */
    if (shader->key.ge.part.tcs.epilog.tes_reads_tess_factors) {
@@ -806,8 +806,8 @@ static void si_write_tess_factors(struct si_shader_context *ctx, LLVMValueRef re
                                    : util_next_power_of_two(outer_comps);
       outer_vec = ac_build_gather_values(&ctx->ac, outer, outer_vec_size);
 
-      ac_build_buffer_store_dword(&ctx->ac, buf, outer_vec, outer_comps, tf_outer_offset, base, 0,
-                                  ac_glc);
+      ac_build_buffer_store_dword(&ctx->ac, buf, outer_vec, outer_comps, NULL, tf_outer_offset,
+                                  base, 0, ac_glc);
       if (inner_comps) {
          param_inner = si_shader_io_get_unique_index_patch(VARYING_SLOT_TESS_LEVEL_INNER);
          tf_inner_offset = get_tcs_tes_buffer_address(ctx, rel_patch_id, NULL,
@@ -815,8 +815,8 @@ static void si_write_tess_factors(struct si_shader_context *ctx, LLVMValueRef re
 
          inner_vec =
             inner_comps == 1 ? inner[0] : ac_build_gather_values(&ctx->ac, inner, inner_comps);
-         ac_build_buffer_store_dword(&ctx->ac, buf, inner_vec, inner_comps, tf_inner_offset, base,
-                                     0, ac_glc);
+         ac_build_buffer_store_dword(&ctx->ac, buf, inner_vec, inner_comps, NULL,
+                                     tf_inner_offset, base, 0, ac_glc);
       }
    }
 

diff --git a/src/gallium/drivers/radeonsi/si_shader_llvm_vs.c b/src/gallium/drivers/radeonsi/si_shader_llvm_vs.c
index d14f8a7..1fed7cb 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader_llvm_vs.c
+++ b/src/gallium/drivers/radeonsi/si_shader_llvm_vs.c
@@ -310,7 +310,7 @@ void si_llvm_streamout_store_output(struct si_shader_context *ctx, LLVMValueRef
       break;
    }
 
-   ac_build_buffer_store_dword(&ctx->ac, so_buffers[buf_idx], vdata, num_comps,
+   ac_build_buffer_store_dword(&ctx->ac, so_buffers[buf_idx], vdata, num_comps, NULL,
                                so_write_offsets[buf_idx], ctx->ac.i32_0, stream_out->dst_offset * 4,
                                ac_glc | ac_slc);
 }