From fd102343632f06b3c782275dc92d753ebe1b4ec7 Mon Sep 17 00:00:00 2001
From: Georgi Gerganov <ggerganov@gmail.com>
Date: Wed, 17 Jan 2024 18:46:30 +0200
Subject: [PATCH] imatrix : offload to GPU support (llama/4957)

* backend : add eval callback

ggml-ci

* backend : group nodes in a single compute when user don't need them

* backend : clean-up the implementation

ggml-ci

* simple : do not perform tensor data copy if not needed

* simple : fix

* imatrix : offload to GPU support

* imatrix : fix ggml_mul_mat_id hanlding

ggml-ci

* ci : add imatrix test

ggml-ci

* ci : rearrange output

ggml-ci
---
 ggml.c | 14 --------------
 ggml.h |  6 ------
 2 files changed, 20 deletions(-)

diff --git a/ggml.c b/ggml.c
index d7e01b81..35fd29a9 100644
--- a/ggml.c
+++ b/ggml.c
@@ -394,12 +394,6 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
 static void ggml_vec_dot_f32(const int n, float * restrict s, const float * restrict x, const float * restrict y);
 static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t * restrict x, ggml_fp16_t * restrict y);
 
-ggml_collect_imatrix_t g_imatrix_collect = NULL;
-
-void ggml_set_imatrix_collection(ggml_collect_imatrix_t imatrix_collect) {
-    g_imatrix_collect = imatrix_collect;
-}
-
 static const ggml_type_traits_t type_traits[GGML_TYPE_COUNT] = {
     [GGML_TYPE_I8] = {
         .type_name                = "i8",
@@ -9790,10 +9784,6 @@ static void ggml_compute_forward_mul_mat(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    if (ith == 1 && g_imatrix_collect) {
-        g_imatrix_collect(src0, src1);
-    }
-
     const enum ggml_type type = src0->type;
 
     const bool src1_cont = ggml_is_contiguous(src1);
@@ -10097,10 +10087,6 @@ static void ggml_compute_forward_mul_mat_id(
 
         const struct ggml_tensor * src0_cur = dst->src[cur_a + 2];
 
-        if (ith == 1 && g_imatrix_collect) {
-            g_imatrix_collect(src0_cur, src1);
-        }
-
         const void * wdata    = (src1->type == vec_dot_type) ? src1->data : params->wdata;
         const size_t row_size = ggml_row_size(vec_dot_type, ne10);
 
diff --git a/ggml.h b/ggml.h
index 837c52e6..27daf6fd 100644
--- a/ggml.h
+++ b/ggml.h
@@ -2085,12 +2085,6 @@ extern "C" {
     GGML_API void ggml_init_iq2_quantization(enum ggml_type type);
     GGML_API void ggml_deinit_iq2_quantization(enum ggml_type type);
 
-    //
-    // Importance matrix
-    //
-    typedef void(*ggml_collect_imatrix_t)(const struct ggml_tensor * src0, const struct ggml_tensor * src1);
-    GGML_API void ggml_set_imatrix_collection(ggml_collect_imatrix_t imatrix_collect);
-
     //
     // gguf
     //